Conversation
…be (#1359) The pre-flight binary smoke does a bare `bun build --compile` — it deliberately skips `scripts/build-binaries.sh` to stay fast. That means packages/paths/src/bundled-build.ts retains its dev defaults, including BUNDLED_IS_BINARY = false. version.ts branches on BUNDLED_IS_BINARY: when true it returns the embedded string; when false it calls getDevVersion(), which reads package.json at `SCRIPT_DIR/../../../../package.json`. Inside a compiled binary SCRIPT_DIR resolves under `$bunfs/root/`, the walk produces a CWD- relative path that doesn't exist, and the smoke aborts with "Failed to read version: package.json not found" — a false positive. Hit during the 0.3.8 release attempt: the real Pi lazy-load fix was working end-to-end; the smoke test was the only thing failing. Use --help instead. It exercises the same module-init graph (so it still catches the real failure modes the skill lists — Pi package.json init crash, Bun --bytecode bugs, CJS wrapper issues, circular imports under minify) but has no dev/binary branch, so no false positive. Also add a longer comment block explaining why --help is preferred, so this doesn't get "normalized" back to `version` by a future drive-by.
The brew path of /test-release runs `brew uninstall` in Phase 5 to leave the system in its pre-test state. For operators using the dual-homebrew pattern (renamed brew binary at `/opt/homebrew/bin/archon-stable` so it coexists with a `bun link` dev `archon`), that uninstall wipes the Cellar dir the `archon-stable` symlink points into → `archon-stable` becomes dangling → `brew cleanup` sweeps it away on the next brew op. Next time the operator wants stable, they have to manually re-run `brew-upgrade-archon`. Fix: make the skill aware of `archon-stable` and restore it transparently. - Phase 2 item 4: detect the `archon-stable` symlink before any brew op; export `ARCHON_STABLE_WAS_INSTALLED=yes` so Phase 5 knows to restore it. Only triggers for the brew path (curl-mac/curl-vps don't touch brew so they leave `archon-stable` alone). - Phase 5 brew path: after `brew uninstall + untap`, if the flag was set, re-tap + re-install + rename. Verifies the restored `archon-stable` reports a version and warns (non-fatal) if the rename target is missing. Documents the tradeoff: the restored version is "whatever the tap ships today", not necessarily the pre-test version — usually that's what the operator wants (the release they just tested becomes stable) but the back-version-QA case requires a manual `brew-upgrade-archon` after. - Phase 1 confirmation banner now mentions that `archon-stable` will be preserved so the operator isn't surprised by the reinstall during Phase 5. No changes to curl-mac/curl-vps paths. No changes to Phase 4 test suite.
… a compiled binary (#1360) v0.3.9 made Pi boot-safe: lazy-loading its imports meant `archon version` no longer crashed on `@mariozechner/pi-coding-agent/dist/config.js`'s module-init `readFileSync(getPackageJsonPath())`. That's what the `provider-lazy-load.test.ts` regression test guards. The fix was only half the problem though. When a Pi workflow actually runs, sendQuery() triggers the dynamic import — and Pi's config.js module-init fires then, hitting the exact same ENOENT on `dirname(process.execPath)/package.json`. Discovered by running `archon workflow run test-pi` against a locally-compiled 0.3.9 binary: [main] Failed: ENOENT: no such file or directory, open '/private/tmp/package.json' at readFileSync (unknown) at <anonymous> (/$bunfs/root/archon-providertest:184:7889) at init_config Boot-safe ≠ runtime-safe. The `/test-release` run for 0.3.9 passed because it only exercised `archon-assist` (Claude); Pi was never actually invoked on the released binary. Fix: before the dynamic `import('@mariozechner/pi-coding-agent')` in sendQuery, install a PI_PACKAGE_DIR shim. Pi's config.js checks `process.env.PI_PACKAGE_DIR` first in its `getPackageDir()` and short-circuits the `dirname(process.execPath)` walk. We write a minimal `{name, version, piConfig:{}}` stub to `tmpdir()/archon-pi-shim/package.json` (idempotent — existsSync check) and set the env var. Pi only reads `piConfig.name`, `piConfig.configDir`, and `version` from that file, all optional, so the stub surface is genuinely minimal. Localized to PiProvider: no global state, no mutation of any shared config, no upstream fork. Claude and Codex providers are unaffected (their SDKs don't have this class of module-init side effect). Verified end-to-end: built a compiled archon binary with this patch, ran `archon workflow run test-pi --no-worktree` (Pi workflow with model `anthropic/claude-haiku-4-5`), got a clean response. Before the patch, same binary crashed at `dag_node_started` with the ENOENT above. Regression test added: asserts `PI_PACKAGE_DIR` is set after sendQuery hits even its fast-fail "no model" path. Together with the existing `provider-lazy-load.test.ts` (boot-safe) this covers both halves.
… and Codex (#1361) Both binary resolvers previously stopped at env-var + explicit config and threw a "not found" error when neither was set. Users who followed the upstream-recommended install flow (Anthropic's `curl install.sh` for Claude, `npm install -g @openai/codex`) still had to manually set either `CLAUDE_BIN_PATH` / `CODEX_BIN_PATH` or the corresponding config field before any workflow could run. Add a tier-N autodetect step between the explicit config tier and the install-instructions throw. Purely additive: env and config still win when set (precedence covered by new tests). On autodetect miss, the same install-instructions error fires as before. Claude probe list (verified against docs.claude.com "Uninstall Claude Code → Native installation" section): - $HOME/.local/bin/claude (mac/linux native installer) - $USERPROFILE\.local\bin\claude.exe (Windows native installer) Codex probe list (verified against openai/codex README; npm global- install puts the binary at `{npm_prefix}/bin/<name>` on POSIX, `{npm_prefix}\<name>.cmd` on Windows): - $HOME/.npm-global/bin/codex (user-set `npm config set prefix`) - /opt/homebrew/bin/codex (mac arm64 with homebrew-node) - /usr/local/bin/codex (mac intel / linux system node) - %APPDATA%\npm\codex.cmd (Windows npm global default) - $HOME\.npm-global\codex.cmd (Windows user-set prefix) Not probed (explicit override still required): - Custom npm prefixes — `npm root -g` would need a subprocess per resolve, too much surface for a probe helper - `brew install --cask codex` — cask layout isn't a PATH binary - Manual GitHub Releases extracts — placement is user-determined - `~/.bun/bin/codex` — not documented in openai/codex README Pi provider intentionally has no equivalent change: the Pi SDK is bundled into the archon binary (no subprocess), so there's no "binary" to resolve. Pi auth lives at `~/.pi/agent/auth.json` which the SDK already finds by default, and the PR A shim (`PI_PACKAGE_DIR`) handles the package-dir case via Pi's own documented escape hatch. E2E verified: removed both config entries from ~/.archon/config.yaml, rebuilt compiled binary, ran `archon workflow run archon-assist` and a Codex workflow. Logs showed `source: 'autodetect'` for both, responses returned cleanly.
…ry autodetect test The native-installer autodetect test computed its expected path from process.env.HOME, but the implementation uses node:os homedir(). On Windows, HOME is typically unset (Windows uses USERPROFILE), so the test fell back to '/Users/test' while the resolver returned the real home dir — making the spy's path-equality check fail and breaking CI on windows-latest. Mirror the implementation by importing homedir() from node:os and joining with node:path so the expected path matches the actual platform-resolved home and separator. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…ver (#1365) Reported in #1365: a user running `archon serve` with DISCORD_BOT_TOKEN set but the "Message Content Intent" toggle disabled in the Discord Developer Portal saw the entire server crash with `Used disallowed intents`. Discord rejects the gateway connection (close code 4014) when a privileged intent is requested without being enabled, and the unguarded `await discord.start()` propagated the error all the way up, taking the web UI down with it. Wrap discord.start() in try/catch — log the failure with an actionable hint (special-cased for the disallowed-intent error) and continue running. Other adapters and the web UI come up regardless. The shutdown handler already uses optional chaining (`discord?.stop()`) so nulling discord after a failed start is safe. Other adapters (Telegram, Slack, GitHub, Gitea, GitLab) have the same unguarded-start pattern but are out of scope for this fix — addressing them is tracked separately. Also expanded the Discord setup docs with a caution callout that names the exact error string and the new log event so users can grep for both. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* docs(script-nodes): add dedicated guide and teach the archon skill how to write them Script nodes (script:) have been a first-class DAG node type since v0.3.3 but were documented only as one-liners in CLAUDE.md and a CI smoke test. Claude Code reading the archon skill would see "Four Node Types: command, prompt, bash, loop" and reach for bash+node/python one-liners instead of a proper script node — losing bun's --no-env-file isolation, uv's --with dependency pins, and the .archon/scripts/ reuse story. - New packages/docs-web/src/content/docs/guides/script-nodes.md mirroring the structure of loop-nodes.md / approval-nodes.md: schema, inline vs named dispatch, runtime/deps semantics, scripts directory precedence (repo > home), extension-runtime mapping, env isolation, stdout/stderr contract, patterns, and the explicit list of ignored AI fields. - guides/authoring-workflows.md and guides/index.md updated so the new guide is discoverable from both the node-types table and the guides landing page. - reference/variables.md calls out the no-shell-quote difference between bash: and script: substitution — a subtle correctness trap when adapting a bash pattern into a script node. - Sidebar order bumped +1 on hooks/mcp-servers/skills/global-workflows/ remotion-workflow to slot script-nodes at order 5 next to the other node-type guides. - .claude/skills/archon/SKILL.md: replaces stale "Four Node Types" (which also silently omitted approval and cancel) with the accurate seven, with a script-node code block showing both inline and named patterns. - references/workflow-dag.md: full Script Node section covering dispatch, resolution, deps, stdout contract, and the list of AI-only fields that are ignored; validation-rules list updated. - references/dag-advanced.md and references/variables.md: retry-support line corrected; no-shell-quote note added. - examples/dag-workflow.yaml: added an extract-labels TypeScript script node and updated the header comment. * fix(docs): review follow-ups for script-node guide - skills example: extract-labels was reading process.env.ISSUE_JSON which is never set; use String.raw`$fetch-issue.output` so the upstream bash node's JSON is actually consumed - guides/script-nodes.md + skills/workflow-dag.md: idle_timeout is accepted but ignored on script (and bash) nodes — executeScriptNode only reads node.timeout. Clarify that script/bash use `timeout`, not idle_timeout - archon-workflow-builder.yaml: prompt enumerated only bash/prompt/command/loop, so the AI builder could never propose script or approval nodes. Add both (plus examples + rule about script output not being shell-quoted) and regenerate bundled defaults - book/dag-workflows.md + book/quick-reference.md + adapters/web.md: fill in the node-type references that were missing script, approval, and cancel. adapters/web.md also overclaimed "loop" in the palette — NodePalette.tsx only drags command/prompt/bash, so note that the other kinds are YAML-only
