The Claude Code skills I actually run every day.
Not a demo pack — these were extracted from months of daily driving Claude Code across real projects: shipping PRs, launching sites, recovering crashed sessions, and auditing my own prompting. Each one was born from a friction pattern that showed up repeatedly, got codified, and has been battle-tested since. Anything personal has been generalized; what's left is the mechanism.
Maintained by Dylan Pulver.
/plugin marketplace add dylanpulver/claude-skills
Then install any plugin you want:
/plugin install session-continuity@claude-skills
/plugin install shipping@claude-skills
/plugin install thinking@claude-skills
/plugin install launch@claude-skills
/plugin install meta@claude-skills
Each plugin is independent — install only what you'll use.
| Skill | What it does |
|---|---|
whereami |
Answers "where are we at / whats next" from a durable per-repo state ledger that outlives sessions, verified against git/PR reality |
handoff |
Writes (or consumes) a standardized handoff doc so planning in one chat and executing in a fresh one just works |
session-merge |
Appends each parallel chat's contribution as a structured block to one consolidation file |
recover-chats |
After a crash, finds every real interactive chat across config dirs and emits labeled claude --resume commands |
| Skill | What it does |
|---|---|
repo-launch |
Takes a repo public: kill-gate the pitch, scrub secrets from git history, draft the Show HN post — grounded in data from 28k+ Show HN posts |
pr-fix |
Reviews a GitHub PR, posts findings as a comment, applies the fixes, pushes, and comments what was fixed and why |
pr-safety |
Assesses the risk of merging a PR: regressions, blast radius, rollback difficulty — posted as a PR comment |
pr-shepherd |
Shepherds open PRs through a manual-merge workflow: content-risk assessment, consensus comments, merge order. Never merges |
continuous-improvement |
Audits a codebase for improvements and implements the best ones as PRs, with approval gates |
codebase-tasks |
Scans a codebase for TODOs, tech debt, and improvements — prioritizes and optionally PRs the quick wins |
| Skill | What it does |
|---|---|
blueprint |
Design a new thing BEFORE building it: fuzzy idea → one executable blueprint with a kill switch |
second-opinion |
Cross-checks the current work against an independent model (Codex/GPT) as an adversarial reviewer |
tool-triage |
Decides how a discovered tool enters your stack — PORT / ADOPT / WATCH / PASS, standing bloat cost counted |
prose-restraint |
Writing discipline for substantive prose: state it once, show the implication, trust the reader |
prompt-retro |
Mines your own Claude Code chat history for repeated prompts, corrections, and friction — then proposes the skills and hooks to fix them |
| Skill | What it does |
|---|---|
domain-scout |
Finds an available, brandable domain: generates names across angles, checks whois/RDAP/DNS, screens trademark collisions, ranks |
site-launch |
Concept → live landing page: buy the domain, build the page, deploy to Vercel, wire DNS, SSL, and a form backend — verified end to end |
| Skill | What it does |
|---|---|
skill-governance |
Router for the skills system: index, audit, conflict-check |
index-skills |
Scans every installed plugin and project-level skill and generates a comprehensive skills-index.md |
audit-skill |
Scores any skill against a quality rubric (trigger crispness, scope, clarity, tool permissions) and detects conflicts |
Standalone skills that live in their own repos (and are therefore not duplicated here):
- prior-art — scans the open-source/research landscape for projects doing what your repo does; novelty verdict + steal list
- keypop — credential intake via native dialogs instead of pasting secrets in chat
- prompt-creole — the prompt-form conventions models read best
- supabase-keepalive — keeps free-tier Supabase projects from pausing
MIT
Built by Dylan Pulver, software engineer and entrepreneur in Toronto.