Merge pull request #5 from 8thlight/eval/improve-rpi-skills

Evaluate and improve RPI pipeline skills

Author: ericjohnolson

.github/workflows/bump-versions.yml

@@ -25,7 +25,7 @@ jobs:
         run: |
           git config user.name "github-actions[bot]"
           git config user.email "github-actions[bot]@users.noreply.github.com"
-          git add .claude-plugin/marketplace.json
+          git add .claude-plugin/marketplace.json plugins/*/.claude-plugin/plugin.json
           git diff --staged --quiet && echo "No version changes" && exit 0
           git commit -m "Bump marketplace and plugin versions"
           git push

AGENTS.md

@@ -4,7 +4,7 @@ This file provides guidance to Claude Code when working with this repository.
 
 This project is a Claude Code plugin marketplace containing **skills** — structured markdown workflow documents. There is no application code or build system. Skills are validated by a three-layer test pipeline (see Testing Skills below).
 
-For project architecture, key concepts (RPI methodology, testing philosophy, hexagonal architecture), artifact lifecycle, and installation commands, see `docs/architecture.md`.
+The core workflow follows the **RPI methodology** (Research → Plan → Implement) for non-trivial features. Skills are validated by a three-layer test pipeline (see Testing Skills below).
 
 ## Skill Authoring Guidelines
 
@@ -32,16 +32,18 @@ allowed-tools: Read Glob Write
 - `triggers` — List of phrases that activate the skill. Use specific multi-word phrases; avoid bare single-word triggers that risk false activation.
 - `allowed-tools` — Space-separated list of Claude Code tools the skill may use.
 
-## Core Workflow (Plan Mode Native)
+## Core Workflow (RPI)
 
-The primary workflow aligns with Claude Code's native plan mode:
+The primary workflow flows seamlessly through three phases, with artifacts at `.light/sessions/` as handoff points:
 
-1. **Research** (outside plan mode) — Assess complexity, dispatch parallel subagents if warranted, write temporary artifact to `.light/sessions/`
-2. **Plan** (inside plan mode) — Draft behavior activates, produces plan summarizing research with Agent Context blocks
-3. **Execute** (`/implement`) — Creates task graph from approved plan, runs three-agent TDD orchestration, writes session artifact to `.light/sessions/`
+1. **Research** (`/research`) — Assess complexity, dispatch parallel subagents if warranted, write research artifact to `.light/sessions/`
+2. **Plan** (`/plan-tasks`) — Consumes research artifact, produces plan with Agent Context blocks + task graph
+3. **Execute** (`/implement`) — Executes task graph with three-agent TDD orchestration, writes session artifact to `.light/sessions/`
 4. **Post-execution** — code-review, simplify, reflect recommendations
 5. **Reflect** (`/reflect`) — Optional post-session learning loop
 
+Each phase flows directly into the next. Context clearing is only suggested when the conversation is extensive — the artifacts carry all needed context forward.
+
 ## Creating New Skills
 
 Use the Anthropic `/skill-creator` skill to guide you through creating new skills. It provides an interactive workflow for designing effective skill documents.
@@ -51,8 +53,7 @@ Before creating a skill, review the Skill Authoring Guidelines and Frontmatter F
 ### Where Skills Live
 
 ```
-plugins/praxis/skills/   — research, plan-tasks, implement, tdd, refactor, reflect
-plugins/harness/skills/  — adr, diagram, distill, hexagonal-architecture, scaffold
+plugins/praxis/skills/   — research, plan-tasks, implement, tdd, adr, reflect
 ```
 
 Each skill directory contains:

README.md

@@ -9,10 +9,9 @@ A Claude Code plugin providing skills for agentic engineering patterns and pract
 ```
 /plugin marketplace add 8thlight/lightfactory
 /plugin install lightfactory@praxis
-/plugin install lightfactory@harness
 ```
 
-After installation, skills are available as `praxis:skill-name`, `harness:skill-name` and activate automatically when relevant to your task.
+After installation, skills are available as `praxis:skill-name` and activate automatically when relevant to your task.
 
