bwkw
UserPersonal AI development toolkit — skills, hooks, and verification gates that work in both Claude Code and Cursor. Installed once globally; product repositories stay untouched.
Categories
Indexed Skills (6)
design-review
Review a plan or spec before any code exists. Use when a design doc is ready, before implementation starts, or when asked whether an approach is sound. Catches what code review cannot fix later — one-way doors, migration order, rollback, and what the plan omitted. Read-only.
investigate
Map what a change would touch, or trace how something actually works. Use when asked where something lives, what depends on it, or what would break. Answers with file:line evidence under a fixed exploration budget, and names what it could not confirm. Read-only.
pr-describe
Write or update the current branch's PR title and description. Use when preparing a PR for review, or when the description is stale. Produces something a reviewer can read before opening the diff. Touches only the PR, never repository files.
skills-audit
Audit installed skills for bloat and breakage. Use before adding a skill, when skills stop firing automatically, or for periodic clean-up. Checks description budget, overlapping coverage, Cursor incompatibility, oversized bodies, and disuse.
verify
Run this repository's own verification commands and report with evidence. Use before claiming work is done, before committing, or before opening a PR. Resolves commands from the repo's profile rather than guessing, and refuses what the repo forbids.
review-all
Review a change across every layer it touches, as a tech lead. Use for code review, PR review, or checking work before shipping — especially when a change spans backend, frontend, and infrastructure. Reviews each layer in its own subagent, then finds the irreversible risks that fall between them, like a contract and its consumer shipping out of order. Read-only.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.