Lhwcv Tinykaggleclaw Tinykaggleclaw

Bulk Optimize

Improve external SKILL.md files for PR contributions using tessl skill review plus safety checks that prevent rejected PRs. Respects maintainers by detecting auto-generated files, preserving domain expertise, and suggesting appropriate PR scope.

For External PR Contributions

Adds safety checks for contributing to other people's repositories. See TEMPLATES.md for lessons from past PR rejections.

Workflow

Phase 1: Safety Checks (CRITICAL)

Run before making changes to prevent rejected PRs:

# 1. Auto-generated file detection
grep -i "auto-generated\|generated by\|DO NOT EDIT\|generator:" <skill-path>/SKILL.md

# 2. Recent migration detection
git log --oneline -10 --all -- <skill-directory>/ | grep -i "migrate\|rename\|move"

# 3. Repository pattern analysis
find . -name "SKILL.md" -type f | head -5 | xargs head -20

If auto-generated detected: STOP - changes must go to generator source If recent migration found: WARN - may cause merge conflicts Note domain patterns: Preserve expert framing and terminology in Phase 4

Phase 2: Baseline Evaluation

Run review for each skill:

tessl skill review <skill-path>/SKILL.md

Parse output to extract:

• Overall score
• Validation issues
• Description/content dimension scores
• Judge suggestions

If improving multiple skills:

• Track baseline scores for all
• Identify which need improvement most
• Consider PR scope (see Phase 7)

Phase 3: Discover Skill Bundle

Read SKILL.md and list files in its directory. Bundle = SKILL.md + sibling files + referenced files. Use for progressive disclosure improvements.

Use bundle context for recommendations.

Phase 4: Generate Recommendations (Preserve Domain Expertise)

Review recommendations to preserve valuable domain knowledge. Grep for specialized terminology (CRO, funnel, A/B test, etc.).

KEEP: Expert framing with domain concepts ("CRO expert who understands funnel optimization, conversion psychology...") REMOVE: Generic expert statements ("You are an expert at helping users...")

For each issue: what to change, why (dimension + score), before/after, impact, domain check.

Phase 5: Validate Recommendations

# Validate Python syntax
import ast
ast.parse(code_snippet)  # Raises SyntaxError if invalid

# Validate command flags
gh pr create --help | grep -q "\-\-flag" || echo "Invalid"

# Validate file references
[[ -f "$SKILL_DIR/$FILE" ]] || echo "Missing file"

Phase 6: Present Recommendations

Show recommendations with:

• Current state (score)
• Issue explanation
• Recommended change
• Impact
• Domain preservation note

Get approval per skill or in batch (user choice).

Phase 7: PR Scope Assessment

Multiple skills (15+): Recommend one-skill-per-PR for easier review. Start with 2-3 highest-impact, get feedback, iterate.

High-scoring skills (90%+): Open issue first before unsolicited PR. Risk: maintainer may not want cosmetic changes.

Phase 8: Apply Changes

Apply approved changes to approved skills only.

Phase 9: Verify Improvements

# For each changed skill
tessl skill review <skill-path>/SKILL.md

Show before/after scores.

Phase 10: Final Accuracy Review

Re-run validation from Phase 5 on updated SKILL.md. Check:

• ✓ Code syntax valid
• ✓ Command flags correct
• ✓ File references exist
• ✓ Third-person voice
• ✓ "Use when..." clause present

Fix issues, then re-run tessl skill review.

Phase 11: PR Guidance

Use commit message format, PR description template, and best practices from TEMPLATES.md. Display ready-to-create PR info for user confirmation.

Notes

• Uses tessl skill review plus PR-specific safety checks
• For external PR work on repos you don't maintain
• For personal skill improvement, use skill-optimizer instead

Source: lhwcv/tinyKaggleClaw — distributed by TomeVault.