The scripts in this directory are the deterministic shell/Node/Python layer around Hive's agent runtime. They enumerate work, classify it, gate merges, enforce GitHub permissions, launch/supervise CLIs, and publish operational telemetry before an LLM is asked to act.
Most production scripts are installed under /usr/local/bin by bin/hive-deploy.sh; several write machine-readable state under /var/run/hive-metrics for the dashboard, governor, and agent prompts. For the architecture context, see src/docs/architecture.md.
| Script | Stage | Purpose |
|---|---|---|
run-pipeline.sh |
Orchestrator | Runs configured pre-kick stages in dependency order: enumerator, classifier, gate, and monitor. Supports --agent and --stage selectors. |
enumerate-actionable.sh |
Enumerator | Builds /var/run/hive-metrics/actionable.json with actionable issues and PRs, excluding holds, blocked external-dependency work, drafts, selected labels, ADOPTERS-only PRs, and external issues missing a commit SHA. |
issue-classifier.sh |
Classifier | Enriches actionable.json with deterministic metadata such as complexity tier, model recommendation, tracker status, cluster key, lane, and architecture-review flag. |
architecture-detector.sh |
Classifier | Adds architecture signals to actionable issues from hive-project.yaml rules so the classifier can route them to the architect lane. |
pr-cluster-detector.sh |
Classifier | Groups related actionable issues into clusters using component, reporter timing, label-combo, and failure-mode signals. |
hive-baseline-check.sh |
Classifier | Compares one exact failing check with the repository's default branch and open sibling PRs, returning shared/isolated/unknown plus an action (DEFER_TO_INCIDENT, MERGE_BASE, FIX_DIFF, NOT_REACHABLE_FORK, RERUN_BASELINE) so agents do not retry a repository-wide incident per PR — and do not "repair" branches that are merely behind the default branch, trust a days-old green on a time-gated check, or push to a fork head (#7397). Pass the PR number as a third argument for the per-PR verdicts. |
merge-gate.sh |
Gate | Writes /var/run/hive-metrics/merge-eligible.json; PRs qualify only when required CI passes, they are not drafts or excluded, and author/review policy allows merge. |
conflict-sweeper.sh |
Gate/enforcer | Processes AI-authored PRs with mergeable=CONFLICTING, attempts a rebase, force-pushes clean rebases, or closes unrebasable PRs and reopens the original issue. |
copilot-comment-checker.sh |
Monitor | Prefetches unaddressed Copilot review comments into /var/run/hive-metrics/copilot-comments.json for reviewer agents. |
ga4-anomaly-detector.sh |
Monitor | Compares recent GA4 event counts with a 7-day baseline and writes /var/run/hive-metrics/ga4-anomalies.json; emits a no-data result if GA4 is unavailable. Flags both error-count spikes and traffic collapses (the drop arm added for #6430) by delegating the computation to ga4_anomaly_lib.py. |
ga4_anomaly_lib.py |
Monitor | Pure, network-free anomaly computation for ga4-anomaly-detector.sh: the spike arm (recent error count exceeds a multiple of the 7-day baseline) and the drop arm (#6430 item 4 — an event/page collapsing below a floor of its baseline). Stdlib-only, split out so test_ga4_anomaly_detector.sh can unit-test it without a live GA4 property, a service-account key, or the google-analytics-data library. |
outreach-tracker.sh |
Monitor | Tracks outreach PRs opened or merged on external repositories and writes /var/run/hive-metrics/outreach-prs.json. |
contributor-dispatcher.sh |
Monitor | Reports current contributor assignment state from /var/run/hive-metrics/contributors.json into /var/run/hive-metrics/contributor-assignments.json; assignment itself happens in the dashboard WebSocket server. |
| Script | Stage | Purpose |
|---|---|---|
agent-launch.sh |
Launcher | Unified AI CLI launcher for backends such as Claude and Copilot. Reads --backend/--model flags or AGENT_BACKEND/AGENT_MODEL environment variables. |
supervisor.sh |
Supervisor | Keeps a tmux session alive with an AI CLI agent and supports optional rate-limit failover between backends. |
supervisor-kick.sh |
Supervisor | Reliably restarts or creates a Copilot-backed session, sends one kick message, and verifies the agent began processing. |
kick-agents.sh |
Kick sender | Sends work orders directly to named tmux sessions (scanner, ci-maintainer, architect, outreach, or all), handles backend rate-limit switching, and calls run-pipeline.sh. |
kick-governor.sh |