…nv gaps, add good-practices + troubleshooting (#1363) * fix(skill/when): document the full `when:` operator set and compound expressions The skill reference previously stated "operators: ==, != only" which is materially wrong — the condition evaluator supports ==, !=, <, >, <=, >= plus && / || compound expressions with && binding tighter than ||, plus dot-notation JSON field access. An agent authoring a workflow from the skill would think half the operators don't exist. Replaces the single-sentence section with a structured reference covering: - All six comparison operators (string and numeric modes) - Compound expressions with precedence rules and short-circuit eval - JSON dot notation semantics and failure modes - The fail-closed rules in full (invalid expression, non-numeric side, missing field, skipped upstream) Grounded in packages/workflows/src/condition-evaluator.ts. * feat(skill): document Approval and Cancel node types Approval and cancel nodes are first-class DAG node types (approval since the workflow lifecycle work in #871, cancel as a guarded-exit primitive) but the skill never described either one. An agent reading the skill and asked to "add a review gate before implementation" or "stop the workflow if the input is unsafe" would fall back to bash + exit 1, losing the proper semantics (cancelled vs. failed, on_reject AI rework, web UI auto-resume). Approval node coverage (references/workflow-dag.md, SKILL.md): - Full configuration block with message, capture_response, on_reject - The interactive: true workflow-level requirement for web UI delivery - Approve/reject commands across all platforms (CLI, slash, natural language) and the capture_response → $node-id.output flow - Ignored-fields list + the on_reject.prompt AI sub-node exception Cancel node coverage (references/workflow-dag.md, SKILL.md): - Single-field schema (cancel: "<reason>") - Lifecycle: cancelled (not failed); in-flight parallel nodes stopped; no DAG auto-resume path - The "cancel: vs bash-exit-1" decision rule (expected precondition miss vs. check itself failing) - Two canonical patterns — upstream-classification gate, pre-expensive-step gate Validation-rules list updated to enumerate approval/cancel constraints (message non-empty, on_reject.max_attempts range 1-10, cancel reason non-empty), plus a forward note that script: joins the mutually-exclusive set once PR #1362 lands. Placement in both files is after the Loop section and before the validation section, so this commit stays additive with respect to PR #1362's Script node insertion between Bash and Loop — rebase is clean. * feat(skill): document workflow-level fields beyond name/provider/model The skill's Schema section previously showed only name, description, provider, and model at the workflow level — which is most of a stub. Agents asked to "use the 1M-context Claude beta" or "run this under a network sandbox" or "add a fallback model in case Opus rate-limits" had no way to discover that any of these fields existed at the workflow level. Adds a comprehensive Workflow-Level Fields section covering: - Core: name, description, provider, model, interactive (with explicit callout that interactive: true is REQUIRED for approval/loop gates on web UI — a common footgun) - Isolation: worktree.enabled for pin-on/pin-off (the only worktree field at workflow level; baseBranch/copyFiles/path/initSubmodules are config.yaml only, so a cross-reference points there) - Claude SDK advanced: effort, thinking, fallbackModel, betas, sandbox, with explicit per-node-only exceptions (maxBudgetUsd, systemPrompt) - Codex-specific: modelReasoningEffort (with note that it's NOT the same as Claude's effort — this has confused users), webSearchMode, additionalDirectories - A complete worked example combining sandbox + approval + interactive All fields cross-referenced against packages/workflows/src/schemas/workflow.ts and packages/workflows/src/schemas/dag-node.ts. * feat(skill/loop): document interactive loops and gate_message Interactive loop nodes pause between iterations for human feedback via /workflow approve — used by archon-piv-loop and archon-interactive-prd. The skill's Loop Nodes section previously omitted both interactive: true and gate_message entirely, so an agent writing a guided-refinement workflow wouldn't know the feature exists or that gate_message is required at parse time. Adds: - interactive and gate_message rows to the config table (marking gate_message as required when interactive: true — enforced by the loader's superRefine) - A dedicated "Interactive Loops" subsection explaining the 6-step iterate-pause-approve-resume flow - Explicit call-out that $LOOP_USER_INPUT populates ONLY on the first iteration of a resumed session — easy to miss and a common surprise - Workflow-level interactive: true requirement for web UI delivery (loader warning otherwise) so the full-flow example is complete - Note that until_bash substitution DOES shell-quote $nodeId.output (unlike script bodies) — called out since the audit surfaced this inconsistency * fix(skill/cli): complete the CLI command reference with missing lifecycle commands The CLI reference previously documented only list, run, cleanup, validate, complete, version, setup, and chat — missing nearly every workflow lifecycle command an agent needs to operate a paused, failed, or stuck run. The interactive-workflows reference assumed these commands existed without actually documenting them. Adds full documentation for: - archon workflow status — show running workflow(s) - archon workflow approve <run-id> [comment] — resume approval gate (also populates $LOOP_USER_INPUT on interactive loops and the gate node's output when capture_response: true) - archon workflow reject <run-id> [reason] — reject gate; cancels or triggers on_reject rework depending on node config - archon workflow cancel <run-id> — terminate running/paused with in-flight subprocess kill - archon workflow abandon <run-id> — mark stuck row cancelled without subprocess kill (for orphan-cleanup after server crashes — matches the #1216 precedent) - archon workflow resume <run-id> [message] — force-resume specific run (auto-resume is default; this is for explicit override) - archon workflow cleanup [days] — disk hygiene for old terminal runs (with explicit callout that it does NOT transition 'running' rows, a common confusion) - archon workflow event emit — used inside loop prompts for state signalling; documented so agents don't invent their own mechanism - archon continue <branch> [flags] [msg] — iterative-session entry point with --workflow and --no-context flags Also: - Adds --allow-env-keys flag to the `workflow run` flag table with audit-log context and the env-leak-gate remediation use case - Adds an "Auto-resume without --resume" note disambiguating when --resume is needed vs. when auto-resume handles it - Adds --include-closed flag to `isolation cleanup`, which was previously missing; converts the flag list to a structured table - Explains the cancel/abandon distinction (live subprocess vs. orphan) All grounded in packages/cli/src/commands/workflow.ts, continue.ts, and isolation.ts. * feat(skill/repo-init): add scripts/ and state/, three-path env model, per-project env injection The repo-init reference was missing two first-class .archon/ directories (scripts/ since v0.3.3, state/ since the workflow-state feature) and had nothing to say about env — the #1 thing a user hits on first-run when their repo has a .env file with API keys. Directory tree updates: - Adds .archon/scripts/ with the extension->runtime rule (.ts/.js -> bun, .py -> uv) so agents know where to put named scripts referenced by script: nodes. - Adds .archon/state/ with explicit "always gitignore" callout — these are runtime artifacts, not source. Previously undocumented in the skill. - Adds .archon/.env (repo-scoped Archon env) and distinguishes it from the target repo's top-level .env. - Adds a "What each directory is for" list so the structure isn't just a tree with no narrative. .gitignore guidance: - state/ and .env added as must-gitignore (state/ matches CLAUDE.md and reference/archon-directories.md — skill was lagging). - mcp/ demoted to conditional — gitignore only if you hardcode secrets. New "Three-Path Env Model" section: - ~/.archon/.env (trusted, user), <cwd>/.archon/.env (trusted, repo), <cwd>/.env (UNTRUSTED, target project — stripped from subprocess env). - Precedence (override: true across archon-owned paths) and the observable [archon] loaded N keys / stripped K keys log lines so operators can verify what actually happened. - Decision tree for where to put API keys vs. target-project env vs. things Archon shouldn't touch. - Links to archon setup --scope home|project with --force for writing to the right file with timestamped backups. New "Per-Project Env Injection" section: - Documents both managed surfaces: .archon/config.yaml env: block (git-committed, $REF expansion) and Web UI Settings → Projects → Env Vars (DB-stored, never returned over API). - Names every execution surface that receives the injected vars: Claude/Codex/Pi subprocess, bash: nodes, script: nodes, and direct codebase-scoped chat. - Documents the env-leak gate with all 5 remediation paths so an agent hitting "Cannot register: env has sensitive keys" knows the options. Grounded in CHANGELOG v0.3.7 (three-path env + setup flags), v0.3.0 (env-leak gate), and reference/security.md on the docs site. * fix(skill/authoring-commands): correct override paths and add home-scoped commands The file-location and discovery sections described an override layout that does not match the actual resolver. It showed: .archon/commands/defaults/archon-assist.md # Overrides the bundled and claimed `.archon/commands/defaults/` was where repo-level overrides lived. In fact the resolver (executor-shared.ts:152-200 + command- validation.ts) walks `.archon/commands/` 1 level deep and uses basename matching — putting `archon-assist.md` at the top of `.archon/commands/` is the canonical way to override the bundled version. The `defaults/` subfolder is a Archon-internal convention for shipping bundled defaults, not a user-facing override pattern. Also, home-scoped commands (`~/.archon/commands/`, shipped in v0.3.7) were completely absent — agents authoring personal helpers wouldn't know they could live at the user level and be shared across every repo. Changes: - File Location section now shows all three discovery scopes (repo, home, bundled) with precedence ordering and 1-level subfolder rules - Duplicate-basename rule documented as a user error surface - Discovery and Priority section rewritten with accurate 3-step lookup order — no more references to the nonexistent defaults/ override path - Adds the Web UI "Global (~/.archon/commands/)" palette label note so users authoring helpers for the builder know what to expect No code changes — this is a pure fix of stale/incorrect skill reference material. * feat(skill): add workflow good-practices and troubleshooting reference pages Closes two gaps from the audit. The skill previously had zero guidance on designing multi-node workflows (what to avoid, what to reach for first, how to structure artifact chains) and zero guidance on where to look when things go wrong (log paths, env-leak gate remediations, orphan-row cleanup, resume semantics). New references/good-practices.md (9 Good Practices + 7 Anti-Patterns): - Use deterministic nodes (bash:/script:) for deterministic work, AI for reasoning — the single biggest quality lever - output_format required whenever downstream when: reads a field — the most common source of "workflow silently routes wrong" - trigger_rule: none_failed_min_one_success after conditional branches — the classic bug where all_success fails because a skipped when:-gated branch doesn't count as a success - context: fresh requires artifacts for state passing — commands must explicitly "read $ARTIFACTS_DIR/..." when downstream of fresh - Cheap models (haiku) for glue, strong for substance - Workflow descriptions as routing affordances - Validate (archon validate workflows) + smoke-run before shipping - Artifact-chain-first design - worktree.enabled: true for code-changing workflows (reversibility) - Anti-patterns with before/after YAML examples for each (AI-for-tests, free-form when: matching, context: fresh without artifacts, long flat AI-node layers, secrets in YAML, retry on loop nodes, tiny max_iterations, missing workflow-level interactive:, tool-restricted MCP nodes) New references/troubleshooting.md: - Log location (~/.archon/workspaces/<owner>/<repo>/logs/<run-id>.jsonl) with jq