skills-gc

Custom skills for Claude Code. Each skill gives Claude a specific behaviour — triggered automatically from context, or invoked manually as a slash command.

Skills

Skill	What it does
checkpoint	End-of-session handoff — writes a one-line entry to CHECKPOINTS.md plus a matching git commit (subject = same line, body = full narrative). Next session reads the index first, opens full commit bodies only for the matches
verify	Audits Claude's own output against the user's original request. PASS/PARTIAL/FAIL per requirement, flags scope creep
plan-challenger	Stress-tests implementation plans before execution — researches alternatives, challenges complexity, returns a revised plan
perspective	Strategic reality check — checks whether the solution already exists, whether there's a better approach, against live docs not stale training data
cc-audit	Audits a project's Claude Code setup against official Anthropic docs across 9 areas, produces a scored AUDIT-REPORT.md
coherency-check	Cross-document and cross-code coherency analysis — finds contradictions, conflicting values, and overlapping definitions, resolves each interactively
philosophier	Compresses verbose text into the shortest precise phrasing that preserves every constraint. Searches for established philosophical principles (YAGNI, Chesterton's Fence, Occam's Razor) when they're exact matches; falls back to raw distillation otherwise
website-audit	SEO, AEO, GEO, and structured-data audit. Crawls with Playwright, runs Lighthouse, checks Perplexity citations, scores deterministically
interactive-qa	Interactive Q&A that collects your decisions when agents present multiple options or disagree. Plain-language choices, recommendation flagged, dissent between agents surfaced. Produces a brief for Plan Mode, a specialist agent, or just context

Install

ln -s /path/to/skills-gc/skills/verify ~/.claude/skills/verify

Each skill directory has a SKILL.md with the full definition and an evals/ folder with test cases.

License

MIT