docs(harness): add plugin README and update root README

README.md

@@ -9,6 +9,7 @@ Productivity and security plugins for [Claude Code](https://claude.ai/code).
 | [opaq](plugins/opaq/) | Secure credential access — use secrets in commands without ever exposing them |
 | [git-pilot](plugins/git-pilot/) | Automated git workflow — branch creation, commit formatting, push and MR prompts |
 | [craft](plugins/craft/) | Generate multi-phase implementation prompts from technical specs |
+| [harness](plugins/harness/) | Agent orchestration harness — task classification, verification gates, session bridging |
 
 ## Installation
 
@@ -26,6 +27,7 @@ Install plugins:
 /plugin install opaq@moukrea-plugins
 /plugin install git-pilot@moukrea-plugins
 /plugin install craft@moukrea-plugins
+/plugin install harness@moukrea-plugins
 ```
 
 ### From a local clone

plugins/harness/README.md

@@ -0,0 +1,99 @@
+# harness — Agent Orchestration Harness
+
+Transparent harness for long-running Claude Code agents. Automatically classifies tasks, decomposes complex work, tracks progress, verifies output, bridges sessions, and adapts to any project type. Zero configuration required.
+
+## What it does
+
+- Classifies prompt complexity (simple → massive) and recommends an execution strategy
+- Guards against test removal, lock file edits, and incomplete implementations
+- Runs tests and lint before allowing a session to stop or a task to be marked complete
+- Tracks failure patterns and suggests recovery after repeated errors
+- Detects project type at session start (Node.js, Rust, Go, Python, etc.) and injects verification commands
+- Preserves git state across context compaction so you don't lose orientation
+- Provides 9 specialized agent definitions for parallel team workflows
+
+## In practice
+
+```
+You: "Refactor the authentication system to use JWT tokens,
+      update all API endpoints, and add comprehensive tests"
+
+Claude: [HARNESS] Task classified as complex.
+        Consider: decompose into tasks, use subagents
+        for parallel work, verify each piece.
+        -> Decomposes into auth module, endpoint migration, test suite
+        -> Spawns implementer agents in isolated worktrees
+        -> Verifies tests pass before marking each task complete
+        -> Blocks session stop until all tests and lint pass
+```
+
+## Skills
+
+| Skill | Description |
+|-------|-------------|
+| `/harness:init` | Initialize harness for a project — detects languages, creates rules |
+| `/harness:session-bridge` | Resume work from a previous session with live state injection |
+| `/harness:task-analyze` | Analyze task complexity and recommend execution strategy |
+| `/harness:task-decompose` | Break complex work into independent, parallelizable units |
+| `/harness:requirements-interview` | Gather detailed requirements for vague tasks |
+| `/harness:spec-ingest` | Ingest a specification document into granular feature units |
+| `/harness:verify-work` | Run comprehensive verification (tests, lint, build, types) |
+| `/harness:progress-report` | Generate a summary of completed tasks, pending work, blockers |
+| `/harness:recovery` | Recover from stuck state — analyze errors, revert, re-plan |
+| `/harness:reflect` | Suggest configuration improvements after completing complex work |
+| `/harness:deployment-monitor` | Monitor a deployment, CI pipeline, or PR status |
+| `/harness:logs` | Summarize recent harness hook activity for debugging |
+
+## Hooks
+
+| Hook | Trigger | Action |
+|------|---------|--------|
+| Session start | New session | Detects project type, test/lint/build commands, git status |
+| Prompt classify | Each prompt | Classifies complexity, recommends execution strategy |
+| Write guard | Write, Edit | Blocks test removal and warns on lock file edits |
+| Post-edit | Write, Edit | Runs per-file type checking (TypeScript, Python, Go, JS) |
+| Failure tracker | Bash failures | Tracks patterns, warns after 3+ similar failures |
+| Task gate | Task completion | Verifies tests pass before allowing task completion |
+| Stop gate | Session stop | Verifies tests and lint pass before allowing stop |
+| Implementation check | Implementer stop | Blocks incomplete work (TODO, placeholder, stub) |
+| Pre/post compact | Context compaction | Preserves and restores git state across compaction |
+
+## Agent Definitions
+
+The harness includes 9 specialized agent roles for team workflows:
+
+| Agent | Role | Model |
+|-------|------|-------|
+| Architect | System design, task decomposition, planning | Opus |
+| Implementer | Focused single-task coding in isolated worktrees | Default |
+| Tester | Test writing and verification | Default |
+| Reviewer | Code quality, security, and consistency review | Sonnet |
+| Debugger | Root cause analysis and minimal fixes | Default |
+| Integrator | Merge conflict resolution and integration validation | Default |
+| Researcher | Deep codebase exploration and analysis | Sonnet |
+| Monitor | Long-running process and CI pipeline tracking | Haiku |
+| UI Verifier | Visual verification via browser automation | Sonnet |
+
+## Project Detection
+
+The harness auto-detects the project environment at session start:
+
+| Project Type | Test | Lint | Build |
+|-------------|------|------|-------|
+| Node.js (TypeScript) | `npm test` | `npm run lint` | `npm run build` |
+| Rust | `cargo test` | `cargo clippy` | `cargo build` |
+| Go | `go test ./...` | `golangci-lint run` | `go build ./...` |
+| Python | `pytest` | `ruff check .` | — |
+| Make-based | `make test` | — | `make build` |
+
+## Rule Templates
+
+Language-specific rules are available for project initialization (`/harness:init`):
+
+`database` · `frontend` · `go` · `infrastructure` · `python` · `rust` · `testing` · `typescript`
+
+## Installation
+
+```
+/plugin install harness@moukrea-plugins
+```