Bootstrap or upgrade a software repository for agent-first engineering. Use when a user wants to improve project-wide development discipline around `AGENTS.m...
Documents
Harness Engineering
Try itEvolve an existing repository toward Harness Engineering by making it more legible to agents, moving critical knowledge into repo-local artifacts, adding pre...
What it does
Evolve an existing repository toward Harness Engineering by making it more legible to agents, moving critical knowledge into repo-local artifacts, adding pre...
The skill document
Harness Engineering
Evolve an existing repository toward Harness Engineering without pretending it must be rebuilt as a blank template. Prefer repo-local knowledge, predictable structure, and executable checks over aspirational documentation and hidden context.
Workflow
1. Audit the current repo
- Read
AGENTS.mdand the top-level docs underdocs/. - Identify missing source-of-truth artifacts, dead references, placeholder CI, and stack-specific assumptions.
- Confirm whether the repo is a real app, a governance layer, or a starter template.
- Preserve existing product and runtime structure where possible. Do not force a greenfield layout onto a mature codebase.
2. Establish the minimum repo-local artifacts
- Keep
AGENTS.mdshort and map-like. - Add only the missing artifacts that materially improve agent legibility, such as:
docs/ARCHITECTURE.mddocs/core-beliefs.mddocs/quality-score.mddocs/observability.mddocs/worktrees.mddocs/skills.mddocs/product-specs/index.mddocs/product-specs/template.mddocs/exec-plans/index.mddocs/exec-plans/template.mddocs/exec-plans/tech-debt-tracker.mddocs/generated/README.md
- Match the artifact set to the repository's maturity. A small service may need fewer documents than a large multi-domain product.
- If generated artifacts are referenced but not yet produced, add explicit placeholders rather than leaving silent gaps.
3. Replace policy-only claims with enforcement
- Use the repo's existing stack where needed, but prefer
bunfor lightweight governance scripts when that does not conflict with the project. - Add executable checks for required files, stale placeholders, broken references, and obvious contract drift.
- Wire CI to run the checks for every push and pull request.
- Do not leave comments like "run lint later" as the only quality gate.
4. Preserve generality
- Preserve real project names, domains, and runtime details when the repository is already in active use.
- Remove only accidental one-off assumptions, stale tribal knowledge, and undocumented conventions.
- Write docs so a future agent can continue work from the repository itself.
- Prefer incremental hardening over broad rewrites.
5. Make the repo legible to agents
- Keep one clear location for each category of knowledge.
- Put product intent in
docs/product-specs/. - Put multi-step delivery work in
docs/exec-plans/. - Put detailed external library notes in
docs/references/. - Prefer explicit directory conventions over clever local variation.
6. Verify the template
- Run
bun run template:checkor the project equivalent. - Fix all hard failures before stopping.
- Treat warnings as backlog only if they are genuinely optional.
- If the repository is not a git checkout, note that you cannot report git status.
Editing rules
- Use imperative language in docs.
- Prefer concise checklists over essays.
- Keep examples aligned with
bun. - Avoid claiming that a metric, score, or automation is "automatic" unless code actually updates it.
- Add placeholder files when the template references an artifact that new projects will generate later.
Reference
Read references/checklist.md when you need a compact audit list while editing or reviewing a template repo.
Deliverables
- Updated repo-local docs
- Missing placeholder artifacts filled in
- Executable validation or governance checks
- CI wired to run real checks
- A short summary of what now supports Harness Engineering and what still depends on hidden context
Related skills
Local GitHub repository helper for search, clone, sync, and issue/PR inspection workflows. Use when users mention github/repo/repository, ask to download or...
engineering execution guardrails for coding, implementing features, fixing bugs, debugging failures, reviewing diffs, refactoring code, simplifying overbuilt...
Turn mobile product signals into execution-ready backlog drafts across analytics, billing, crashes, and stores.
Use when evaluating, testing, and optimizing an agent architecture or multi-agent system. Best for reviewing planning, routing, memory, tool use, reliability...
Register your AI agent on a model-agnostic network to discover peers, route messages, and collaborate on tasks.