Scheduler | Legacy adaptive governor that measures issue/PR backlog, chooses surge/busy/quiet/idle cadences, records kick audit entries, and calls kick-agents.sh. |
agent-healthcheck.sh |
Health | Watches an agent log's mtime, kills stalled tmux sessions for supervisor respawn, and escalates after repeated failed respawns. |
kick-outcome-tracker.sh |
Metrics | Correlates governor kicks with queue, MTTR, and token outcomes after a cooldown period. |
gh-rate-check.sh |
Health | Scans tmux panes for GitHub API rate-limit messages, writes /var/run/hive-metrics/gh_rate_limits.json, and temporarily pauses affected CLI cohorts. |
gh-zombie-reaper.sh |
Health | Kills long-running gh processes older than the configured threshold to stop rate-limit retry storms. |
ttyd-tmux.sh |
Terminal UI | Wraps ttyd-to-tmux sessions and disables tmux mouse mode while the browser session is attached so normal text selection works. |
| Script | Stage | Purpose |
|---|---|---|
gh-app-token.sh |
Credentials | Generates and caches (0600, hub-only) a GitHub App installation token; --export prints shell exports for callers; --scoped <tier> [repos] prints a JSON tier-scoped token for a contributor agent and never touches the shared cache. |
git-credential-hive.sh |
Credentials | Git credential helper that serves cached GitHub App tokens and honors the host requested by Git. |
repo-toolchain.sh |
Contributor runtime | Installs a task repository's declared .hive/tools pip requirements into the container's venv before the relay types the prompt (#7925). Bounded: PEP 508 name/extras/version only, apt lines recorded not installed (no root), devcontainer postCreateCommand ignored, time-boxed, exit 0 always. |
gh-wrapper.sh |
Enforcement | gh wrapper that injects App tokens and enforces global/per-agent restriction rules from /etc/hive/restrictions/<agent-id>.json. |
hive-open-pr.sh |
Enforcement | Agent-side wrapper for PR creation requests. It writes a request file for the Hive watcher so PRs are opened by the GitHub App bot and pass the same ACMM authorization checks. See src/docs/hive-open-pr.md. |
hive-open-issue.sh |
Enforcement | Agent-side wrapper for issue creation and comments. Agents call it INSTEAD of gh issue create / gh issue comment; it writes a request file for the Hive watcher so the work is attributed to the GitHub App bot and passes the same ACMM authorization checks. --repo is validated before anything is written (real owner/name, not a placeholder such as owner/repo, inside $HIVE_REPOS); --dry-run previews a request without writing it, which is the only safe way to test the tool since a live call creates a real GitHub object. See src/docs/hive-open-issue.md. |
hive-merge.sh |
Enforcement | Agent-side wrapper for merging a PR. Agents call it INSTEAD of the GitHub MCP merge_pull_request tool, so the merge is performed by the App bot under the same ACMM gating rather than with an agent's own credential. See src/docs/hive-merge.md. |
hive-review.sh |
Enforcement | Agent-side wrapper for gh pr review. Writes a review-request file the Hive submits with the App token and audits as agent_pr_reviewed, so PR-review activity is attributed and visible to hive-health (a direct gh pr review is invisible). Prints the matching .result.json path so callers can require confirmed delivery before dependent writes. --comment --thread <id> replies inside one review-bot thread and --resolve-thread <id> resolves it; the watcher refuses both on any thread a human opened (see src/docs/review-bot-threads.md). |
agent-env-scrub.sh |
Enforcement | Sourced (never executed) at the start of every shell in an agent's process tree, via BASH_ENV/ENV from agent-launch.sh and an /etc/bash.bashrc guard, to unset the live GitHub credentials backend CLIs re-export into agent tool shells (#4045). |
hive-config.sh |
Config | Shared shell config reader that exposes project, repo, agent, dashboard, health, and policy values parsed from hive-project.yaml. |
| Script | Stage | Purpose |
|---|---|---|
hive.sh |
CLI | Operator command wrapper for starting the supervisor, checking status, attaching, kicking agents, reading logs, and stopping agents. |
hive-setup.sh |
Bootstrap | All-in-one Ubuntu 24.04 LXC setup for a Docker-based Hive v2 instance, including generated hive.yaml and env files. Docker-only: it installs Docker Engine and drives docker compose, and predates HIVE_DEPLOY_RUNTIME, so that variable does not redirect it. For a Podman install follow the Quick Start (Podman); see Deployment helper scripts. |
hive-prereq-check.sh |
Bootstrap | Validates host prerequisites for a Docker-based Hive v2 install and can attempt fixes with --fix. |
