Claude Skills by compforge

Code taste and software design judgment: solution/design review, module & interface boundary shaping, refactoring tradeoffs, naming, maintainability, and documentation layering. First-principles: clarify the original requirement before discussing solutions. Skip for debugging, triage, call-path inspection, or routine small edits.

Commit current changes without pushing. Use when the user says "gcam" or wants to commit only.

Commit current changes and push, without creating a pull/merge request. Use when the user says "gcamp" or wants to commit + push (e.g. add a commit to an existing PR/MR branch).

Commit current changes, push, and create/reuse a pull/merge request (GitHub PR or GitLab MR). Use when the user says "gcampr" or asks to commit + push + open/raise a PR / MR / pull request / merge request.

Commit, push, safely rebase an existing branch, create/read/update/close a pull/merge request (GitHub PR or GitLab MR), cut a feature branch, create/resume a managed worktree, or release. Triggers — gcam / gcamp / gcampr / rebase / 解决冲突 / 提 PR / 提 MR / pull request / merge request / 看 PR / 看 MR / 关 PR / 关 MR / checkout / worktree / 切新分支 / 起新分支 / 发版.

Reconcile local branches and worktrees with GitHub/GitLab PR/MR state, report lifecycle changes, or configure Codex Scheduled tasks for recurring devloop checks. Use when asked to monitor/check PR or MR status, clean finished worktrees, reconcile local PRs, or set up recurring PR/MR monitoring.

Manage devloop automated code-review runs and findings on GitHub/GitLab PRs/MRs: inspect or run a review, independently verify every published finding, record a Verdict, handle valid findings, resolve completed threads, or report missed findings. Use when Review/Review findings context surfaces, or the user asks to run review, process review comments/findings, review and fix, adjudicate, label, or 打标. Not for a direct architecture/code review without devloop artifacts; use code-taste for that.

Normalize a repo Component, run its lint and test validation checks, and report each result. Use when the user asks to validate changes, lint/format code, run tests, or verify before committing or pushing.

Discover, operate, and incrementally extend a project's existing end-to-end test capability. Use when the user asks to run, check, inspect, or report E2E tests; validate a project or deployed environment end to end; identify E2E coverage gaps; or add and strengthen project-owned E2E cases. Follow project-owned tests, operating knowledge, and canonical entrypoints; do not create a parallel test system or present unexecuted coverage as verified.

Discover, run, interpret, compare, and incrementally extend a project's existing system performance capability. Use when the user asks to run or report load, capacity, stress, soak, or performance-regression tests; compare performance runs; or add bounded project-owned perf coverage. Follow project-owned workloads, profiles, SLOs, targets, and entrypoints; do not invent a missing capability or run material load without target authorization.

Discover, run, interpret, and iteratively tune a project's existing agent trajectory evaluation capability. Use when the user asks to evaluate agent trajectories, compare effect and cost across runs, find trajectory smells or the next trajectory problem, or identify optimizations for a known trajectory issue. Operate project-owned trajectory evidence and canonical entrypoints; do not invent missing evidence, verdicts, or a parallel evaluation framework.