damonzs/claude-codex-pro-tool/harness-engineering

Claude Code and agent skill by damonzs/claude-codex-pro-tool

Trust tier 3
  • community
Source check pending

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-engineering

SKILL.md

Harness Engineering

Core Rule

Treat every non-trivial task as a controlled engineering loop:

  1. Read project context.
  2. Define or locate the spec.
  3. Define or locate acceptance criteria.
  4. Implement the smallest necessary change.
  5. Verify with real evidence.
  6. 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:

  1. Inspect the target project root.
  2. Generate the Harness Engineering baseline documents:
  • AGENTS.md
  • docs/harness-engineering-theory.md
  • spec/harness-engineering-baseline.md
  • acceptance/harness-engineering-baseline.md
  1. Do not overwrite an existing AGENTS.md unless the user explicitly asks for overwrite.
  2. Prefer running scripts/init_harness.py <project-root> for deterministic initialization.
  3. 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 typeExamplesRequired workflow
Microrun a command, explain code, typo fixUse existing context, verify if claiming a result
Normallocal bug fix, small UI copy changeSummarize scope and run targeted verification
Importantpage, API, module, cross-file behaviorCreate/read spec/*.md and acceptance/*.md first
High riskauth, secrets, deletion, installer, release, migrationFull spec, acceptance, rollback/risk notes, stronger verification

When uncertain, classify upward.

Required Workflow

1. Gather Context

Read in this order when present:

  • AGENTS.md
  • README.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:

  1. Task conclusion.
  2. Files changed and why.
  3. Verification commands/results.
  4. Acceptance criteria mapping.
  5. 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

MistakeCorrection
Coding before the spec is clearStop and write/update the spec
Acceptance says "works correctly"Replace with observable pass/fail criteria
Reporting "done" after reading codeRun verification or state why it cannot run
Expanding scope during implementationMove extra ideas to follow-up
Treating failed verification as a footnoteFix 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

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
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/damonzs/claude-codex-pro-tool/harness-engineering/badge)](https://verified-skill.com/skills/damonzs/claude-codex-pro-tool/harness-engineering)

Something wrong with this skill or its page? Tell us.