hive-deploy.sh |
Deploy | Pulls the latest Hive repository and syncs scripts to /usr/local/bin; also restarts Discord bot components when their files change. |
hive-checkout-guard.sh |
Deploy | ExecStartPre guard for systemd units that execute code out of a checkout (e.g. hive-discord.service, hive-snapshot.service): refuses to let the unit start when the checkout directory or its entrypoint files could have been planted by an unprivileged local user (#5435). Its exit status is the assertion — every unsafe path exits non-zero, so a violation actually stops ExecStart. |
hive-standalone-runtime.sh |
Deploy | Selects the standalone container engine from HIVE_DEPLOY_RUNTIME (docker by default) and runs commands against it without ever falling back to the other engine. |
hive-podman-cleanup.sh |
Deploy | Defines the Hive ownership labels for Podman containers, pods, networks, volumes, and images, and guards cleanup so it can never widen into a store-wide Podman/Buildah operation. See src/docs/podman-ownership-cleanup.md. |
hive-podman-teardown.sh |
Deploy | Tears down one standalone Podman deployment through the #4210 ownership contract: every resource is selected by the ownership label, every command is passed through the cleanup guard before it runs, and a command the guard rejects aborts the teardown. Plans by default; run --yes removes. Unlabelled containers, pods, volumes, and networks are invisible to it. See src/docs/podman-ownership-cleanup.md. |
hive-podman-preflight.sh |
Bootstrap | Read-only Podman diagnostics before a lifecycle runs: engine/version, the connection it is actually talking to, rootless vs rootful, and cgroup version. Runs only when HIVE_DEPLOY_RUNTIME selects podman; hive-prereq-check.sh invokes it. |
hive-podman-preflight-host.sh |
Deploy | Read-only Podman preflight for SELinux state and mount labeling, configuration/secrets readability, and published host-port availability. Runs only when HIVE_DEPLOY_RUNTIME=podman. See src/docs/podman-preflight-host.md. |
hive-podman-preflight-ids.sh |
Deploy | Read-only Podman preflight for rootless subordinate UID/GID delegation, unsupported (NFS and other distributed) container storage, and the rootless network backend/helper. Never edits /etc/subuid or /etc/subgid. Runs only when HIVE_DEPLOY_RUNTIME=podman. See src/docs/podman-preflight-ids.md. |
hive-podman-setup.sh |
Bootstrap | One-command standalone Podman install (#4470), the Podman counterpart to hive-setup.sh's Docker path. |
hive-podman-update.sh |
Deploy | Deliberate manual update and rollback for the Hive Quadlet unit (#4378). Updates are explicit rather than automatic, per ADR-0017. reconcile re-copies the repo-owned files a Quadlet host runs from — the gateway config and the unit files — which no update path refreshed before (#6078). |
hive-dashboard-upgrade-helper.sh |
Deploy | Closed host-side executor for dashboard-triggered Hive upgrades. Accepts exactly one operation (upgrade), an allow-listed runtime (podman-quadlet or docker-compose), and a ghcr.io/hivecommons/hive tag or sha256 digest, then execs the existing reviewed lifecycle script (hive-podman-update.sh pin or blue-green-deploy.sh) with the ref as argv — it never evaluates shell text from Hive. See src/docs/dashboard-standalone-upgrades.md. |
hive-podman-lifecycle-probe.sh |
Deploy | Exercises the Quadlet lifecycle — stop, start, restart, recreate, and boot wiring (#4377) — to verify the unit behaves correctly across each transition. |
federation-heartbeat.sh |
Federation | Sends live contributor and actionable-work stats to the Hive federation registry. |
notify.sh |
Notifications | Shared Bash notification library for ntfy, Slack incoming webhooks, and Discord webhooks. |
vibe-kanban-mirror.js |
Integrations | Report-only, default-off mirror of one repository's admitted hive queue (/api/contribute/queue + /api/contribute/fleet) onto a local vibe-kanban board via its MCP stdio server (#10641). Never writes back to hive. With VIBE_KANBAN_SHADOW_LOG, records board pick-ups of queued items as ext_work_shadow_observed events in a local JSONL file. See src/docs/vibe-kanban.md. |
| Script | Stage | Purpose |
|---|---|---|
contributor-agent.sh |
Contributor runtime | Contributor-container entrypoint: detects authenticated CLI backend, starts the relay, launches the CLI in tmux, and creates ${HOME}/agent.md only from a verified live knowledge export. |
contributor-local-relay.sh |
