From capybaraa
Review the current diff against capybaraa's rules: lean, ask, optimal, terse, clean, sync. Finds guessed specs, speculative abstractions, bad complexity, filler comments, mess left behind, and docs/tests left out of sync. One line per finding, lists only, does not apply fixes. Use when the user says "review against the rules", "capybaraa review", "/capybaraa-review", or asks what's wrong with a change before they ship it.
How this skill is triggered — by the user, by Claude, or both
Slash command
/capybaraa:capybaraa-reviewThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Review the current change (the staged/unstaged diff, or the files the user names)
Review the current change (the staged/unstaged diff, or the files the user names)
against capybaraa's rules. This is a quality pass, not a correctness audit: it hunts
the things a calm senior dev would flag in review, not runtime bugs. For correctness
bugs use /code-review. To scan the whole repo instead of the diff, use
/capybaraa-audit.
Detailed guidance: references/principles.md.
Get the diff: git diff plus git diff --staged (or read the files the user points at).
Read each changed hunk. For every problem, emit ONE line:
path:line: <tag> <what's wrong>. <the fix>.
Group nothing, pad nothing. If a hunk is clean, say so and move on.
End with a one-line verdict: ship it, or fix the flagged lines first.
ask: the spec was guessed where the code or the request did not say. Name the
assumption that should have been a question.lean: something that does not need to exist: an abstraction with one caller, a
config for a constant, a dependency for what stdlib or a few lines already do, dead
flexibility "for later". Name what to delete or inline.optimal: wrong data structure or needless O(n^2) on a hot path. Name the better one.terse: filler comments that restate the code, dead prose, a comment that explains
the obvious. Name the lines to cut.clean: the old version left next to the new, a stale comment, dead code the change
orphaned, or an unsanitized input at a trust boundary. Name what to remove or guard.sync: the change left something that described the old shape behind: a doc or README
naming a renamed symbol, a test asserting the old return shape, a sibling caller of a
deleted export, a lagging version string. Name the reference to update or delete.Spotted a dropped guard (validation, error handling, security)? Flag it in one line too.
src/auth.js:42: lean: AuthFactory wraps one constructor. Inline it, call the class directly.
src/auth.js:88: ask: retry count hardcoded to 3, the ticket never said. Confirm the value.
src/parse.js:15: optimal: nested .find in a loop, O(n^2). Build a Map once, look up in O(1).
src/parse.js:60: clean: old parseLine() still here, dead since line 15 replaced it. Delete it.
verdict: fix the four lines above, then ship.
Lists findings only, does not edit files. Over-engineering, mess, and unfinished work
are in scope. Correctness bugs, security exploits, and performance profiling are not,
that is what /code-review and /security-review are for. If you spot a real security
hole while reading, flag it in one line and tell the user to run /security-review,
do not fix it silently.
npx claudepluginhub katipally/capybaraa --plugin capybaraaProvides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.