 ### Install from Local Clone
 
@@ -23,48 +22,47 @@ git clone https://github.com/8thlight/lightfactory
 ```
 /plugin marketplace add /path/to/lightfactory
 /plugin install lightfactory@praxis
-/plugin install lightfactory@harness
 ```
 
 ### Update
 
 ```
 /plugin marketplace update lightfactory@praxis
-/plugin marketplace update lightfactory@harness
 ```
 
 ## Available Skills
 
-### Praxis Plugin: used for agentic engineering flow
+### Praxis Plugin
 
 | Skill | Command | Description |
 |-------|---------|-------------|
-| **tdd** | `/tdd` | Boundary-focused TDD workflow enforcing L3/L4 altitude testing and property-based tests |
 | **research** | `/research` | Spawns parallel subagents to explore a codebase and produce a compact research artifact |
 | **plan-tasks** | `/plan-tasks` | Consumes research artifact and produces a compact implementation plan with L3/L4 test specs |
 | **implement** | `/implement` | Executes an implementation plan phase by phase with strict test-first discipline |
-| **refactor** | `/refactor` | Refactoring process with test safety and incremental commits |
+| **tdd** | `/tdd` | Boundary-focused TDD workflow enforcing L3/L4 altitude testing and property-based tests |
+| **adr** | `/adr` | Guides writing minimal Architecture Decision Records |
 | **reflect** | `/reflect` | Post-session reflection that mines git history and artifacts to produce improvement proposals |
 
-#### RPI Methodology (Research → Plan → Implement)
+### RPI Methodology (Research → Plan → Implement)
 
-The light factory praxis plugin's core workflow for non-trivial features:
+The praxis plugin's core workflow for non-trivial features follows three phases:
 
-1. **Research** (`/research`) — Explore the codebase with parallel subagents, output a compact research artifact
-2. **Plan** (`/plan-tasks`) — Consume the research artifact, produce a compact implementation plan with test specs
+1. **Research** (`/research`) — Explore the codebase with parallel subagents, output a compact research artifact to `.light/sessions/`
+2. **Plan** (`/plan-tasks`) — Consume the research artifact, produce a compact implementation plan with test specs and Agent Context blocks
 3. **Implement** (`/implement`) — Execute the plan phase by phase with strict RED → GREEN → VALIDATE discipline
 
+#### Key Concepts
 
-### Harness Plugin: used for harness engineering flow
+- **Plan Mode Native** — The workflow aligns with Claude Code's native plan mode. Research happens outside plan mode; planning activates draft behavior inside plan mode; implementation executes the approved plan.
+- **Three-Agent TDD Isolation** — During implementation, each TDD phase dispatches isolated agents: `agent-test` (writes failing tests), `agent-impl` (writes minimal code to pass), and `agent-validate` (runs full test suite). Agents never modify each other's artifacts.
+- **Context Compaction** — Research produces a ~200-line artifact that carries forward into planning, replacing unbounded codebase exploration with a focused summary. Plan mode entry is the compaction boundary.
+- **Tracker Detection Chain** — The implement and plan-tasks skills auto-detect available task trackers: yaks (preferred) → beads (fallback) → native tasks (last resort).
 
+#### Standalone Skills
 
-| Skill | Command | Description |
-|-------|---------|-------------|
-| **diagram** | `/diagram` | Creates architecture diagrams (C4 structural, dynamic flows, data flow) with subtype dispatch |
-| **scaffold** | `/scaffold` | Scaffolds DDD projects from Gherkin feature files with language subtype dispatch |
-| **hexagonal-architecture** | `/hexagonal-architecture` | Applies hexagonal (ports & adapters) architecture with domain-first design |
-| **adr** | `/adr` | Guides writing minimal Architecture Decision Records |
-| **distill** | `/distill` | Distills an agentic engineering harness from a project with DDD structures and fitness tests |
+- **TDD** (`/tdd`) — For interactive, human-in-the-loop test-driven development outside the full RPI flow. Enforces boundary-focused testing at the L3/L4 altitude with ZOMBIES progression.
+- **ADR** (`/adr`) — Guides writing Architecture Decision Records following the Harmel-Law signal check pattern. Ensures decisions are actually made before documenting them.
+- **Reflect** (`/reflect`) — Post-session learning loop that mines git history and session artifacts to produce improvement proposals for skills, CLAUDE.md, and hooks.
 
 ## Testing
 
@@ -87,5 +85,3 @@ cd tests/evals && promptfoo eval
 ## Contributing
 
 See `AGENTS.md` for skill authoring guidelines, testing workflows, and the pre-ship checklist.
-
-See `docs/architecture.md` for project structure and key concepts.