recipes for common queries (last assistant message, failed events, full stream) - Artifact location for cross-node handoff debugging - 9 Common Failure Modes, each with root cause + concrete fix: - $BASE_BRANCH unresolvable - Env-leak gate (5 remediations) - Claude/Codex binary not found (compiled-binary-only) - "running" forever (AI working / orphan / idle_timeout) - Mid-workflow failure and auto-resume semantics - Approval gate missing on web UI (workflow-level interactive:) - MCP plugin connection noise (filtered by design) - Empty $nodeId.output / field access (4 causes) - Diagnostic command cheat sheet (list, status, isolation list, validate, tail-log, --verbose, LOG_LEVEL=debug) - Escalation protocol (version + validate + log tail + CHANGELOG + issue) SKILL.md routing table now dispatches "Workflow good practices / anti-patterns" and "Troubleshoot a failing / stuck workflow" to the new references so an agent can find them without having to know they exist. * docs(book): update node-types coverage from four to all seven The book is the curated first-contact reading path (landing page → "Get Started" → /book/). Both dag-workflows.md and quick-reference.md were stuck on "four node types" — missing script, approval, and cancel. A user reading the book as their first introduction would form an incomplete mental model, then find three more node types in the reference section later with no explanation of when they arrived. book/dag-workflows.md: - "four node types" → "seven node types. Exactly one mode field is required per node" - Table now lists Command, Prompt, Bash, Script, Loop, Approval, Cancel with one-line "when to use" for each, and cross-links to the dedicated guide pages for Script / Loop / Approval - New sections below the table for Script (inline + named examples with runtime and deps), Approval (with the interactive: true workflow-level note that's easy to miss), and Cancel (guarded-exit pattern) — keeping the existing narrative shape for Bash and Loop book/quick-reference.md: - Node Options table now includes script, approval, cancel rows - agents row added (inline sub-agents, Claude-only) - New "Script-specific fields" and "Approval-specific fields" subsections so the cheat-sheet is actually complete rather than pointing users elsewhere for the required constraints - Retry row callout that loop nodes hard-error on retry — previously omitted - bash timeout note widened to cover script timeout (same semantics) Both files are docs-web content; the CI build on the docs-script-nodes PR (#1362) previously validated the Starlight build path with a similar table addition, so this should render clean. * fix(skill/cli): remove nonexistent \`archon workflow cancel\`, fix workflow status jq recipe Two accuracy issues from the PR code-reviewer (comment 4311243858). C1: \`archon workflow cancel <run-id>\` does NOT exist as a CLI subcommand. The switch at packages/cli/src/cli.ts:318-485 dispatches on list / run / status / resume / abandon / approve / reject / cleanup / event — running \`archon workflow cancel\` hits the default case and exits with "Unknown workflow subcommand: cancel" (cli.ts:478-484). Active cancellation is only available via: - /workflow cancel <run-id> chat slash command (all platforms) - Cancel button on the Web UI dashboard - POST /api/workflows/runs/{runId}/cancel REST endpoint cli-commands.md: removed the \`### archon workflow cancel <run-id>\` subsection; kept the \`abandon\` subsection but made it explicit that abandon does NOT kill a subprocess. Added a call-out box at the bottom of the abandon section explaining where to go for actual cancellation. troubleshooting.md "running forever" section: split the original cancel-vs-abandon advice into three bullets — Web UI / CLI abandon (for orphans, no subprocess kill) / chat \`/workflow cancel\` (for live runs that need interruption). Added an explicit "there is no archon workflow cancel CLI subcommand" parenthetical since the wrong command was being suggested in flow. I1: the \`archon workflow list --json\` diagnostic used an incorrect jq filter. workflow list's --json output (workflow.ts:185-219) has shape { workflows: [{ name, description, provider?, model?, ... }], errors: [...] } with no \`runs\` field — \`jq '.workflows[] | select(.runs)'\` returns empty unconditionally. Replaced with \`archon workflow status --json | jq '.runs[]'\`, which matches the actual shape of workflowStatusCommand at workflow.ts:852+ ({ runs: WorkflowRun[] }). Also tightened the narration to distinguish JSON from human-readable status output. No change to the commit history in this PR — these are follow-up fixes to claims I introduced in earlier commits of this branch (f10b989 for C1, 66d2b86 for I1). * fix(skill): remove env-leak gate references (feature was removed in provider extraction) C2 from the PR code-reviewer (comment 4311243858). The pre-spawn env-leak gate was removed from the codebase during the provider-extraction refactor — see TODO(#1135) at packages/providers/src/claude/provider.ts:908. Zero hits for --allow-env-keys / allowEnvKeys / allow_env_keys / allow_target_repo_keys across packages/. The CLI's parseArgs (cli.ts:182-208) has no --allow-env-keys option, and because parseArgs uses strict: false, an unknown --allow-env-keys would be silently ignored rather than error. What remains accurate and is NOT touched: - Three-Path Env Model section (user/repo archon-owned envs are loaded; target repo <cwd>/.env keys are stripped from process.env at boot) still correctly describes current behavior, grounded in packages/paths/src/strip-cwd-env.ts + env-integration.test.ts - Per-Project Env Injection section (Option 1: .archon/config.yaml env: block; Option 2: Web UI Settings → Projects → Env Vars) is unchanged — both remain the sanctioned way to get env vars into subprocesses Removed claims (all three files): - cli-commands.md: --allow-env-keys flag row in the workflow run flags table - repo-init.md: the "Env-leak gate" subsection at the end of Per-Project Env Injection listing 5 remediations (all of which reference UI/CLI/ config surfaces that don't exist). Replaced with a succinct callout that explains the actual current behavior — target repo .env keys are stripped, workflows that need those values should use managed injection — so the reader still gets the "where to put my env vars" answer - troubleshooting.md: the "Cannot register: codebase has sensitive env keys" section (error message that can no longer be emitted) If the env-leak gate is ever resurrected per TODO(#1135), the docs can be re-added then. The CHANGELOG v0.3.0 entry describing the gate is a historical record of past behavior and does not need to be rewritten. * fix(skill/troubleshooting): correct JSONL event type names and field name C3 from the PR code-reviewer (comment 4311243858). The troubleshooting reference's event-types table used _started / _completed / _failed suffixes, but packages/workflows/src/logger.ts:19-30 shows the actual WorkflowEvent.type enum is: workflow_start | workflow_complete | workflow_error | assistant | tool | validation | node_start | node_complete | node_skipped | node_error The second jq recipe also queried `.event` but the discriminator is `.type`. Fixes: - Event table: renamed columns (_started → _start, _completed → _complete, _failed → _error). Explicitly called out the field name as `type` so the reader knows what jq selector to use - Replaced the "tool_use / tool_result" row with a single `tool` row and listed its actual payload fields (tool_name, tool_input, duration_ms, tokens) — tool_use/tool_result are SDK message kinds that appear within the AI stream, not top-level log event types - Added a `validation` row (was missing; it's emitted by workflow-level validation calls with `check` and `result` fields) - Removed `retry_attempt` row — this event type is not emitted to the JSONL file. Retry bookkeeping goes through pino logs, not the workflow log file - Added an explicit callout that loop_iteration_started / loop_iteration_completed (and other emitter-only events) go through the workflow event emitter + DB workflow_events table, NOT the JSONL file. Pointed readers to the DB or Web UI for loop-level detail. This distinguishes the two parallel event systems — easy to conflate (store.ts:11-17 uses _started/_completed/_failed for the DB side, logger.ts uses _start/_complete/_error for JSONL) - Fixed the "all failed events" jq recipe: .event → .type and _failed → _error - Minor cleanup: the inline "tool_use events" mention in the "running forever" section said the wrong event name — updated to "tool or assistant events in the tail" Grounded in packages/workflows/src/logger.ts (canonical JSONL event shape) and packages/workflows/src/store.ts (the parallel DB event naming, which the reviewer correctly flagged as different and worth keeping distinct). * fix(skill): two stragglers from the code-reviewer audit Cleanup of two references that slipped through the earlier C1 and C3 fixes: - references/troubleshooting.md:126: \`node_failed\` → \`node_error\` (the "Node output is empty" diagnostics section references the JSONL log, which uses the logger.ts enum — not the DB workflow_events table which does use \`node_failed\`). The C3 fix corrected the event table and one jq recipe but missed this inline mention. - references/interactive-workflows.md:106: removed \`archon workflow cancel <run-id>\` (nonexistent CLI subcommand) from the troubleshooting bullet. This was pre-existing before the hardening PR but fell within the C1 remediation scope. Replaced with the correct triage: reject (approval gate only) vs abandon (orphan cleanup, no subprocess kill) vs chat /workflow cancel (actual subprocess termination). Grounded in the same sources as the earlier C1/C3 commits: packages/cli/src/cli.ts:318-485 (no cancel case) and packages/workflows/src/logger.ts:19-30 (JSONL type enum). * feat(skill): point to archon.diy as the canonical docs source The skill had no reference to archon.diy (the live docs site built from packages/docs-web/). Several reference files said "see the docs site" without naming the URL, leaving the agent to guess or grep the repo for the hostname. An agent with the skill loaded should know that when the distilled reference pages don't cover a case, the full canonical docs are one WebFetch away. SKILL.md: new "Richer Context: archon.diy" section between Routing and Running Workflows. Covers: - When to reach for the live docs (longer examples, tutorial framing, features the skill only mentions in passing, "where's that documented?" user questions) - URL map — 13 starting points covering getting-started, book (tutorial series), guides/ (authoring + per-node-type + per-node-feature), reference/ (variables, CLI, security, architecture, configuration, troubleshooting), adapters/, deployment/ - Precedence: skill refs first (context-cheap, tuned for agents), docs site as escalation. Prevents agents defaulting to WebFetch when a local skill ref already covers the answer Also upgrades the 5 existing generic "docs site" mentions across reference files to concrete archon.diy URLs with anchor fragments where helpful: - good-practices.md: Inline sub-agents pattern → archon.diy/guides/ authoring-workflows/#inline-sub-agents - troubleshooting.md: "Install page on the docs site" → archon.diy/ getting-started/installation/ - workflow-dag.md: "Workflow Description Best Practices" → anchor link; sandbox schema reference → archon.diy/guides/authoring-workflows/ #claude-sdk-advanced-options - repo-init.md: Security Model reference → archon.diy/reference/ security/#target-repo-env-isolation (deep-link into the section that covers the <cwd>/.env strip behavior) URL source of truth: astro.config.mjs:5 (site: 'https://archon.diy'). URL structure mirrors packages/docs-web/src/content/docs/<section>/ <page>.md — verified by the 62 pages the docs build produces.
Anthropic's Opus 4.7 landed 2026-04-16; on the Anthropic API, opus / opus[1m] now resolve to 4.7 with a 1M context window at standard pricing. Using the alias instead of the hard-pinned claude-opus-4-6[1m] lets bundled default workflows auto-track the recommended Opus version. No explicit effort is set, so nodes inherit the per-model default (xhigh on 4.7, high on 4.6).
