Contribution

Contribution

Teaches an AI agent how to contribute correctly to this repository (agents-skills), following its own conventions.

Purpose

Ensure contributions are consistent, well-structured, and merge-ready by grounding the agent in the repository's actual rules rather than generic GitHub advice. Reduces review churn and keeps the skill catalog coherent.

When to Use

• User says they want to contribute, add a skill, fix a skill, or open a PR against this repo
• User asks "how do I contribute?", "how do I add a new skill?", "what are the conventions?"
• Agent is about to scaffold a new skills/<name>/SKILL.md in this repository
• Before the agent commits, pushes, or opens a PR in this repo

Workflow

1. Orient the contributor and set up the workspace

• Confirm the repo root is agents-skills (contains skills/, CONTRIBUTING.md, README.md)

• If cloning fresh:

  git clone https://github.com/marwan562/agents-skills.git
  cd agents-skills
  

• For fork workflow, verify remotes:

  git remote -v
  # expect origin -> contributor fork, upstream -> marwan562/agents-skills
  

• Ensure working from latest main:

  git checkout main
  git fetch upstream  # or origin if no upstream
  git pull --ff-only
  

2. Read the source of truth

• Open and follow CONTRIBUTING.md (the canonical workflow) and docs/architecture.md (why the layout is this way)
• Check root README.md → Available Skills table for naming collisions
• Inspect skills/<existing>/SKILL.md count and style to match tone
• Note commit convention (Conventional Commits) and branch naming (feat/<skill>, fix/<skill>-...)

3. Plan the change

• For new skill: propose kebab-case name, verify it matches directory, draft description (20-300 chars, trigger style: "Use when...")
• For fix/docs: identify the minimal diff; note if README.md catalog or docs/architecture.md also needs an update
• For large or cross-cutting change: advise opening an issue first (Proposal: <skill-name> skill) before coding — per CONTRIBUTING.md

4. Scaffold or edit

• New skill:

  mkdir -p skills/my-new-skill
  # Create skills/my-new-skill/SKILL.md with required frontmatter
  

• Frontmatter must be at line 1:

  ---
  name: my-new-skill
  description: Use when you need to do X efficiently
  ---
  

• Body sections in order: Purpose → When to Use → Workflow → Instructions → Constraints → Examples (plus optional References)

• Keep the skill independent and reusable; no hard dependency on another skill without declaration

5. Update catalog and cross-references

• If adding a skill: add a row to README.md → Available Skills table

  | **my-new-skill** | `skills/my-new-skill/` | Short purpose summary |
  

• Add installation examples for the new skill:

  npx skills add marwan562/agents-skills --skill my-new-skill
  

• Adjust structure diagrams if needed; do not restructure unrelated sections

6. Validate locally (never skip)

Run these before committing:

# 1. Each skill has SKILL.md and name matches directory
for f in skills/*/SKILL.md; do
  dir=$(basename $(dirname "$f"))
  name=$(grep "^name:" "$f" | awk '{print $2}')
  [ "$dir" = "$name" ] && echo "✓ $f" || echo "✗ MISMATCH $f dir=$dir name=$name"
done

# 2. No secrets / tokens in diff
git diff --cached  # review; ensure no .env, keys, tokens

# 3. Frontmatter parses
python3 -c "
import pathlib, yaml
for p in pathlib.Path('skills').glob('*/SKILL.md'):
    data = yaml.safe_load(p.read_text().split('---')[1])
    assert 'name' in data and 'description' in data
    print(f'✓ {p}')
"

# 4. Links resolve
# Click through README links to CONTRIBUTING.md and docs/architecture.md

7. Commit, push, and open PR

• Create a topic branch from main:

  git checkout -b feat/my-new-skill
  

• Commit with Conventional Commits:

  git add skills/my-new-skill/SKILL.md README.md
  git commit -m "feat(my-new-skill): add my-new-skill skill"
  

• Push and open PR:

  git push -u origin feat/my-new-skill
  gh pr create --title "feat(my-new-skill): add my-new-skill skill" --body "$(cat <<'EOF'
  ## Summary
  Adds `my-new-skill`...
  
  ## Changes
  - ...
  
  ## Validation
  - [x] Frontmatter valid
  - [x] No secrets in diff
  - [x] Links checked
  EOF
  )"
  

• If gh is not authenticated, instruct the user to run gh auth login and retry; do not fabricate a PR URL

8. Optional Jev pre-PR check (skipped without JEV_API_KEY)

When configured, send the SKILL.md diff + intent + validation results to jev_review; fix concrete flags, re-validate, rescore with previousEvaluation. CONTRIBUTING.md and the validator still decide. Never send secrets. Full protocol: contribute Jev gates.

Instructions

• Follow CONTRIBUTING.md first. If this skill and CONTRIBUTING.md disagree, CONTRIBUTING.md wins — update this skill to match.
• Keep diffs minimal. One skill per PR; do not bundle unrelated reformatting.
• Preserve existing style. Mirror headings, tone, and table formatting from existing skills.
• Validate file paths. Use forward slashes, skills/<name>/SKILL.md exactly; do not create skill/ or SKILL.MD.
• Warn before external actions. Confirm with the user before git push or gh pr create — these are public and irreversible.

Constraints

• Do NOT overwrite origin or upstream remotes without asking
• Do NOT commit secrets, tokens, .env, or private URLs
• Do NOT change LICENSE or .gitignore unless explicitly requested
• Do NOT create a skill without name ≡ directory-name invariant
• Do NOT invent a CONTRIBUTING.md workflow that contradicts the actual file in the repo

Examples

Example 1 — Contributor wants to add a skill

User: "I want to add a testing skill to this repo."

Agent:

1. Reads CONTRIBUTING.md:1-50, checks skills/ for collision (testing not present)
2. Scaffolds skills/testing/SKILL.md with required sections
3. Updates README.md Available Skills table
4. Runs validation loop; shows git diff --stat and asks confirmation before push

Example 2 — Quick fix to an existing skill

User: "Fix a typo in the code-review skill."

Agent:

1. Reads skills/code-review/SKILL.md:45, patches typo
2. Validates frontmatter still parses, name unchanged
3. Commits as fix(code-review): correct typo in workflow step 3
4. Pushes only after confirmation

Example 3 — Contributor asks how to contribute

User: "How do I contribute here?"

Agent:

1. Summarizes CONTRIBUTING.md: clone → create skills/<name>/SKILL.md → validate → commit (feat:) → PR
2. Shows scaffold command and validation checklist
3. Points to docs/architecture.md for deeper design rationale

References

• Canonical workflow: CONTRIBUTING.md
• Architecture rationale: docs/architecture.md
• Skill spec and install guide: README.md