agentleFS
ffy6511ffy6511/harness-creator-skill/skills/harness-creator/SKILL.md

harness-creator

Skill4 starsChanged 5 months agoMITView on GitHub

Use when creating or adapting a coding-agent harness for either an existing repository or a new project, especially when you need to preserve working flows such as spec-driven development, verification paths, or session continuity.

5 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions5 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 5

    1. walkinglabs/learn-harness-engineering★ 21ksame name
    2. TheAatroxking1/harness_creator_skill★ 48same name
    3. anthony-chaudhary/fak★ 47same name
    4. idoforgod/cys-harness-creator★ 4same name
    5. tjdwls101010/Harness-Creator★ 4same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

The file

Show its rules and text

We split this skill into the rules it gives an agent

23 rules. 0 are common, 19 are distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 19

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill when:

    line 20
  2. Distinctive:

    Do not use this skill for prompt engineering, model choice, or generic software architecture.

    line 26
  3. Distinctive:

    Treat these bundled templates as the highest-fidelity scaffold sources:

    line 34
  4. Distinctive:

    Use them when generating:

    line 40
  5. Distinctive:

    Do not collapse them into short summaries.

    line 46
  6. Very specific:

    Use docs/decisions/ as the long-term home for durable design decisions instead of a separate specs constitution folder.

    line 70
  7. Very specific:

    If the repo is already spec-driven, extend that path instead of adding parallel feature_list.json, progress.md, or handoff files by default.

    line 77
  8. Distinctive:

    If the repo is empty or very early, gather enough product and workflow context first.

    line 80
  9. Distinctive:

    If the user has not defined architecture, lifecycle, or execution style, ask for that context before generating files.

    line 81
  10. Distinctive:

    Only scaffold after the working model is clear enough.

    line 82
  11. Distinctive:

    If the repo already has a coherent harness, improve it in place.

    line 96
  12. Distinctive:

    Use Repo Adaptation Pattern for this pass.

    line 98
  13. Distinctive:

    Use Specs Directory Pattern and Docs Directory Pattern when the repo already separates planning from knowledge capture.

    line 99
  14. Distinctive:

    If the repo already has a good state source, keep it.

    line 109
  15. Distinctive:

    Use Spec-Driven Harness Pattern when long-running work is phase-based.

    line 111
  16. Distinctive:

    Use readiness gates that match the repo, not a fixed bootstrap ritual.

    line 122
  17. Very specific:

    Treat specs and docs as two different but related systems:

    line 138
  18. Distinctive:

    When the repository uses structured planning, prefer this split:

    line 143
  19. Distinctive:

    Do not bury durable decisions inside stale specs if they need to outlive one implementation plan.

    line 150
  20. Distinctive:

    Use Gotchas before finalizing a design.

    line 161
  21. Distinctive:

    Use these bundled assets selectively.

    line 165
  22. Distinctive:

    Never copy every template by default.

    line 203
  23. Very specific:

    When generating specs/AGENTS.md or docs/AGENTS.md, prefer the detailed templates over a minimal directory guide.

    line 205

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Harness Creator

Build or adapt a repo-local harness without breaking the repo's current working model.

Language rule: Unless the user explicitly asks otherwise, follow this logic to choose the scaffold language:

  1. Existing repo mode — detect the harness language by scanning AGENTS.md, CLAUDE.md, README.md, docs/AGENTS.md, specs/AGENTS.md, or other top-level prose files. If the majority of these files are written in Chinese, recommend Chinese templates and confirm with the user. Otherwise default to English.
  2. Greenfield mode — ask the user which language they prefer before generating any scaffold. Offer English and Chinese as choices.

Templates are organized under templates/en/ (English) and templates/zh/ (Chinese). When scaffolding, pick the subdirectory that matches the chosen language.

When to use

Use this skill when:

  • A user wants AGENTS.md, CLAUDE.md, progress tracking, verification rules, or session continuity.
  • A repository already has some harness structure, and you need to improve it without replacing its source of truth.
  • A new or early-stage project needs an initial harness, but the workflow and architecture still need to be clarified.

Do not use this skill for prompt engineering, model choice, or generic software architecture.

Core rule

Do not force one harness shape onto every repository.

High-fidelity templates

Treat these bundled templates as the highest-fidelity scaffold sources:

  • templates/{en,zh}/architecture.md
  • templates/{en,zh}/specs-agents.md
  • templates/{en,zh}/docs-agents.md

Use them when generating:

  • ARCHITECTURE.md
  • specs/AGENTS.md
  • docs/AGENTS.md

Do not collapse them into short summaries. Match their level of specificity, section granularity, and formatting discipline, then adapt the content to the target repository.