Contributor runtime | Sourced by just contribute-hive <cli> local and just contribute-stop: refuses a local start while contributor-relay.pid names a live local relay, stops that relay and its tmux session, and runs a sandboxed true through bubblewrap before Claude-family local mode advertises readiness (#14832, #14833). |
contributor-relay.js |
Contributor runtime | Node.js WebSocket client for ClankeR contributor agents. It authenticates to one or more hubs, receives tasks, injects GitHub tokens, reports progress/results, supports The Commons multi-hive routing strategies (ranked, spread, neediest) between tasks, and supports interactive tmux or headless one-shot delivery. Pure pane-classification (readiness/login/onboarding, the busy/idle/blocked/error state machine) lives in lib/pane-classifier.js, which this file requires. |
contributor-relay.sh |
Contributor runtime | Thin POSIX-sh compat wrapper that execs contributor-relay.js with the same argv (#6429) — kept because the .sh path is baked into existing deployments. New callers should invoke contributor-relay.js directly. |
lib/pane-classifier.js |
Contributor runtime | Pure (no process/tmux) pane-classification library required by contributor-relay.js: classifyReadiness, classifyPane, blockingPromptKey, classifyBlockedOnHumanReason, and the pane-tail/API-error-line detectors they share. Kept test-for-test in sync with src/crux/agent/manager.go's Go port via the shared fixtures in testdata/pane-fixtures/ (#6427). |
hivectl-bootstrap.sh |
Contributor runtime | Resolves hivectl for contributor Justfile recipes (just contribute-tui, just contribute-hives) and execs it with the caller's argv (#8200). Resolution order: HIVECTL override, ./bin/hivectl, user/system install, PATH; when none is found it stages a checkout-local binary from the Hive image (HIVECTL_BOOTSTRAP_IMAGE, default ghcr.io/hivecommons/hive:stable) via a throwaway Podman container carrying the #4210 ownership labels, and records the source digest in bin/.hivectl.source so later runs refresh or refuse instead of using a stale binary. No Go toolchain required. |
contributor-quota-control.js |
Contributor runtime | Out-of-band contributor quota override command (#6953). The relay usually runs detached (background, container, K8s pod), so #6833's scoped overrides cannot depend on its stdin; this writes continue-once / continue-until-reset / disable-session / pause-until-reset / resume overrides into the same per-pool directory (lib/quota-pool-store.js) the relay reads on every guard evaluation. Stores only the relay's own published hold state — never a provider credential, raw account identifier, or billing figure. |
lib/quota-pool-store.js |
Contributor runtime | Shared per-pool quota-override store used by both contributor-relay.js (guard evaluation) and contributor-quota-control.js (override writes), so the attached banner and the out-of-band command share exactly one mechanism with no second code path to drift (#6953). |
pi-backend.js |
Contributor runtime | Pi contributor adapter contract (#5039). AGENT_MODEL is the one contributor-owned selection input; the adapter derives the provider's official credential variable names from it so only the selected provider's keys are handed to the container. |
mlx-backend.js |
Optional local inference | Apple Silicon runtime/config checks, explicit tool-call verification and offline loopback server launch for Pi's mlx provider. Use just contribute-mlx-serve, just contribute-mlx-setup <model> --verify-tools, then just contribute-mlx; see MLX setup. No automatic installation or model downloads. |
mlx-pi-extension.js |
Optional local inference | Delegates streaming/tools to Pi's maintained OpenAI adapter with a loopback-only, redirect-rejecting transport; required for the private hive-mlx API. |
omp-backend.js |
Contributor runtime | omp (Oh My Pi) contributor adapter (#7678). Container mode stages the host's omp sign-in — ~/.omp/agent/config.yml plus the agent.db SQLite store whose auth_credentials rows hold every provider's OAuth record or API key; omp has no auth.json — into the ephemeral CLI staging dir and, like pi-backend.js --stage, keeps only the selected provider's rows (AGENT_MODEL=provider/model, else config.yml's modelRoles). --describe is what just contribute-setup omp prints so a missing host sign-in is caught before a container opens on omp's setup wizard. --stage and --describe refuse/flag a row omp has marked disabled_cause (a revoked OAuth refresh), and --sync-back writes a credential the container refreshed back over the host row it was staged from — only the selected provider's existing row, only when strictly newer — so a single-use refresh token consumed in the container does not strand the host; just contribute-hive omp runs it every HIVE_OMP_CREDENTIAL_SYNC_SECONDS (default 300) while the container runs and once more at exit (#7922). |
