Claude Skills

August 24, 2026 · View on GitHub

Reusable agent skills, references, scripts, and assets for Claude Code, Codex, and related CLI-based development workflows.

This repo is the canonical source for the maintained skill set. Install copies into the active skill root for each provider, or symlink them while iterating.

Last sync: 2026-08-24 — added battle-test, visual-route-sweep, vibe-to-prod, frontendgame; refreshed handoff (v2), validation, and the full audit bundle (21 skills).

Catalog

Ship & verify — prove it works before calling it done

Skill	What it does
`battle-test`	Fail-closed end-to-end hardening cycle: static audit → exhaustive real-browser validation → semantic screenshot review → multi-role/security/visual-regression probes → isolated remediation → re-verification. `--quick` exists but can never emit a battle-tested verdict.
`validation`	Evidence-based browser validation: route/element/flow/API inventory, screenshots, console+network checks, resumable state, redaction, auditable HTML report. Modes: smoke / standard / exhaustive.
`visual-route-sweep`	Authenticated light/dark × desktop/tablet/mobile screenshots of every safe route, responsive+runtime failure detection, semantic review of every image — the visual slice of battle-test, standalone.
`vibe-to-prod`	Guided 13-layer production-readiness self-check for an AI-generated MVP: per-layer PASS/GAP scorecard (security, errors, logging, deploy, compliance…). Lighter than the audit bundle.
`deploy-verify`	Post-deploy verification workflow.
`security-review`	Focused security pass over pending changes.

Audit bundle — technical due diligence

Location: skills/audit/ (21 skills). Full-repository due diligence: the orchestrator, method setup, hard-stop checks, Tambon LLM-signature hunt, blind-spot walk (B1-B19), 13 domain audits, decision handling, fix-prompt generation, and the audit→remediate→rerun loop. Domains follow enumeration-before-verdict.

Primary entry points:

audit-orchestrator: full dual-layer audit (founder view + technical evidence).
audit-loop: audit → roadmap → remediation → rerun → missed-finding patch loop.
audit-fix-generator: remediation prompts for individual findings.

Install the children of skills/audit/ into the provider skill root:

cp -r skills/audit/* ~/.claude/skills/
cp -r skills/audit/* ~/.codex/skills/

Build — opinionated implementation skills

Skill	What it does
`frontendgame`	Premium mobile-first product UI: design-system discipline, advanced UX patterns, motion engineering, screenshot-to-code, anti-generic-AI-look guardrails. References for interaction patterns, motion, stack recipes.
`mcp-builder`	Build MCP servers correctly.
`tdd-workflow`	Red-first test-driven flow.
`debug-resolve`	Systematic debugging to resolution.
`supabase-debug`	Supabase-specific diagnosis.
`feature-validation`	Validate a feature against its spec.
`mercadopago-integration`	MercadoPago OAuth/checkout/webhooks for Supabase/React SaaS, with production pitfalls from 7+ real Argentine projects. Assets: edge function, hook, migration SQL.
`checklist-design-app-audit`	Audits against a validated Checklist Design snapshot: 703 auditable items, evidence-aware scoring, deterministic search scripts.

Session & project workflow

Skill	What it does
`handoff`	Structured session handoff (v2: pre-degradation contract — write BEFORE compaction, greppable header, corr id). The next session starts from the file, not from vibes.
`skillify`	Turn a proven repeated workflow into a reusable skill.
`skill-creator`	Create and validate skills locally (progressive disclosure).
`project-setup`	Generate/refresh AGENTS.md / CLAUDE.md from the real codebase.
`project-curate`	Hand-tune a project's `.claude/` directory.
`project-doctor`	Read-only orchestration health check.

Collaboration

Skill	What it does
`debate`	Three-way Claude / Gemini / Codex debate with rounds, synthesis, session persistence and a web viewer.

Repository Layout

skills/
  audit/                      21-skill due diligence bundle (see above)
  battle-test/                Fail-closed e2e hardening cycle
  checklist-design-app-audit/ Checklist Design evidence audits
  debate/                     Multi-provider debate workflow
  debug-resolve/              Systematic debugging
  deploy-verify/              Post-deploy verification
  feature-validation/         Feature-vs-spec validation
  frontendgame/               Premium mobile-first UI engineering
  handoff/                    Structured session handoffs (v2)
  mcp-builder/                MCP server construction
  mercadopago-integration/    MercadoPago OAuth/payment integration
  project-curate/             .claude/ curation workflow
  project-doctor/             Read-only orchestration health check
  project-setup/              AGENTS.md / CLAUDE.md setup workflow
  security-review/            Security pass over pending changes
  skill-creator/              Skill creation and validation tools
  skillify/                   Workflow → reusable skill
  supabase-debug/             Supabase diagnosis
  tdd-workflow/               Red-first TDD flow
  validation/                 Evidence-based browser validation
  vibe-to-prod/               13-layer production-readiness self-check
  visual-route-sweep/         Visual route sweep with semantic review
tools/                        Supporting docs and legacy helpers
tests/                        Tests for repo tooling

Reference docs at skills/ root: backend-patterns.md, frontend-patterns.md, coding-standards.md, clickhouse-io.md, project-guidelines-example.md, AUTO-HEALING-INTEGRATION-GUIDE.md.

Quick starts

# full hardening cycle on the current project
/battle-test

# browser validation
/validate http://localhost:3000
/validate --backend playwright-local --mode standard http://localhost:5173
/validate --resume

# production-readiness self-check
/vibe-to-prod

# full due diligence audit
/audit

# three-way debate
/debate Should we use Redis or in-memory cache for sessions?

Skill Development

Use skills/skill-creator/ before creating or changing skills. The local guidance follows progressive disclosure:

Keep SKILL.md concise.
Move long reference material into references/.
Put deterministic repeated operations in scripts/.
Put reusable output files in assets/.

Validate changed skills with:

python skills/skill-creator/scripts/quick_validate.py skills/<skill-name>

Package a distributable skill with:

python skills/skill-creator/scripts/package_skill.py skills/<skill-name>

Installation

Install only the skills you actually want active. Copy or symlink from this repo into each provider's skill root (~/.claude/skills/, ~/.codex/skills/).

# ship & verify set
cp -r skills/battle-test skills/validation skills/visual-route-sweep skills/vibe-to-prod ~/.claude/skills/

# audit bundle (installs as 21 sibling skills)
cp -r skills/audit/* ~/.claude/skills/

# workflow set
cp -r skills/handoff skills/skillify skills/project-setup skills/project-curate skills/project-doctor ~/.claude/skills/

# build set
cp -r skills/frontendgame skills/mercadopago-integration ~/.claude/skills/

Symlink while iterating:

ln -s "$(pwd)/skills/validation" ~/.claude/skills/validation

Verification

Run focused validation after edits:

python skills/skill-creator/scripts/quick_validate.py skills/validation
git diff --check

Compare repo skills against installed provider copies:

python skills/skill-creator/scripts/compare_installed_skills.py --target all
python skills/skill-creator/scripts/compare_installed_skills.py --target claude --skill validation

To overlay repo copies into a provider skill root, use --sync. Existing installed copies are backed up under _backup_sync/ first.

Notes

Do not commit secrets, .env files, session transcripts, or private provider state. Personal paths are genericized (~, <projects-root>) before every sync.
Prefer small skill patches over rewrites.
Keep provider-specific behavioral notes in the relevant skill rather than in global instructions.