* fix(workflow): migrate piv-loop plan handoff to $ARTIFACTS_DIR (#1380) The create-plan node used a relative path (.claude/archon/plans/{slug}.plan.md) that the AI agent would sometimes write to a different location, breaking all downstream nodes that glob for the plan file. Migrated all plan/progress file references to $ARTIFACTS_DIR/plan.md and $ARTIFACTS_DIR/progress.txt, matching the pattern used by archon-fix-github-issue and other workflows. Changes: - Replace slug-based plan path with $ARTIFACTS_DIR/plan.md in create-plan node - Replace ls -t glob discovery with direct $ARTIFACTS_DIR/plan.md reads in refine-plan, code-review, and fix-feedback nodes - Replace empty-string guard with file-existence check in implement-setup bash - Migrate progress.txt references in implement loop to $ARTIFACTS_DIR/ - Add explicit plan/progress paths in finalize node - Regenerated bundled-defaults.generated.ts Fixes #1380 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix(workflow): address review findings in archon-piv-loop - Rename 'Step 2: Write the Plan' to 'Step 2: Plan File Location' to eliminate the duplicate heading that collided with Step 3's identical title in the create-plan node - Guard implement-setup against a 0-task plan file: exit 1 with a clear error when no '### Task N:' sections are found, preventing a silent no-op implement loop - Remove 2>/dev/null from code-review commit so pre-commit hook failures and other stderr are visible to the agent instead of silently swallowed - Replace '|| true' on git push in finalize with an explicit WARNING echo so push failures (auth, upstream conflict, no remote) surface to the agent rather than being silently ignored - Regenerate bundled-defaults.generated.ts Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * chore(workflows): regenerate bundled defaults to match opus[1m] alias The bundle was stale relative to the YAML sources after #1395 merged — check:bundled was failing CI. Regenerated; no YAML edits. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…cutor (#1403) PIV Task 1: Adds three new tests in a dedicated describe block 'executeDagWorkflow -- final status derivation' covering the anyFailed branch (dag-executor.ts ~line 2956) that previously had no direct test: - one success + one independent failure calls failWorkflowRun (not completeWorkflowRun) - multiple successes + one failure calls failWorkflowRun (not completeWorkflowRun) - trigger_rule: none_failed skips dependent node but anyFailed still marks run failed Fixes #1381.
New reference for the archon skill: a single-glance lookup of which parameter works on which node type, an intent-based "how do I..." table, a consolidated silent-failure catalog, and an inline agents: section (previously only referenced via archon.diy). Purpose is complementary, not duplicative: - workflow-dag.md remains the authoring guide - dag-advanced.md remains the hooks/MCP/skills/retry deep-dive - good-practices.md remains the patterns and anti-patterns - parameter-matrix.md is the grep-this-first lookup when you know the outcome you want but not which field gets you there Also registers the new reference in SKILL.md routing table.
Add explicit references to .github/PULL_REQUEST_TEMPLATE.md in both CONTRIBUTING.md and CLAUDE.md, plus a reminder to link issues with Closes/Fixes/Resolves so they auto-close on merge. Repo-triage runs were flagging dozens of partially-filled or unlinked PRs each cycle.
…riage (#1428) * feat(workflows): add maintainer-standup workflow for daily PR/issue triage Daily morning briefing that pulls origin/dev, triages all open PRs and assigned issues against direction.md, and surfaces progress vs. the previous run. Designed for live-checkout use (worktree.enabled: false) so it can read its own state. Layout under .archon/maintainer-standup/: - direction.md (committed) — project north-star: what Archon IS / IS NOT. Drives PR P4 polite-decline classification with cited clauses. - README.md / profile.md.example — setup docs and template for new maintainers. - profile.md, state.json, briefs/YYYY-MM-DD.md — gitignored, per-maintainer. Engine: - 3 parallel gather scripts in .archon/scripts/maintainer-standup-*.ts (git-status, gh-data, read-context) — bun runtime, JSON stdout. - Synthesis node: command file with output_format schema for { brief_markdown, next_state }. - Persist node: tiny inline bun script writes both to disk. Run-to-run continuity: state.json carries observed_prs/issues snapshots, so the next run can detect what merged, what closed, what the maintainer shipped, and which carry-over items aged past N days. Also adds .archon/** to the ESLint global ignore list (matches the existing .claude/skills/** pattern) since .archon/ is user content and not part of any tsconfig project. * fix(maintainer-standup): address CodeRabbit review on #1428 - gh-data: bump --limit 100 → 1000 on all_open_prs and warn loudly when the cap is hit; preserves the observed_prs invariant the next-run "resolved since last run" diff depends on. (CodeRabbit critical) - maintainer-standup.md: clarify P1 CI signal — the gathered payload only carries mergeStateStatus, not statusCheckRollup; for borderline P1s, drill in via `gh pr checks <n>`. (CodeRabbit minor) - workflow.yaml persist: write briefs under local YYYY-MM-DD (sv-SE locale) instead of UTC ISO date, so an evening run doesn't file tomorrow's brief and break recent_briefs lookups. (CodeRabbit minor) - workflow.yaml persist: wrap state/brief writes in try/catch; on failure dump brief_markdown and next_state to stderr so a 5-minute Sonnet synthesis isn't lost to a transient disk error. (CodeRabbit minor) - gh-data + git-status: switch from execSync (shell-string) to execFileSync (argv array) for git/gh invocations. Defense-in-depth against shell metacharacters in values that pass through (esp. the gh_handle from profile.md). (CodeRabbit nitpick)
Add optional `tags: string[]` to `workflowBaseSchema`. Explicit values take precedence over keyword inference; `tags: []` suppresses inference end-to-end; omitting the field falls back to inference (backwards compatible). Non-array values warn-and-ignore matching the sibling `worktree`/`additionalDirectories` patterns.
…ows under maintainer/ (#1430) * feat(workflows): add maintainer-review-pr and group maintainer workflows under .archon/workflows/maintainer/ Adds the maintainer-review-pr workflow — a Pi/Minimax-based PR triage flow that gates on direction alignment, scope focus, and PR-template quality before doing any deep review. If the gate clears, runs the five review aspects (code/error-handling/test-coverage/comment-quality/ docs-impact) as parallel Archon nodes and auto-posts a synthesized review comment. If the gate fails (direction conflict, multiple concerns, sprawling scope), drafts a polite-decline comment and pauses for the maintainer's approval before posting. Reorganizes the existing maintainer-standup workflow into the same subfolder so all maintainer-facing workflows live together. Subfolder grouping is supported by the workflow loader (1 level deep, resolution by filename). What lands: - .archon/workflows/maintainer/maintainer-standup.yaml (moved from .archon/workflows/maintainer-standup.yaml) - .archon/workflows/maintainer/maintainer-review-pr.yaml (new) - .archon/commands/maintainer-review-{gate,code-review,error-handling, test-coverage,comment-quality,docs-impact,synthesize,report}.md (new, Pi-tuned variants of the existing review-agent commands so they avoid Claude-only Task / sub-agent patterns) Pi/Minimax integration: - Uses provider: pi, model: minimax/MiniMax-M2.7 — verified via the e2e-minimax-smoke test that Pi correctly routes to Minimax (session jsonl confirms provider=minimax) and that Pi's best-effort output_format parser handles the gate's nested schema. - Two test runs landed real comments: a direction-decline on PR #1335 and a deep-review on PR #1369. Both were posted to GitHub via the workflow's gh pr comment node. * chore(workflows): also group repo-triage under .archon/workflows/maintainer/ repo-triage is the third maintainer-facing workflow alongside maintainer-standup and maintainer-review-pr; group it in the same subfolder for consistency. Subfolder resolution is by filename so the workflow name is unchanged.
…r unmapped providers (#1284) Closes #1096. - Switch Pi provider model lookup from pi-ai's getModel() (static catalog only) to ModelRegistry.create(authStorage).find() so user-configured custom models in ~/.pi/agent/models.json (LM Studio, ollama, llamacpp, custom OpenAI-compatible endpoints) are discoverable. - Remove the local lookupPiModel helper. - For env-var-mapped providers (anthropic, openai, etc.) still throw with a pi /login hint when credentials are missing. For unmapped providers, log pi.auth_missing at info and continue so local models that don't need credentials work without ceremony. - Surface modelRegistry.getError() in the not-found message and emit pi.model_not_found so users debugging custom-provider configs see the real cause (e.g. missing baseUrl in models.json). - Guard AuthStorage.create() and ModelRegistry.create() with try/catch so a malformed ~/.pi/agent/auth.json surfaces with Pi-framed context instead of a raw SDK stack trace. - Document the credential-free path for local providers in ai-assistants.md. Co-authored-by: Matt Chapman <Matt@NinjitsuWeb.com>
…add e2e-minimax-smoke (#1431) * chore(workflows): group all smoke-test workflows under .archon/workflows/test-workflows/ Move the 7 existing e2e-*.yaml smoke tests plus the new e2e-minimax-smoke test into a dedicated subfolder. Subfolder grouping is supported by the workflow loader (1 level deep, resolution by filename) so workflow names are unchanged. Mirrors the .archon/workflows/maintainer/ split landing in #1430. Also adds e2e-minimax-smoke.yaml — a sanity check that Pi correctly routes to Minimax M2.7 via the user's local pi auth, and that Pi's best-effort output_format parser handles a small nested schema. Asserts routing by reading the most recent Pi session jsonl rather than asking the model to self-identify (LLMs are unreliable narrators about their own identity, especially when Pi's system prompt mentions other providers as defaults). * fix(e2e-minimax-smoke): address CodeRabbit review on #1431 - Widen find window from -mmin -3 to -mmin -10. The smoke's three Pi nodes plus the assert can collectively run several minutes on slow networks; 3 minutes was tight enough to false-FAIL on a healthy run. (CodeRabbit minor) - Drop non-deterministic `head -1` over `find` output. find doesn't guarantee any order; on a tie, the wrong file would be picked. Now iterates all matching sessions and breaks on first one carrying the routing signal — any match is sufficient evidence. (CodeRabbit minor) - Replace single-regex `'"provider":"minimax".*"modelId":"MiniMax-M2.7"'` with two separate greps joined by `&&`. JSON field order isn't part of Pi's contract; a future Pi release reordering `provider` and `modelId` in the model_change event would silently false-FAIL the original pattern. The new check is order-independent. (CodeRabbit major)
Six findings, two majors and four minors/nitpicks: - gate.md L17 vs L77: resolved conflicting input-source instructions. Body claimed "all inline, no extra fetch" while a later phase permitted reading PULL_REQUEST_TEMPLATE.md. Now: explicit "one allowed extra read" callout in Phase 1 + matching wording in Gate C. (CodeRabbit major) - gate.md fenced blocks: added missing language identifiers (text/json/ markdown) to satisfy markdownlint MD040. (CodeRabbit minor) - gate.md L155 + read-context.ts: deterministic clock. The 3-day deadline was anchored to prior_state.last_run_at, which can be stale and produce past-dated deadlines. Moved both today and deadline_3d into the read-context.ts output (computed via sv-SE locale → ISO date in local time) and instructed the gate to use $read-context.output.deadline_3d directly. LLMs are unreliable at calendar arithmetic; this avoids it entirely. (CodeRabbit major) - maintainer-review-pr.yaml fetch-diff: dropped 2>/dev/null on gh pr diff so auth / network / deleted-PR failures fail the node instead of feeding an empty diff to the gate. Empty-but-successful diff (PR has no changes) is now an explicit marker the gate can detect. (CodeRabbit minor) - maintainer-review-pr.yaml approve-unclear: added capture_response: true so the maintainer's approve comment flows to the report node. Reject reasoning is already captured by Archon's run record. (CodeRabbit minor) - maintainer-review-pr.yaml post-decline + report.md: the gh pr edit --add-label call previously swallowed all errors with || true and the report still claimed the label was applied. Now writes applied/skipped to $ARTIFACTS_DIR/.label-applied + the gh stderr to .label-error so the report can describe the actual outcome. (CodeRabbit nitpick)