| Script | Stage | Purpose |
|---|---|---|
copilot-models.mjs |
Model discovery | Uses @github/copilot-sdk and the installed Copilot CLI's own auth to print the available Copilot model catalog as one JSON line. |
token-collector.sh |
Cost metrics | Reads scanner beads and token metrics to attribute issue cost; supports recent windows and --all. |
token-usage.py |
Cost metrics | Aggregates Claude and Copilot CLI session token usage by agent and rolling time window. |
nous-install.sh |
Nous | Clones the external Nous strategy-evolution framework under NOUS_DIR, fetches and verifies the reviewed NOUS_PIN_SHA commit (override for testing with HIVE_NOUS_PIN), creates a venv, installs it editable, and prepares run directories. Bump the pin only by reviewing the new upstream commit and updating the script in a PR. |
nous-runner.sh |
Nous | Strategist helper that invokes the installed Nous framework on each kick. |
nous-hive-gate.py |
Nous | Implements the Nous Gate protocol, posting suggest-mode decisions to the dashboard and polling for operator approval. |
nous-sync.py |
Nous | Reads Nous governor/repo outputs, applies confidence decay and conflict detection, and writes principles, recommendations, and ledger data. |
| Script | Covers |
|---|---|
contributor-agent.test.sh |
Contributor-agent regression for knowledge export handling. |
contributor-relay.test.js |
Contributor relay task/restart/headless behavior; loads contributor-relay.js with stubs, plus direct-require coverage of lib/pane-classifier.js. |
vibe-kanban-mirror.test.js |
vibe-kanban-mirror.js against a fake MCP stdio server (testdata/vibe-kanban-mcp/): status mapping, default-off, --dry-run, idempotent --once reruns, and shadow pick-up observation. |
gh-wrapper.test.sh |
gh-wrapper.sh author-gate and restriction regressions using a mock gh binary. |
repo-toolchain.test.sh |
repo-toolchain.sh manifest parsing: what never reaches pip (URLs, paths, options, unknown directives), apt lines recorded not run, pip failure never fails the task. |
test_agent_env_scrub.sh |
agent-env-scrub.sh (#4045): backend CLIs must not re-export live GitHub credentials into the tool shells they spawn. Behavioural plus source assertions. |
test_gh_auth_native_no_cat.sh |
The N14 (#3842) fix: a native/systemd-install agent kicked via kick-agents.sh — no Go AgentManager, no per-agent HIVE_AGENT_TOKEN_CACHE — still authenticates without leaking the token through cat. |
test_gh_wrapper_gates.sh |
gh-wrapper.sh's enforcement gates. |
test_git_credential_hive.sh |
git-credential-hive.sh behaviour. |
test_hive_open_issue.sh |
hive-open-issue.sh, the agent issue-creation chokepoint. See src/docs/hive-open-issue.md. |
test_hive_open_pr.sh |
hive-open-pr.sh, the agent PR-creation chokepoint: CLI parsing, required-field validation, body sourcing (--body, --body-file, stdin) and the empty-body refusal, --issues normalization, JSON request structure, and the python3-less awk fallback escaper. Pins two regressions that already shipped (dropped --body-file; invalid JSON for multi-line bodies without python3). See src/docs/hive-open-pr.md. |
test_hive_review.sh |
hive-review.sh, the agent PR-review chokepoint. |
test_hive_podman_setup.sh |
Contract tests for hive-podman-setup.sh (#4470). |
test_hive_podman_update.sh |
Contract tests for hive-podman-update.sh (#4378). |
test_hive_podman_lifecycle_probe.sh |
Contract tests for hive-podman-lifecycle-probe.sh (#4377). |
test_hive_baseline_check.sh |
Shared-CI classifier regressions for red default branches, exact-name sibling thresholds, reruns, pending checks, fail-closed API errors, and the #7397 verdicts: red siblings all behind a fresh green base are drift (MERGE_BASE), all behind a stale green base are unknown (RERUN_BASELINE), an up-to-date red sibling is shared regardless of base age, and a fork head is NOT_REACHABLE_FORK. |
test_bin_suites_wired.sh |
Fails when a test suite in this directory is not run by any workflow, Justfile target, or hook (#4363). |
test_hive_standalone_runtime.sh |
hive-standalone-runtime.sh engine selection: Docker default, explicit Podman, and no silent fallback. |
test_hive_podman_cleanup.sh |
hive-podman-cleanup.sh ownership labels and cleanup guard. Analyses arguments only: it contacts no container engine and deletes nothing. |
test_hive_podman_teardown.sh |
