aurelienbobenrieth/harness/build

Claude Code and agent skill by aurelienbobenrieth/harness

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Implement an authorized feature, fix, or refactor test-first, with designs that stay deep, isolated, and small at their public surface. Use when changing code behavior, adding modules or exports, or reshaping boundaries. Not for explanation-only requests or unresolved requirements (see align).

Install

npx vskill@latest install aurelienbobenrieth/harness/build

SKILL.md

Build

Smallest change that proves itself, in code that's easy to change next time. A small edit stays a small edit.

Before editing

  • Inspect the working tree; existing work survives. Read the real validation commands and their side effects from config.
  • "Can you fix this?" → act. Ask only for an outcome-changing decision or an action outside authorization; several open, interdependent decisions → align first.
  • Unknown cause or unfamiliar boundary → a small probe before choosing the design.

Loop

  1. Contract: trigger → expected outcome, from the request or spec, before reading the implementation.
  2. Red: one failing test per behavior slice, oracle per testing. Fails on the assertion, not a fixture.
  3. Green: minimal code; tests read-only. Spec and test disagree → stop and report.
  4. Refactor on green; structure and behavior changes in separate commits.
  5. Prove it: doubt a green test → break the code, it must go red. Pick extra strategies (property, state model, contract, packed install) by the failure you fear, not by habit.

Defect → reproduce the symptom at its boundary first, cheapest distinguishing observation first; no reproduction → name the missing condition and keep the cause provisional. Covered, low-impact, reversible edit → no new test, still the full gate.

Design

A module hides one decision, its interface stays much smaller than what it hides, and dependencies point inward. Before shaping a module, an export, or a package boundary, load design: each rule there names the lint that backs it.

Implement and verify

  • Runtime limits, retry identity, cancellation, partial failure: design them in when the workload needs them.
  • Stale generated code → run the generator (check whether it syncs externally). Never hand-write or widen substitute types.
  • Parallel workers: bounded scope, file ownership, a verification task each; trust their evidence, not their confidence. Serialize shared mutable state.
  • In a worktree, servers take their ports from worktree-hygiene ports with a strict port; before handoff worktree-hygiene done exits 0 (--reap stops what you started). Never stop a process by name or port alone.
  • Prove at the failure boundary: real filesystem for a writer, consumer install for exports, browser for focus. A fake, static render, or compile proves only itself.
  • Skipped, empty, stale, or unavailable ≠ pass. Rerun checks a later change invalidated; full gate on the final state.
  • Final diff: no debug output, temp files, formatter churn, or scope creep.

Handoff: resulting behavior, checks run and results, material limits; readable without earlier progress messages. Never "done" with a failing gate.

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

Repository stars
1
7-day trend score
0

Stars and forks are for the whole GitHub repository, not this skill alone. 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/aurelienbobenrieth/harness/build/badge)](https://verified-skill.com/skills/aurelienbobenrieth/harness/build)

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