…ume (#1435) * fix(workflows): approval gate bypass after reject-with-redraft on resume When an approval node was rejected with on_reject.prompt, the synthetic PromptNode built to run the on_reject prompt reused the approval gate's own node ID. executeNodeInternal then wrote a node_completed event with that ID, causing getCompletedDagNodeOutputs to treat the gate as already completed on the next resume — bypassing the human gate entirely. Fix: give the synthetic node the ID `${node.id}:on_reject` so its node_completed event has a distinct step_name that won't match the approval gate slot in priorCompletedNodes. Adds a regression test asserting no node_completed event with the approval gate's ID is written during on_reject execution. Fixes #1429 * test(workflows): add positive assertion and SSE side-effect comment for on_reject synthetic node Add complementary positive assertion to the regression test to verify that node_completed is written exactly once with step_name 'review:on_reject', ensuring future refactors that suppress the event entirely would be caught. Add inline comment in executeApprovalNode documenting the known SSE side-effect: node_started/node_completed events with nodeId='review:on_reject' flow through the SSE pipeline into the web UI, resulting in a transient phantom node in the execution view. This is cosmetic-only — the human gate contract is preserved. * simplify: reduce duplicate cast pattern in on_reject test assertions
…e checkout (#1438) * feat(workflows): add mutates_checkout field to skip path-lock for concurrent runs Add `mutates_checkout: boolean` (optional, default true) to the workflow schema. When set to false, the executor skips the path-exclusive lock that serializes all runs on the same working path, allowing N concurrent runs on the same live checkout. The primary use case is `maintainer-review-pr`, which reads shared state but writes only to per-run artifact paths and GitHub PR comments — two parallel reviews of different PRs should not fail with "Workflow already active on this path". Changes: - `schemas/workflow.ts`: add optional `mutates_checkout` field - `loader.ts`: parse and propagate the field (warn-and-ignore on invalid values) - `executor.ts`: wrap path-lock guard in `if (workflow.mutates_checkout !== false)` - `executor.test.ts`: two new tests in the concurrent-run guard suite - `maintainer-review-pr.yaml`: opt in with `mutates_checkout: false` * test(workflows): add loader tests for mutates_checkout parsing - Add 5 tests covering false, true, omitted, and invalid (string "yes") values - Invalid non-boolean values are silently dropped with warn — now explicitly tested - Remove the // end mutates_checkout guard trailing comment (no precedent in file) - Clarify loader comment: "parse/warn pattern" not "warn-and-ignore pattern" to avoid implying the return style matches interactive * simplify: collapse nodeType/aiFields pair into single nonAiNode object in parseDagNode
…es (#1434) * docs: replace String.raw with direct assignment in script node examples String.raw`$nodeId.output` fails silently when substituted output contains a backtick, terminating the template literal early and producing cryptic parse errors. JSON is valid JS expression syntax, so direct assignment is safe for all valid JSON values including those with backticks. - Replace String.raw pattern in dag-workflow.yaml example - Replace String.raw pattern in archon-workflow-builder.yaml template - Add CAUTION bullet in workflow-dag.md Script Node section - Add Silent Failures item #14 in parameter-matrix.md - Add Starlight caution aside in script-nodes.md - Extend script bodies bullet in variables.md - Regenerate bundled-defaults.generated.ts Fixes #1427 * docs: fix Rule 6 in generate-yaml prompt to distinguish bun vs uv patterns Rule 6 still referenced JSON.parse after the example was updated to direct assignment, creating a contradiction for the AI code generator. Update the prose to explicitly distinguish TypeScript/bun (direct assignment) from Python/uv (json.loads), matching the updated embedded example.
…s/experimental/ Move two repo-scoped workflows that were sitting untracked at the workflow root into a dedicated subfolder. Subfolder grouping is supported by the loader (1 level deep, resolution by filename), so workflow names are unchanged and the /release skill still resolves archon-release correctly. Files moved: - archon-fix-github-issue-experimental.yaml — Path-A variant of the issue-fix workflow used today to land #1434, #1435, #1438. - archon-release.yaml — the live release workflow used by the /release skill end-to-end (validate -> binary smoke -> version bump -> changelog -> approval -> commit -> PR -> tag -> Homebrew formula update).
…des (#1387) executeBashNode previously only merged explicit envVars on top of process.env. The three well-known workflow directories (artifactsDir, logDir, baseBranch) were passed as function parameters and used for compile-time substitution of $ARTIFACTS_DIR / $LOG_DIR / $BASE_BRANCH in the script body, but were never added to the subprocess environment. As a result, any script that relied on shell-runtime expansion — e.g. JSON_FILE="${ARTIFACTS_DIR}/foo.output.json" inside a heredoc, an inherited helper script, or a `bash -c` subshell — saw the variable unset and silently fell back to its default (typically an empty string or "."), writing artifacts to the workflow cwd instead of the nominal artifacts directory. Always build subprocessEnv from process.env plus the three well-known directories, then allow explicit envVars to override. Compile-time substitution behavior is unchanged; existing scripts that do not reference these variables are unaffected; user-supplied envVars still win on conflict.
…1426) * fix(workflow): substitute \$nodeId.output refs in approval messages Approval node messages were emitted as raw strings, bypassing the substituteNodeOutputRefs() pass that prompt/bash/loop/cancel nodes all run. This made interactive workflows like atlas-onboard show literal "\$gather-context.output.repo_name" placeholders to humans at HITL gates, leaving them unable to know what they were approving. Fix: rendered the approval.message through substituteNodeOutputRefs once at the top of the standard approval gate path, then used the resolved string in all 4 emission sites (safeSendMessage, createWorkflowEvent, pauseWorkflowRun, event-emitter). Test: new dag-executor.test case wires a structured-output upstream node into an approval node and asserts pauseWorkflowRun receives the substituted message ("Repo: hcr-els | App: CCELS | Port: 3012") rather than the literal placeholders. Repro: any workflow with an approval node whose message references \$nodeId.output[.field]. Observed in the wild on atlas-onboard's confirm-context HITL gate. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * test(workflow): extend approval-substitution test to cover all 4 emission sites Per CodeRabbit review: the original test only verified pauseWorkflowRun received the substituted message, but the fix touches 4 emission sites. A future regression at safeSendMessage / createWorkflowEvent / event-emitter would silently leave the test passing while users still saw raw $node.output placeholders. Adds two additional assertions: - platform.sendMessage prompt contains substituted message + does NOT contain literal $gather-context.output placeholders - The persisted approval_requested workflow event's data.message is substituted Event-emitter assertion deferred (no existing pattern for spying on the global emitter in this test file). Two of three secondary surfaces covered closes the practical regression risk — both are user-visible (chat prompt + audit-log event); the emitter is internal only. Test count: 7 pass / 22 expect() (was 18). Full suite 193 pass / 353 expect() — no regressions. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
#1367) * feat(workflows): expose $LOOP_PREV_OUTPUT in loop node prompts (#1286) Adds a new substitution variable that carries the previous loop iteration's cleaned output into the next iteration's prompt. Empty on iteration 1; the prior iteration's output (after stripCompletionTags) on iteration 2+. Why: fresh_context: true loops have no way to reference what the previous pass produced or why it failed without dragging the full session forward. $LOOP_PREV_OUTPUT closes that gap with zero session-cost — same trust boundary as $nodeId.output, no new external surface. Changes: - packages/workflows/src/executor-shared.ts: substituteWorkflowVariables accepts a 10th positional loopPrevOutput arg and substitutes $LOOP_PREV_OUTPUT (defaults to ''). - packages/workflows/src/dag-executor.ts: executeLoopNode passes lastIterationOutput on iteration 2+ (and explicit '' on iteration 1 / the first iteration of an interactive resume, since lastIterationOutput is a per-call variable that does not survive resume metadata). - Unit tests: 3 new cases in executor-shared.test.ts. - Integration tests: 2 new cases in dag-executor.test.ts verifying the prompt sent to the AI on iter 1 vs iter 2, and that the value reflects cleaned output (no <promise> tags). - Docs: variables.md, loop-nodes.md (new "Retry-on-failure" pattern), CLAUDE.md variable reference. Backward compatibility: prompts that don't reference $LOOP_PREV_OUTPUT are unaffected. All 843 workflow tests + type-check + lint + format:check + bun run validate pass locally. * docs: address coderabbit review on variables/loop-nodes - variables.md: include $LOOP_PREV_OUTPUT in substitution-order list and availability table to match the new variable row at line 30 - loop-nodes.md: document the interactive-resume exception where the first iteration after an approval-gate resume still receives an empty $LOOP_PREV_OUTPUT regardless of iteration number (per dag-executor.ts L1781-1783 where i === startIteration always clears prev output) * docs(changelog): add Unreleased entry for $LOOP_PREV_OUTPUT (#1367 review) * test(loop): add resume-from-approval integration test for $LOOP_PREV_OUTPUT (#1367 review) Per maintainer-review-pr suggestion (Wirasm): two-call integration test covering the resume-from-approval scenario. - Call 1: fresh interactive loop pauses at the gate after iteration 1 and asserts $LOOP_PREV_OUTPUT substitutes to empty on iter 1 (no prior output) plus the gate pause is recorded. - Call 2: resumed run with metadata.approval populated. The first resumed iteration must substitute $LOOP_PREV_OUTPUT to '', NOT to the paused run's iter-1 output (which lived in a different process and is not persisted). $LOOP_USER_INPUT still flows through as normal. Locks the documented invariant at dag-executor.ts:1769-1772. --------- Co-authored-by: voidborne-d <DottyEstradalco@allergist.com>
…1457) The brief was missing a key signal — when contributors reply on PRs or issues, the maintainer wouldn't see it explicitly. Empirically reviewed PR replies were buried under aggregate updatedAt timestamps with no indication of WHO replied or WHAT they said. This adds a new "Replies waiting on you" section to the daily brief, sourced from two paginated GitHub API calls scoped by since=last_run_at: - /repos/{o}/{r}/issues/comments PR + issue conversation comments - /repos/{o}/{r}/pulls/comments inline code-review comments Filters applied: - Skip the maintainer's own comments (gh_handle from profile.md) - Skip GitHub bot accounts (login ending in [bot]) — coderabbitai, chatgpt-codex-connector, dependabot, etc. They post a constant churn of automated review tooling that drowns out human replies; the maintainer wants the latter. Output is grouped by PR/issue number with kind classification: - issue comment on a non-PR issue - pr_conversation PR conversation-level comment - pr_review inline code-review comment (most actionable — usually needs a code-level response, so kind upgrades to pr_review whenever review comments arrive on a PR that also has conversation ones) Sorted by recency (newest reply first). Synthesizer reads gh-data.output.replies_since_last_run and renders a section. Verified on a backdated state.json (last_run_at = yesterday morning): 22 human replies on 22 PRs/issues, bot noise filtered (32 → 22 after the [bot] filter). Surfaces exactly the contributor responses to yesterday's review comments and direction questions.
