From harness
Use this skill whenever the user mentions setup, initialization, starting a new project, CI setup, Codex CLI setup, harness-mem, agent configuration, symlinks, mirrors, or runs /harness-setup. Also use when the user needs to configure the harness environment or onboard a new repository. Do NOT load for: code implementation (use harness-work), code review (use harness-review), release (use harness-release), or planning (use harness-plan). Unified setup skill for Harness v3 — project initialization, tool configuration, 2-agent setup, memory config, symlinks, and mirror sync.
npx claudepluginhub tim-hub/powerball-harness --plugin harnessThis skill is limited to using the following tools:
Unified setup skill for Harness v3.
Creates isolated Git worktrees for feature branches with prioritized directory selection, gitignore safety checks, auto project setup for Node/Python/Rust/Go, and baseline verification.
Executes implementation plans in current session by dispatching fresh subagents per independent task, with two-stage reviews: spec compliance then code quality.
Dispatches parallel agents to independently tackle 2+ tasks like separate test failures or subsystems without shared state or dependencies.
Unified setup skill for Harness v3. Consolidates the following legacy skills:
setup — Unified setup hubharness-init — Project initializationharness-update — Harness updatesmaintenance — File cleanup and organization| Subcommand | Behavior |
|---|---|
harness-setup init | New project initialization (CLAUDE.md + Plans.md + hooks) |
harness-setup ci | CI/CD pipeline configuration |
harness-setup codex | Codex CLI installation and configuration |
harness-setup harness-mem | harness-mem integration and memory configuration |
harness-setup mirrors | skills/ → public mirror bundle update |
harness-setup agents | agents/ agent configuration |
harness-setup localize | CLAUDE.md rule localization |
Introduce Harness v3 to a new project.
Generated files:
project/
├── CLAUDE.md # Project configuration
├── Plans.md # Task management (empty template)
├── .claude/
│ ├── settings.json # Claude Code configuration
│ └── hooks.json # Hook configuration (v3 shim)
└── hooks/
├── pre-tool.sh # Thin shim (→ core/src/index.ts)
└── post-tool.sh # Thin shim (→ core/src/index.ts)
Flow:
Configure GitHub Actions workflows.
# .github/workflows/ci.yml generation example
name: CI
on:
push:
branches: [main]
pull_request:
jobs:
test:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- run: npm ci && npm test
# Installation check
which codex || npm install -g @openai/codex
# Timeout command check (macOS)
TIMEOUT=$(command -v timeout || command -v gtimeout || echo "")
# For macOS: brew install coreutils
Usage patterns (via official plugin):
bash scripts/codex-companion.sh task --write "task content"
# Or via stdin
cat /tmp/prompt.md | bash scripts/codex-companion.sh task --write
Configure Unified Harness Memory.
# Create memory directories
mkdir -p .claude/agent-memory/claude-code-harness-worker
mkdir -p .claude/agent-memory/claude-code-harness-reviewer
# Deploy MEMORY.md template
cat > .claude/agent-memory/claude-code-harness-worker/MEMORY.md << 'EOF'
# Worker Agent Memory
## Project Context
[Project overview]
## Patterns
[Learned patterns]
EOF
On Windows with core.symlinks=false, repository symlinks become regular files, and harness-* skills may not appear in the command list. Public bundles are synced as real directory mirrors.
Codex skills are symlinks to skills/ — no manual sync needed:
ls -la codex/.codex/skills/
Update targets:
skills/ (SSOT)codex/.codex/skills/ (symlinks → ../../../skills/)Configure the 3-agent structure in agents/.
agents/
├── worker.md # Implementation (task-worker + codex-implementer + error-recovery)
├── reviewer.md # Review (code-reviewer + plan-critic)
└── scaffolder.md # Scaffolding (project-analyzer + scaffolder)
Adapt .claude/rules/ rules to the current project.
# Check rule list
ls .claude/rules/
# Add project-specific rules
cat >> .claude/rules/project-rules.md << 'EOF'
# Project-Specific Rules
[Project-specific rules here]
EOF
Marketplace stability was significantly improved in v2.1.71.
# Pin version with @ref format (recommended)
claude plugin install owner/repo@v3.5.0
# Latest version
claude plugin install owner/repo
The owner/repo@vX.X.X format is recommended. With the @ref parser fix, tags, branches, and commit hashes are all resolved accurately.
claude plugin update owner/repo
Merge conflicts during updates were fixed in v2.1.71, enabling stable updates.
/plugin uninstall uses settings.local.json: Accurately reflected in user-local settingsPeriodic maintenance tasks:
| Task | Command |
|---|---|
| Delete old logs | find .claude/logs -mtime +30 -delete |
| Compress Plans.md | Move completed tasks to an archive section |
| Delete old traces | tail -1000 .claude/state/agent-trace.jsonl > /tmp/trace && mv /tmp/trace .claude/state/agent-trace.jsonl |
harness-plan — Create project plans after setupharness-work — Execute tasks after setupharness-review — Review setup configuration