For specs/AGENTS.md, preserve the detailed rules for:

  • directory responsibilities
  • file naming
  • plan template structure
  • phase and task formatting
  • lifecycle transitions
  • writing guidance

For docs/AGENTS.md, preserve the detailed rules for:

  • writing principles
  • numbering rules
  • frontmatter requirements
  • section templates

Then extend the docs model to cover:

  • docs/lessons/
  • docs/features/
  • docs/decisions/

Use docs/decisions/ as the long-term home for durable design decisions instead of a separate specs constitution folder.

Choose one of two modes:

  1. Existing repo mode
    • Inspect the current harness first.
    • Preserve the existing source of truth if it works.
    • If the repo is already spec-driven, extend that path instead of adding parallel feature_list.json, progress.md, or handoff files by default.
  2. Greenfield mode
    • If the repo is empty or very early, gather enough product and workflow context first.
    • If the user has not defined architecture, lifecycle, or execution style, ask for that context before generating files.
    • Only scaffold after the working model is clear enough.

Workflow

1. Detect the repo mode

Look for:

  • AGENTS.md, CLAUDE.md
  • specs/AGENTS.md, specs/active/, specs/draft/, specs/archive/
  • docs/AGENTS.md, docs/lessons/, docs/features/, docs/decisions/
  • feature_list.json, progress.md, session-handoff.md
  • init.sh, bootstrap, check, or other readiness entrypoints

If the repo already has a coherent harness, improve it in place. Do not create duplicate truth sources.

Use Repo Adaptation Pattern for this pass. Use Specs Directory Pattern and Docs Directory Pattern when the repo already separates planning from knowledge capture.

2. Pick the state source on purpose

The agent needs one primary control plane for scope and status. Common options:

  • Spec-driven: active spec + docs decisions + status checkboxes
  • Feature-list-driven: feature_list.json + progress.md
  • Minimal: AGENTS.md + readiness path + short progress log

If the repo already has a good state source, keep it.

Use Spec-Driven Harness Pattern when long-running work is phase-based.

3. Define readiness instead of blindly adding init.sh

The harness must define how the agent proves the repo is ready to work on. That can be:

  • init.sh
  • a smaller verification script
  • a startup checklist inside AGENTS.md
  • a phase-0 section in an active spec

Use readiness gates that match the repo, not a fixed bootstrap ritual.

4. Encode end-of-session continuity

The repo must answer:

  • What is the current focus?
  • What is done?
  • What is blocked?
  • What proof exists?
  • Where does the next session resume?

That continuity can live in progress.md, session-handoff.md, or an actively maintained spec.

Specs and docs are complementary

Treat specs and docs as two different but related systems:

  • specs/ pushes active development forward.
  • docs/ captures stable explanations, lessons, feature behavior, and durable decisions.

When the repository uses structured planning, prefer this split:

  • specs/active/NN-topic-plan.md for execution
  • docs/features/NN-topic.md for user-facing or system-facing capability docs
  • docs/lessons/NN-topic.md for experience summaries and non-obvious implementation lessons
  • docs/decisions/NN-topic.md for durable cross-spec decisions

Do not bury durable decisions inside stale specs if they need to outlive one implementation plan.

5. Keep governance separate from planning

Spec-driven development can replace some progress and state files, but it does not replace:

  • clean-state discipline
  • verification discipline
  • repository hygiene
  • periodic governance reviews

Use Gotchas before finalizing a design.

Templates and scripts

Use these bundled assets selectively. Prose templates live under language-specific subdirectories:

templates/
├── en/                     # English templates
│   ├── agents.md
│   ├── architecture.md
│   ├── specs-agents.md
│   ├── docs-agents.md
│   ├── docs-decision.md
│   ├── docs-feature.md
│   ├── docs-lesson.md
│   ├── active-spec.md
│   ├── progress.md
│   ├── session-handoff.md
│   └── readiness-check.md
├── zh/                     # Chinese templates
│   ├── agents.md
│   ├── architecture.md
│   ├── specs-agents.md
│   ├── docs-agents.md
│   ├── docs-decision.md
│   ├── docs-feature.md
│   ├── docs-lesson.md
│   ├── active-spec.md
│   ├── progress.md
│   ├── session-handoff.md
│   └── readiness-check.md
├── feature-list.json       # language-neutral
├── feature-list.schema.json
└── init.sh                 # language-neutral

Scripts:

  • scripts/detect-harness.sh
  • scripts/scaffold-greenfield.sh

Never copy every template by default. Pick the smallest set that fits the repo mode.

When generating specs/AGENTS.md or docs/AGENTS.md, prefer the detailed templates over a minimal directory guide. Only emit a minimal version if the user explicitly asks for one.

Common mistakes

  • Replacing a spec-driven workflow with feature_list.json just because the course used it.
  • Adding init.sh when the repo only needs a lightweight readiness check.
  • Creating progress.md and session-handoff.md even though the active spec already tracks phase state well.
  • Letting planning files and governance files drift apart.
  • Creating multiple status sources that disagree.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.