The maintainer-standup brief had no signal for "I already triaged that
PR via maintainer-review-pr 2 days ago" — it just kept listing reviewed
PRs in P1-P4 with no acknowledgement of prior work. Result: maintainer
ends up re-skimming the same PR several mornings in a row.
This adds a shared persistent state file at:
.archon/maintainer-standup/reviewed-prs.json (gitignored, per-maintainer)
shape:
{
"1338": {
"reviewed_at": "2026-04-27T16:34:57Z",
"gate_verdict": "review", // review | decline | needs_split | unclear
"run_id": "..."
},
...
}
Three pieces:
1. WRITER — new `record-review` script node in maintainer-review-pr.yaml,
runs after whichever branch fired (post-review / post-decline /
approve-unclear) with trigger_rule: one_success. Inline bun script;
reads $gate.output.verdict, $ARTIFACTS_DIR/.pr-number, and
$WORKFLOW_ID; appends/upserts the entry. report node now depends on
record-review so the state write happens before the run completes.
2. READER — read-context.ts loads reviewed-prs.json into a new
reviewed_prs field on the standup gather output. Same pattern as
prior_state and recent_briefs.
3. SURFACE — maintainer-standup command file gets a Phase 2h rule:
when listing PRs in P1-P4 / Polite-decline sections, append:
- "✓ reviewed Nd ago" for review-branch entries
- "✓ declined Nd ago" for decline / needs_split branches
- "✓ triaged Nd ago (unclear)" for unclear branch
and a STALENESS marker — compare reviewed_at to PR's updatedAt; if
contributor pushed since the prior review, append
"⚠ contributor pushed since" so the maintainer knows the prior pass
may need to be re-run.
Plus a one-shot backfill script:
.archon/scripts/maintainer-standup-backfill-reviews.ts
Scans the maintainer's gh comments in the last 7 days, pattern-matches
"## Review Summary" / direction-clause-citation / split-up wording, and
populates reviewed-prs.json. Idempotent; existing entries (from real
workflow runs) take precedence over backfilled ones (the writer-node
record is more authoritative than a body-pattern guess). Uses 64MB
maxBuffer on the gh exec because --paginate over 7 days of an active
repo's comments easily exceeds Node's default 1MB.
Backfill verified: 363 comments scanned, 18 matched, 17 unique PRs
populated — exactly the 17 PRs we reviewed via the workflow yesterday.
The new state file is gitignored alongside the existing per-maintainer
files (profile.md, state.json, briefs/).
…1460) Both SDKs were ~30 patch releases behind. Validation suite passes (type-check, lint, format, tests across all 10 packages) without code changes. The only sustained Claude SDK behavior change in the range — v0.2.111's options.env overlay/replace flap, since reverted to overlay — is a no-op for Archon, which already passes { ...process.env } as the SDK env.
…1461) * fix(claude): stop passing --no-env-file to native binary in dev mode The Claude Agent SDK switched from shipping `cli.js` inside the package to per-platform native binaries via optional deps somewhere in the 0.2.x series. As of 0.2.121 there is no `cli.js` in the SDK package; dev mode resolves to `@anthropic-ai/claude-agent-sdk-darwin-arm64/claude` (Mach-O). That native binary rejects `--no-env-file` with `error: unknown option '--no-env-file'` and the subprocess exits 1. `shouldPassNoEnvFile` was returning true on `cliPath === undefined` on the assumption that "dev mode = JS executable run via Bun". That assumption is dead. Tighten the predicate to only return true on an explicit `.js` suffix, so we only emit the flag when the SDK is going to spawn a Bun-runnable script. CWD `.env` leak protection is unaffected. `stripCwdEnv()` in `@archon/paths` (#1067) deletes Bun-auto-loaded `.env`/`.env.local`/ `.env.development`/`.env.production` keys from `process.env` at every Archon entry point before any subprocess is spawned. The native Claude binary does not auto-load `.env` from its cwd either. `--no-env-file` was belt-and-suspenders for the JS-via-Bun case only. Verified end-to-end with a sentinel: added a unique `ARCHON_LEAK_SENTINEL_$$` to Archon's `.env`, ran e2e-claude-smoke with a bash probe checking the subprocess env. stderr shows `[archon] stripped 23 keys from /Users/rasmus/Projects/cole/Archon (.env, .env.local)` — sentinel was deleted. Bash node prints `PASS: simple='4', no sentinel leak`. Workflow completes cleanly, no `--no-env-file` rejection from the SDK binary. bun run validate: green across all 10 packages. * fix(claude): address review on #1461 (stale docs + test gaps) Critical: file-level JSDoc at provider.ts:18 still claimed dev mode resolves cli.js. Updated to reflect SDK 0.2.x's switch to per-platform native binaries. Important: security.md still listed --no-env-file as item 2 of target-repo .env isolation. Scoped that bullet to legacy Bun-runnable JS entry points and called out that native binaries don't auto-load .env from cwd. Added an Unreleased Fixed entry to CHANGELOG.md. Updated binary-resolver.ts JSDoc title that referenced cli.js. Polish: widened the predicate to accept .mjs and .cjs (also Bun-runnable JS — matches the SDK's own internal extension list). Dropped the redundant `passesNoEnvFile` log field that mirrored `isJsExecutable`. Added unit cases for .mjs/.cjs (now true) and .ts/.tsx/.jsx (deliberately false — never SDK entry points). Added an integration test that mocks resolveClaudeBinaryPath to return a .js path and asserts executableArgs: ['--no-env-file'] flows through buildBaseClaudeOptions all the way to the SDK call — catches future regressions in the conditional spread. bun run validate: green across all 10 packages.
* refactor(workflows): trust the SDK for model validation Drops cross-provider model inference and hard-coded model allow-lists. The string a workflow author writes in `model:` is forwarded to the SDK unchanged; the SDK and its API decide whether the model exists. Provider identity is the only thing Archon validates at load time — typos like `provider: claud` are caught early; everything else fails at runtime through the SDK's normal error path. Why this matters: a recent run on Sasha showed `provider: claude` + `model: opus[1m]` getting silently routed to Codex (because Codex's isModelCompatible was defined as the complement of Claude's, so anything not literally `sonnet|opus|haiku` matched). Codex then rejected the model as a `⚠️ ` system warning and the node "completed" in 2.1 seconds with empty output, after which the workflow opened a hallucinated PR. Three stacked bugs and two amplifiers; this commit removes all five. Changes: - Delete model-validation.ts entirely (inferProviderFromModel and isModelCompatible are gone). Drop the matching field from ProviderRegistration and from the claude/codex/pi entries. - Replace the resolver in executor.ts and dag-executor.ts (both the per-node and per-loop paths) with a flat `node.provider ?? workflow.provider ?? config.assistant`. Model never influences provider selection; load-time validation is just isRegisteredProvider on the resolved provider id. - Remove the dag-node Zod superRefine that recomputed model-compat — load-time provider validation moved to loader.ts. - Codex provider: stream loop now matches Claude's contract. error events that aren't followed by turn.completed yield `result.isError: true` (subtype `codex_stream_incomplete`) so the dag-executor's existing isError path catches them. turn.failed becomes `codex_turn_failed` with the same shape. Iterator close without a terminal event is itself a fail-stop. MCP-client errors remain filtered (Codex retries those internally). - dag-executor: AI nodes that exit the streaming loop with empty assistant text and no structured output now fail with `dag.node_empty_output` instead of completing silently — the Sasha bug's final amplifier. Bash/script/approval nodes are unaffected. Tests: model-validation.test.ts and isPiModelCompatible block deleted; codex provider tests rewritten to assert the new fail-stop contract; dag-executor empty-output test flipped to assert failure; new tests cover (a) loader rejecting unknown provider, (b) loader accepting any model string with a known provider, (c) executor passing provider+model through without re-routing, (d) executor throwing on unknown provider, (e) Codex synthesizing fail-stop on iterator close. Two cost-tracking tests adjusted to yield non-empty assistant text since their intent was cost accumulation, not empty-output handling. bun run validate: green (check:bundled, type-check, lint --max-warnings 0, format:check, all packages' test suites — 0 fail). End-to-end smoke (.archon/workflows/test-workflows/): - e2e-deterministic: PASS (engine healthy) - e2e-codex-smoke: PASS (Codex sendQuery + structured output work) - e2e-claude-smoke: FAIL with `error: unknown option '--no-env-file'` — this is a regression from the SDK 0.2.121 bump (#1460), not from this redesign. The Claude provider source is unchanged on this branch. To be fixed separately. * fix(workflows): address review on #1463 Critical: - C1: empty-output guard now skips idle-timeout completions. The on-screen message says "completed via idle timeout"; flipping that to a failure contradicted the user-facing log. Added !nodeIdleTimedOut to the guard. - C2: per-node provider identity is now validated at YAML load time. Loader iterates dagNodes after parsing and rejects any unknown provider id with "Node 'X': unknown provider 'Y'. Registered: ...". The dag-executor's runtime check stays as defense-in-depth. Important: - I1: CHANGELOG entry under [Unreleased] > Changed describing the resolver redesign + an explicit migration line for workflows that relied on cross-provider model inference. - I2: restored the dropped mockLogger.error('turn_failed') assertion in the turn.failed-without-error-message test. - I3: empty-output test now also asserts store.failWorkflowRun was called, matching the parallel error_max_budget_usd test pattern. - I4: new test that proves a node yielding zero assistant text but a valid structuredOutput is treated as a successful completion (not caught by the empty-output guard). - I5: rewrote the post-loop comment in codex/provider.ts to be precise about which dag-executor branch catches the synthesized result chunk (the throwing msg.isError branch, distinct from the empty-output guard's { state: 'failed' } return). - I6: removed PR-era "redesign" / "Sasha workflow" references from three test-file comments. - I7: docs sweep for the deleted isModelCompatible field — six files updated (CLAUDE.md, two docs guides, quick-reference, contributing guide, architecture reference). Polish: - S3: dropped the dead sawTerminal flag in streamCodexEvents — both terminal branches `return`, so reaching the post-loop block always means no terminal fired. Pure simplification. - S4: dropped parsePiModelRef and PiModelRef from community/pi/index.ts exports. The parser is consumed only by Pi's provider.ts; making it package-internal narrows the public surface. - S6: new Codex test for the bare-stream-close case (zero events, iterator just ends) — locks in the default fallback message used when no captured non-MCP error is available. - S7: new dag-executor test for per-node unknown-provider at runtime. Bypasses the loader to exercise resolveNodeProviderAndModel's throw, asserts the node_failed event carries the "unknown provider 'claud'" detail (the workflow-level fail message is a generic summary). bun run validate green across all 10 packages. * fix(workflows): address CodeRabbit review on #1463 Two real issues from CodeRabbit's automated pass on db95e8a: 1. Empty-output fail-stop now applies to loop iterations too. The single-shot AI-node guard at executeNodeInternal only covered prompt/command nodes; executeLoopNode has its own streaming path, so a provider that closed cleanly with zero content could pause an interactive loop with a blank gate or burn the full max_iterations budget. Mirrors the contract of the single-shot guard: `fullOutput.trim() === '' && !iterationIdleTimedOut` fails the iteration with a `loop_iteration_failed` event carrying a clear error. Idle-timeout exits remain exempt for the same reason as single-shot nodes — the on-screen "completed via idle timeout" message would otherwise contradict the failure. 2. Unknown loop providers now throw instead of return-failed. The early-return path bypassed the layer dispatch's outer catch at line 2870, so loop nodes with an invalid per-node `provider:` field skipped the standard `node_failed` event, the user-facing message, and the pre-execution log entry. Throwing reuses the common failure path — same shape as resolveNodeProviderAndModel uses for non-loop nodes. Both align with CLAUDE.md's "fail fast, explicit errors, never silently swallow" principle. The third CodeRabbit finding (boundary violation for `@archon/providers` import in loader.ts) is consistent with existing precedent — `dag-executor.ts`, `executor.ts`, and `validator.ts` already import from the same path; the runtime contract (every entrypoint bootstraps the registry before parseWorkflow runs) is already enforced in tests and documented at `loader.test.ts:31`. bun run validate green across all 10 packages.
