damonzs/claude-codex-pro-tool/harness-engineering
Claude Code and agent skill by damonzs/claude-codex-pro-tool
- community
Checking that the SKILL.md file is still in the repository.
Use when a coding task needs spec-first planning, acceptance criteria, verification evidence, AI agent workflow design, sub-agent coordination, disciplined project delivery, or project workflow initialization.
Install
npx vskill@latest install damonzs/claude-codex-pro-tool/harness-engineeringSKILL.md
Harness Engineering
Core Rule
Treat every non-trivial task as a controlled engineering loop:
- Read project context.
- Define or locate the spec.
- Define or locate acceptance criteria.
- Implement the smallest necessary change.
- Verify with real evidence.
- Report against the acceptance criteria.
Do not claim completion without verification evidence.
Initialization Mode
When the user invokes this skill in a project and asks to initialize, bootstrap, or set up Harness Engineering:
- Inspect the target project root.
- Generate the Harness Engineering baseline documents:
AGENTS.mddocs/harness-engineering-theory.mdspec/harness-engineering-baseline.mdacceptance/harness-engineering-baseline.md
- Do not overwrite an existing
AGENTS.mdunless the user explicitly asks for overwrite. - Prefer running
scripts/init_harness.py <project-root>for deterministic initialization. - After initialization, verify the four files exist and report the generated paths.
If the script is unavailable, use references/initialization.md to create the same baseline manually.
Decision Guide
Use the lightest process that protects the task:
| Task type | Examples | Required workflow | |
|---|---|---|---|
| Micro | run a command, explain code, typo fix | Use existing context, verify if claiming a result | |
| Normal | local bug fix, small UI copy change | Summarize scope and run targeted verification | |
| Important | page, API, module, cross-file behavior | Create/read spec/*.md and acceptance/*.md first | |
| High risk | auth, secrets, deletion, installer, release, migration | Full spec, acceptance, rollback/risk notes, stronger verification |
When uncertain, classify upward.
Required Workflow
1. Gather Context
Read in this order when present:
AGENTS.mdREADME.md- Relevant
docs/ - Relevant
spec/*.md - Matching
acceptance/*.md - Related source, config, and tests
If project docs conflict, follow the higher-priority project rule and state the conflict.
2. Fill Missing Harness
If a non-trivial task lacks a spec or acceptance criteria:
- Create or update a spec before implementation.
- Create or update matching acceptance criteria before delivery.
- Keep both scoped to the current task.
Use references/templates.md for templates.
3. Summarize Before Implementing
Before code changes, summarize:
- Current goal.
- Files likely to change.
- Files or behaviors that must not change.
- Acceptance criteria.
- Key risks.
4. Implement Narrowly
Make the smallest change that satisfies the spec. Avoid unrelated refactors, dependency changes, production config changes, and cleanup outside scope.
5. Verify
Run the narrowest real verification that proves the acceptance criteria. Broaden verification for shared, security-sensitive, release, migration, or user-visible behavior.
Never invent test, build, run, screenshot, or log results.
6. Report
Final response must include:
- Task conclusion.
- Files changed and why.
- Verification commands/results.
- Acceptance criteria mapping.
- Remaining risks or follow-up.
Sub-Agent Use
For complex work, split roles:
- Spec Agent: produces
spec/*.md. - Acceptance Agent: produces
acceptance/*.md. - Implementation Agent: changes code/config/tests.
- Test Agent: verifies against acceptance criteria.
- Review Agent: checks drift, omissions, quality, and risk.
Do not let the same role silently redefine the target, implement it, and judge completion.
References
Read only what is needed:
references/theory.md: deeper theory, task layering, failure modes, and metrics.references/initialization.md: project initialization behavior and generated baseline expectations.references/templates.md: spec, acceptance, task summary, and final report templates.references/checklists.md: quick checklists for creation, implementation, verification, and review.
Common Mistakes
| Mistake | Correction | |
|---|---|---|
| Coding before the spec is clear | Stop and write/update the spec | |
| Acceptance says "works correctly" | Replace with observable pass/fail criteria | |
| Reporting "done" after reading code | Run verification or state why it cannot run | |
| Expanding scope during implementation | Move extra ideas to follow-up | |
| Treating failed verification as a footnote | Fix it or report it as unresolved risk |
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/harness-engineering/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/damonzs/claude-codex-pro-tool/harness-engineering)