hive-podman-teardown.sh against a fake Podman whose store holds Hive's resources next to a Distrobox, an unrelated development container, and somebody's Postgres volume. Asserts the unowned resources survive every teardown, replays every command the teardown issued through the #4210 guard, and checks that a hostile resource name aborts the run instead of widening it. |
test_hive_podman_preflight.sh |
hive-podman-preflight.sh engine, connection, root-mode, and cgroup checks. Drives a stub engine, so the whole matrix runs on a host with no Podman installed. |
test_hive_podman_preflight_host.sh |
hive-podman-preflight-host.sh across enforcing, permissive, and disabled SELinux, wrong and missing mount labels, unreadable and over-permissive secrets, and occupied and sub-floor ports. Every input is mocked; it runs no container and changes nothing on the host. |
test_hive_podman_preflight_ids.sh |
hive-podman-preflight-ids.sh across delegated, missing, short, and multi-range subordinate IDs, NFS and other unsupported graphroots, tmpfs, a missing rootless network helper, and rootful hosts. Every input is mocked; PATH is the fake bin alone, so no real helper on the test machine can mask an absent one. |
test_enumerate_actionable.sh |
Contract tests for enumerate-actionable.sh, which writes /var/run/hive-metrics/actionable.json — the one file merge-gate.sh, kick-agents.sh, and the dashboard's contribute lane all consume. |
test_merge_gate.sh |
Contract tests for merge-gate.sh: how actionable.json is turned into /var/run/hive-metrics/merge-eligible.json. |
test_issue_classifier.sh |
Contract tests for issue-classifier.sh, the pipeline stage that enriches actionable.json with complexity_tier, model_recommendation, lane, cluster, and architecture-review fields that kick-agents.sh reads verbatim into every scanner kick. Executes a path-rewritten copy of the script against a temp dir (and asserts the rewrite landed) rather than grepping it. |
test_hivectl_bootstrap.sh |
Contract tests for hivectl-bootstrap.sh (#8200): resolution order, image staging via a fake podman, source-digest freshness checks, and the Justfile recipes that call it. Hermetic — PATH is a fake bin, no image is pulled. |
test_kick_governor.sh |
Contract tests for kick-governor.sh, the adaptive kick timer that ladders scanner/ci-maintainer/architect/outreach cadence from the actionable backlog. |
test_gh_rate_check.sh |
Contract tests for gh-rate-check.sh: GitHub API rate-limit detection in agent tmux panes and the pause/pullback decisions it makes. |
test_backend_smoke.sh |
End-to-end smoke for the contributor-agent backends: exercises the seam with vendor coding-agent CLIs (completion detection, flags, output shape) that vendors can break on their own release schedule (#5376). |
test_backend_smoke_attribution.sh |
backend-smoke's failure-issue script must attribute failures truthfully (#5883): an infrastructure failure (e.g. missing podman on the fleet image) must not be reported as a CLI-version regression. |
backend_smoke_attribution_harness.js |
Harness used by test_backend_smoke_attribution.sh: executes backend-smoke.yml's failure-issue script against fixtures so its attribution is verified by running it rather than by reading it (#5883). |
test_backend_smoke_image_owner.sh |
backend-smoke's pinned lane must test THIS repository's contributor image, not upstream's (#6086): extracts the image-selection step from backend-smoke.yml and runs it, asserting an uppercase owner is lowercased and the upstream fallback is loud rather than silent. |
test_ga4_anomaly_detector.sh |
Contract test for ga4_anomaly_lib.py's compute_anomalies() (#6430 item 4): hermetic Python assertions against synthetic recent/baseline dicts covering both the spike arm and the traffic-collapse drop arm — no GA4 property, key, or network. |
test_dibs_cutover_preflight.sh |
Contract tests for src/deploy/dibs-domain-cutover/preflight.sh (#5925): every case executes the preflight against stub kubectl/dig/curl and asserts on exit code and report — no cluster, no network. |
test_dibs_cutover_verify.sh |
Contract tests for src/deploy/dibs-domain-cutover/verify.sh (#5925): every case executes the verifier against stub openssl/curl/dig/kubectl and asserts on exit code and report. Pins that each silent-pass failure mode — a self-signed certificate, a certificate missing the new SAN, an inert redirect, a redirect that drops the path and query, a hub on a foreign registrable domain — blocks rather than reading as a finished cutover. No cluster, no network. |