… crash on missing source (#1394) The 18 top-level `import … with { type: 'text' }` statements in `bundled-skill.ts` resolve at module load. For `bun build --compile` that's build time, so the binary embeds the strings and works regardless of any on-disk skill files. For `bun link` (linked-source) installs that's every `archon` invocation — including `archon --help`, which doesn't even use the skill content. If any of the 18 source files are missing or moved, the import fails and the CLI cannot start at all. The skill content is data the binary deploys via `archon setup`, not data the CLI needs at runtime. There's only one consumer in production code: `copyArchonSkill()` in `setup.ts`. Moving the import into that function as a dynamic import preserves the compiled-binary behavior (Bun's bundler statically analyses literal-string `import()` and embeds the chunk — verified by grepping the SKILL.md frontmatter out of a freshly compiled binary) while making the linked-source install resilient: only `archon setup` triggers the bundled-skill module load now. Verified: a known skill string appears in the compiled binary 1×, and `archon --help` no longer needs the source files to start. `copyArchonSkill()` becomes async because the dynamic import is a Promise. The single production call site is already in an async function and gets an `await`. The four `setup.test.ts` cases become async too.
…art (#1307) * fix(docker): register safe.directory for all repos on bind-mount restart (#1279) On macOS bind mounts (VirtioFS), host UIDs do not map to the container appuser (1001). Git 2.35.2+ rejects operations with "dubious ownership". The Dockerfile RUN-layer gitconfig is not inherited by bind mounts on restart, and worktree paths are unknown at build time. Add a find loop in docker-entrypoint.sh that dynamically registers every .git directory under /.archon as a safe directory after the chown block. This is idempotent and handles worktrees at any depth. * chore: add -prune to avoid scanning .git internals
…mat persist (#1480) The original maintainer-standup workflow relies on Claude SDK-enforced output_format (one big JSON wrapping the brief markdown + state). It works on Claude but hangs in nested Claude Code sessions (#1067), and Pi/Minimax can't reliably emit a 30KB JSON wrapper around markdown content. Changes: - Drop output_format from synthesize node. Synthesis now emits brief markdown plain, followed by a delimited ARCHON_STATE_JSON_BEGIN/END block. - Replace persist (script:) with bash: that pipes synth output into a new .archon/scripts/maintainer-standup-persist.ts. The script tries the delimiter format first and falls back to the legacy {brief_markdown, next_state} JSON wrapper Pi/Minimax tends to emit regardless of prompt instructions. Either format yields the same brief + state.json on disk. - Add maintainer-standup-minimax.yaml — a 1:1 copy with provider: pi, model: minimax/MiniMax-M2.7 — for use when the Claude variant hangs or when the maintainer wants to spend Pi tokens instead. - Update the maintainer-standup synthesizer prompt: top-of-file output format directive, explicit "do not wrap in JSON object" guidance, delimiter-based example. Tested end-to-end on the Minimax variant against today's PR/issue payload; brief was extracted via the JSON-wrapper fallback path and written to .archon/maintainer-standup/briefs/2026-04-29.md.
…nses (#1440) * fix(providers/pi): tolerate prose preamble in structured-output responses Pi has no SDK-level JSON-mode parameter and Minimax's Anthropic-compat proxy doesn't expose response_format support, so reasoning models like Minimax M2.7 routinely "think out loud" before emitting the JSON we asked for, e.g.: "Now I have all the inputs. Let me evaluate the three gates: **Gate A — Direction alignment**: ... {\"verdict\":\"review\",\"direction_alignment\":\"aligned\",...}" The previous tryParseStructuredOutput stripped fences then JSON.parse'd the whole string, which fails the moment a single character of prose appears before the {. The maintainer-review-pr workflow's gate node was hitting this on ~70% of runs, propagating to 7-of-10 silent failures in a sequential review batch this afternoon. Add two preamble-tolerant fallback tiers: Tier 1 (existing): clean JSON.parse — fast path for compliant models Tier 2 (new): scan backward to last `{`, parse from there — flat JSON with reasoning preamble (Minimax M2.7 case) Tier 3 (new): scan forward to first `{`, parse from there — nested JSON with preamble (Tier 2 lands inside a child object and fails) The existing "trailing-prose-too" failure case (preamble + JSON + postamble) still returns undefined — handling it would require brace- depth tracking and isn't worth the cost. The two new tiers cover the failure mode actually observed in production. Tested against the real Minimax preamble pattern and a synthetic preamble + nested-JSON case. 39 event-bridge tests pass. No SDK changes, no Pi extensions, no Minimax API dependencies. Pure Archon-side parser hardening — backward-compatible and benefits every verbose-preamble model routed through Pi, not just Minimax. * fix(providers/pi): address CodeRabbit review on #1440 - Drop the redundant `firstBrace !== lastBrace` guard on Tier 3. When the only `{` past position 0 is the same one Tier 2 just tried, Tier 3 will re-run JSON.parse on the same input and fail identically — one redundant call accepted for simpler control flow. (CodeRabbit cleanup) - Tighten the Tier 1 comment to drop the model name list (will date) and tighten the Tier 2 comment to scope its claim to "preamble + flat JSON" rather than overgeneralized "nested JSON". (CodeRabbit comment-quality cleanup) - Replace the bare `// Fall through...` comments with `// fall through` inside the catch blocks. Necessary because eslint's `no-empty` rule flags bare `catch {}` blocks. (lint compliance) - Add a regression test for `{` inside a string value that pins the Tier 2 → Tier 3 cascade composition: lastIndexOf lands inside the string brace, JSON.parse rejects the resulting `{ inside","ok":true}` fragment, Tier 3 forward-scans to the JSON object's outer `{` and parses cleanly. Note: the preamble must not itself contain `{`, otherwise Tier 3 lands on it instead of on the JSON object — covered in the test comment so the constraint isn't lost. (CodeRabbit test coverage) - Update `packages/docs-web/src/content/docs/getting-started/ai-assistants.md:378` — the previous "degrades cleanly when the model emits prose" claim is now partially false (preamble-only patterns are recovered). New text enumerates the three handled forms (bare, fenced, prose-preamble) and clarifies the trailing-text-interleaved case still degrades. (CodeRabbit docs-impact) - Add a CHANGELOG entry under `## [Unreleased] / Fixed`. CodeRabbit's "critical" suffix-check suggestion (`cleaned.slice(brace) .trimEnd().endsWith('}')`) is not applied. JSON.parse already rejects trailing non-whitespace, so any successful parse from a `{`-prefixed slice already ends with `}` after trim — the check is tautological in all cases the function actually reaches it. The actual concern (a self-contained `{...}` fragment in preamble that happens to be valid JSON, with no real answer in the response) wouldn't be caught by the suffix check either, since the fragment IS a valid JSON object that ends with `}`. Real defense against that case would require schema validation, which lives at the call site (executor's `structured_output_missing` warning path), not in this parser. Tests: 40 pass / 0 fail. Lint, format, type-check all clean. * fix(providers/pi): drop Tier 2 backward-scan to avoid brace-postamble footgun CodeRabbit flagged a real correctness issue: a backward scan from the last `{` silently returns the wrong JSON object when the response contains a brace-bearing example after the real payload (e.g. `{"actual":1}\n For example: {"x":2}` parses the trailing example instead of failing). This breaks the conservative-failure contract callers rely on. Tier 2 was always strictly worse than Tier 3 anyway: every preamble pattern Tier 2 handled, Tier 3 handles too, and Tier 3 doesn't have the multi-fragment hazard. Dropping Tier 2 entirely removes the footgun and keeps the parser simple (clean parse → forward scan → undefined). Tests updated: - Renamed/clarified existing tests to reference the surviving forward-scan tier instead of the deleted backward-scan tier (no behavior change for the cases they cover — preamble has no extra braces). - Added a regression test for the brace-postamble footgun: input with a trailing example like `{"actual":"value"}\nFor example: {"verdict":"review"}` must return undefined under the new contract.
#1393) * fix(workflows): concise failure messages for bash/script nodes (#1389) When a `bash:` or `script:` node fails, the executor was embedding `err.message` verbatim into the user-visible error. For inline scripts run via `bash -c <body>` / `bun -e <body>`, Node's `promisify(execFile)` puts the entire substituted script body into `err.message`, `err.cmd`, and the first line of `err.stack` — and the Pino log serialized all three, repeating the body ~3× in one structured log line. The actionable diagnostic was buried under kilobytes of echoed source. Route both catch-block default branches through a new `formatSubprocessFailure(err, label)` helper in `executor-shared.ts` that: - strips the `Command failed: <cmd>` prefix line (which carries the body) - prefers `err.stderr` — Bun/bash emit the actionable diagnostic there - tail-truncates at 2 KB with a `…[truncated]` marker - returns a controlled `logFields` subset (`exitCode`, `killed`, `stderrTail`) so Pino never re-serializes `err.message` / `err.stack` / `err.cmd` Also drops the script-node `stderrHint` concatenation — stderr is already handled by the helper, so the previous code appended it twice. Timeout / ENOENT / EACCES branches are preserved verbatim. Fixes #1389 * fix(workflows): address PR review feedback for #1389 - Run formatSubprocessFailure unconditionally so timeout / ENOENT / EACCES branches also get sanitized log fields (the timeout message also embeds the `Command failed: bash -c <body>` line). - Drop `errType: err.constructor.name` (always 'Error' in production) and replace with `nodeType: 'bash' | 'script'` for actual discriminating value. - Replace chained `||` + ternary in diagnostic selection with explicit if/else for readability. - Simplify exit suffix guard: `err.code != null` instead of double typeof. - Make stderrTail emptiness check explicit. - Drop `RawSubprocessError` export (no external consumer) and widen `code` to `number | string | null` to mirror Node's ExecFileException. - Tighten test length bound to <2100 (was <4000 / <2200) so a doubling of SUBPROCESS_ERROR_MAX_CHARS would actually trip the assertion. - Replace broad regex assertion with `.toContain('[eval]')` — the location marker is the strongest signal that diagnostic content survived. - Strip issue-number citations from describe block and test comments. - Update script-nodes.md to distinguish stderr behavior on success vs failure paths. - Add CHANGELOG Unreleased / Fixed entry for the user-visible change.
