From harness-kit
Analyze codebase and recommend AI agent harness (agents, skills, hooks)
How this skill is triggered — by the user, by Claude, or both
Slash command
/harness-kit:harness-builderThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Bootstrap an AI agent harness for any codebase. The harness-builder agent analyzes your stack, workflows, and domain to recommend agents, skills, and automation hooks.
Bootstrap an AI agent harness for any codebase. The harness-builder agent analyzes your stack, workflows, and domain to recommend agents, skills, and automation hooks.
Get a tailored harness recommendation (agents, skills, hooks, CLAUDE.md skeleton) without starting from scratch.
/commit, /review-pr) rather than generating new ones. Extend these with project-specific variants./harness bootstrap # Triggers harness-builder agent automatically
Or manually:
[Invoke harness-builder agent on current codebase]
The agent returns:
Adjust recommendations:
After approval, agent generates:
.claude/agents/*.md (one per agent, with tool permissions and domain context).claude/skills/*/SKILL.md (one per skill, with PURPOSE/CRITICAL/STANDARD PATH)CLAUDE.md (harness documentation).gitignore updatesRun verification command to confirm setup works:
git add . && nix flake check # Or your stack's validation
The harness-builder recommends agents based on project type and detected stack.
Always recommended:
If GitHub detected:
If frontend detected:
.impeccable.md) — Accessibility and design audit guidelines/ui-review skill — WCAG 2.1 AA compliance checkerIf database-heavy:
The harness-builder recommends skills based on detected workflows and available global skills.
Always recommended:
/commit — Conventional commits with project-specific quality gates (extends global /commit)If GitHub is detected:
/feature-spec — Write feature specs before implementation/issue-triage — Categorize and prioritize issues/standup — Generate daily standup from git + GitHub activityIf frontend is detected:
/ui-review — Accessibility and UX audit (WCAG 2.1 AA)Key principle: Use global skills + project-local wrappers.
✅ Good:
/commit handles conventional format/commit adds quality gates specific to your stack✅ Good:
/harness provides meta-harness framework/harness needed (framework is stack-agnostic)❌ Bad:
/commit workflow in project-local skill (creates maintenance burden)Hooks enable zero-token automation: formatters and validators run silently on every file edit, before Claude sees the content.
Run before Claude sees any edits. Examples:
nix run ".#format" — Format Nix filesprettier --write — Format JavaScript/JSONruff check --fix — Auto-fix Python lintingmarkdownlint --fix — Fix Markdown formattingTrigger: Every Edit/Write tool use. Cost: Zero tokens (silent, happens before Claude reads).
Run after formatting. Output ≤30 lines, gated with conditions. Examples:
nix flake check — Nix syntax validationtype-check — TypeScript/mypy type checkingcargo test --lib — Rust unit tests (sampled)npm test -- --coverage — Test coverage checkTrigger: After edit if conditions match (e.g., "if .nix file was changed"). Cost: Tokens only if errors found.
Orient developers with skill/doc references. Examples:
Trigger: After validation, always shown. Cost: <20 lines, always helpful, never noisy.
{
"hooks": {
"PostToolUse": {
"Edit,Write": [
{
"stage": "formatters",
"name": "format-files",
"command": "nix run \".#format\" -- {file_path} 2>/dev/null || true"
},
{
"stage": "validators",
"name": "nix-check",
"command": "nix flake check --quiet 2>&1 | head -10",
"condition": "file_matches('**/*.nix')"
},
{
"stage": "guidance",
"name": "harness-check",
"command": "echo '📋 Run /harness to audit the agent harness'",
"condition": "file_in(['.claude/skills/**', '.claude/agents/**'])"
}
]
}
}
}
The harness-builder detects your project stack by looking for:
| Detection Method | Detected Stack |
|---|---|
package.json + npm scripts | JavaScript/TypeScript/Node.js |
Cargo.toml | Rust |
go.mod | Go |
pyproject.toml or setup.py | Python |
flake.nix + home.nix | Nix/Home Manager |
Makefile with ci target | Generic make-based |
justfile with check recipe | Just-based |
| File | Detection |
|---|---|
package.json | npm test or npm run check |
Cargo.toml | cargo test |
Makefile | make ci or make check |
justfile | just check or just ci |
flake.nix | nix flake check |
.github/workflows/ or gh CLI integration.impeccable.md or frontend framework dependenciesmigrations/, schema.sql, or ORM configsHarness-builder will ask:
Use these answers to customize recommendations before finalizing.
/harness — Audit harness quality (gates on PURPOSE/CRITICAL/STANDARD PATH)CLAUDE.md — Generated harness documentation (customize after generation)/documentation-writer — Write scannable docs/skill-writer — Create well-structured skillsnpx claudepluginhub dougborg/harness-kit --plugin harness-kitGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.