…o prevent infinite failure loop (#1294) * fix(orchestrator): clear stale session ID on error_during_execution to prevent infinite failure loop When a Claude API session expires (e.g. after container restart), the orchestrator persists the new (failed) session ID from the error result, causing every subsequent message in that conversation to hit the same error — an infinite failure loop. Fix: on error_during_execution result, set assistant_session_id to NULL instead of persisting the failed session ID. The next message starts a fresh session with full context rebuilt from the DB. Conversation history is unaffected since it lives in remote_agent_messages, independent of the Claude session. Changes: - updateSession() and tryPersistSessionId() now accept string | null - Both handleStreamMode and handleBatchMode clear session ID on error_during_execution Fixes #1280 * test(orchestrator): add stale session clearing tests + address review feedback Co-Authored-By: Claude Opus 4 (1M context) <noreply@anthropic.com> Signed-off-by: kagura-agent <kagura.agent.ai@gmail.com> --------- Signed-off-by: kagura-agent <kagura.agent.ai@gmail.com> Co-authored-by: Claude Opus 4 (1M context) <noreply@anthropic.com>
#1481) * fix(claude): honor CLAUDE_BIN_PATH in dev mode for libc-mismatch hosts The Claude Agent SDK auto-resolves its bundled native binary in [linux-x64-musl, linux-x64] order. On glibc Linux hosts (Ubuntu/Debian/ Fedora), Bun installs both via optionalDependencies and the musl variant is picked first; its ELF interpreter (/lib/ld-musl-x86_64.so.1) does not exist on glibc, so spawn fails and the SDK reports a misleading "binary not found" — the file is on disk, the loader is not. The documented escape hatch CLAUDE_BIN_PATH was dead code in dev mode: the resolver early-returned undefined when BUNDLED_IS_BINARY=false before ever reading the env var. The only workaround was patching node_modules. Move the env-var block above the BUNDLED_IS_BINARY return. Config-file path stays binary-mode-only — it's per-repo, not per-machine; env is the right knob for libc mismatches. Behavior preserved: - env unset → unchanged (undefined in dev, autodetect/throw in binary) - env set + file exists → resolved (was binary-only; now also dev) - env set + file missing → clear error (was binary-only; now also dev) Closes #1474 * chore(claude): address CodeRabbit review on #1481 - CHANGELOG entry under [Unreleased] / Fixed describing the dev-mode CLAUDE_BIN_PATH escape hatch (previously ignored). Notes that config-file path remains binary-mode-only and that env-loading + target-repo .env isolation are unchanged downstream. - Empty-string test pinning that CLAUDE_BIN_PATH='' falls through to undefined rather than throwing — protects against a future predicate typo that would treat empty as "set". - One-line note in ai-assistants.md "Binary path configuration" section pointing dev-mode users at the env-var override for the glibc/musl mismatch case. Skipped from the review: - The other two docs-page rewrites (configuration.md / troubleshooting.md): the error message itself names CLAUDE_BIN_PATH, and #1474 documents the use case publicly. One mention in ai-assistants.md is enough for discovery. - Type-style consistency tweaks in the test file: pure bikeshed.
#1478) The DAG-structure validator scans `node.when`, `node.prompt`, and `loop.prompt` strings for `$nodeId.output` references. Prompt bodies in builder-style workflows embed fenced and inline code as documentation for the LLM (e.g. `archon-workflow-builder` shows how to author a script node), and those literal `$<other-node>.output` mentions were being treated as real cross-node references. Result: `archon-workflow-builder` (a bundled default) failed to load, and `bun run cli workflow run archon-workflow-builder ...` reported "references unknown node '$other-node.output'". Strip triple-backtick fenced blocks and single-backtick inline code from prompt and loop.prompt before scanning. `when:` clauses are JS-like expressions and never carry markdown code, so they pass through unchanged. Real cross-node refs in prose continue to validate. Also wraps one bare `$nodeId.output` mention in `archon-workflow-builder.yaml` Rules section in inline backticks so it reads as documentation alongside the surrounding `$nodeId.output` mentions that already use this style. Closes #1413
- Add --base $BASE_BRANCH to gh pr create in archon-architect, archon-refactor-safely, and archon-implement-issue - Add verify-pr-base bash node to all 9 PR-creating workflows that auto-corrects via gh pr edit if the AI mis-targets - Rewire downstream depends_on edges through verify-pr-base - Regenerate bundled-defaults.generated.ts
#1483) * chore(deps): remove stale package-lock.json to clear Dependabot noise This file was deleted in #85 (Bun migration) but accidentally re-committed in #89 unrelated to that PR's actual fix. It hasn't been touched since April and isn't used by anything (CI runs `bun install`), but Dependabot keeps scanning it — every one of the 21 open alerts triaged in #1353 is against this file, not bun.lock. Removing it closes all 21 alerts. The axios `^1.15.0` override in package.json stays — it's doing real work for the bun tree because @slack/bolt pulls in a vulnerable axios transitively (CVE-2025-62718). Add package-lock.json (and yarn/pnpm lockfiles) to .gitignore so this can't silently slip back in. Closes #1353 * chore(deps): patch four runtime CVEs in bun.lock via overrides Targets #1353 alerts that resolve in the actual runtime tree (bun.lock), not just the stale package-lock.json removed in the previous commit. Added overrides: - follow-redirects ^1.16.0 — auth-header leak on cross-domain redirect (GHSA-r4q5-vmmm-2653); via @slack/bolt - path-to-regexp ^8.4.2 — DoS via sequential optional groups (CVE-2026-4926, CVE-2026-4923); via @slack/bolt + claude-agent-sdk - qs ^6.15.1 — arrayLimit bypass DoS (CVE-2025-15284, CVE-2026-2391); via @slack/bolt - flatted ^3.4.2 — prototype pollution in parse() (CVE-2026-33228); dev-only via eslint chain bun audit confirms each resolves to a single non-vulnerable version across the tree. bun run validate green. No code changes — purely transitive bumps; we don't import any of these directly. Skipped (require deeper triage): undici, lodash, picomatch — each has multiple major versions resolved in the bun tree, so a single override would force-downgrade other consumers.
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Regression from #1481 (honor CLAUDE_BIN_PATH in dev mode). The CI workflow set `CLAUDE_BIN_PATH: ~/.local/bin/claude` in YAML `env:` blocks; YAML does not expand `~`, so the literal string was passed to the resolver. Before #1481, dev mode silently ignored the env var and the SDK auto-resolved its bundled binary — so the broken value was harmless. After #1481, dev mode honors it, the file-existence check fails on the literal `~`, and the smoke job aborts with "CLAUDE_BIN_PATH is set ... but the file does not exist". Move the env-var assignment into the run-step shell where `$HOME` resolves. Both e2e-claude and e2e-mixed-providers jobs are affected.
Wirasm
added a commit
that referenced
this pull request
Apr 29, 2026
After v0.3.10's PR #1488 squash-merged into main, dev was reset to main's HEAD via git reset --hard. That rewrote dev's history, severing every open PR's merge-base from its original commit and ballooning their diffs (e.g. #1444 went from +80/-1 to +6626/-300). This merge commit re-attaches the original release commits (db1c005, f51600a) to dev's history while keeping main's content via a 3-way merge. Open PRs' merge-bases revert to their original commits, their diffs shrink back to normal. No force-push: this is a fast-forward over current origin/dev (fd6d75e is one of this commit's parents).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
10 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
[0.3.10] - 2026-04-29
Maintainer workflow suite, loop output variables, and broad workflow engine fixes
Added
maintainer-standupfor daily PR/issue triage (feat(workflows): add maintainer-standup workflow for daily PR/issue triage #1428), contributor-reply surfacing (feat(maintainer-standup): surface contributor replies since last run #1457),maintainer-review-prfor automated code review (feat(workflows): add maintainer-review-pr and group maintainer workflows under maintainer/ #1430), cross-workflow review memory (feat(maintainer-workflows): cross-workflow review memory + backfill #1458), and a Pi/Minimax variant of standup (feat(maintainer): Pi/Minimax variant of maintainer-standup + dual-format persist #1480).$LOOP_PREV_OUTPUTsubstitution variable in loop node prompts, giving each iteration access to the cleaned output of the previous pass (feat(workflows): expose $LOOP_PREV_OUTPUT in loop node prompts (#1286) #1367).mutates_checkoutflag on workflow nodes to permit concurrent runs against a live checkout without requiring worktree isolation (feat(workflows): add mutates_checkout to allow concurrent runs on live checkout #1438).tagsfield in workflow YAML for categorization and filtering (feat(workflows): support explicit tags and category in workflow YAML #1190).ModelRegistrysupport for custom model slugs and automatic auth bypass for unmapped providers (feat(pi): Use ModelRegistry & allow empty auth to support custom models/providers. #1284).Changed
opus[1m]alias (chore(workflows): switch default Opus pin to opus[1m] alias #1395).Fixed
$BASE_BRANCHinstead of a hardcoded branch name (fix(workflows): ensure all PR-creating workflows target $BASE_BRANCH #1479).$nodeId.outputvalues no longer trigger false DAG validation errors (fix(workflows): skip markdown code blocks in $nodeId.output validation #1478).CLAUDE_BIN_PATHenvironment variable now honoured in dev mode on hosts with libc mismatches (fix(claude): honor CLAUDE_BIN_PATH in dev mode for libc-mismatch hosts #1481).error_during_executionto prevent infinite failure loops (fix(orchestrator): clear stale session ID on error_during_execution to prevent infinite failure loop #1294).safe.directoryfor all repos, not only the primary one (fix(docker): register safe.directory for all repos on bind-mount restart #1307).--versionand--helpno longer crash when bundled skill source files are absent (fix(cli): lazy-import bundled skill so non-setup commands work without source files #1394).--no-env-fileflag no longer incorrectly passed to the native Claude binary in dev mode (fix(claude): stop passing --no-env-file to native binary in dev mode #1461).$nodeId.outputreferences now substituted correctly inside approval gate messages (fix(workflow): substitute $nodeId.output refs in approval messages #1426).ARTIFACTS_DIR,LOG_DIR, andBASE_BRANCHnow exported into bash node subprocess environments (fix(workflows): export ARTIFACTS_DIR, LOG_DIR, BASE_BRANCH to bash nodes #1387).Merging this PR releases 0.3.10 to main.