菜单
Use to bring the whole repo back in sync after a batch of changes. Routes to every update-* skill in deterministic order and aggregates the resulting diff.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when the user wants to add, update, or remove entries in the CV (src/data/cv.json), or wants recommendations on what to add, remove, or emphasize. Always validates the result against schemas/cv.schema.json before finishing.
Use when the Visual workflow (.github/workflows/visual.yml) is failing on `main` or on a PR. Reproduces the failure, identifies which commit broke each snapshot, decides whether the diff is intentional, re-records baselines if so, and commits the new pixels.
Use when the user wants to clean up the open Dependabot PR queue. Walks the open Dependabot PRs newest-first, merges the green ones, then rebases and fixes the CI-failing ones, pushes, waits for CI, and merges.
Use when this repo may have drifted out of conformance with OSS_SPEC.md. Runs the bash mirror of the oss-spec validator (no Rust toolchain required), walks the violations, and fixes each one until the validator reports zero violations.
Use when the user wants to refresh the `tagline` and/or `description` of one or more companies in `src/data/cv.json` using their declared `sourceUrls`. Fetches each source, synthesizes a tight bilingual tagline + description pair, propagates the change into any matching `experience[].printDescription` (and assignment `printDescription` when the assignment's `clientId` resolves to a company being refreshed), and validates the result.
Use when files under docs/ may be stale. Discovers commits since the last docs sweep and reconciles each docs/*.md file with its sources of truth (component CSS, schemas, search index, design tokens).
| name | maintenance |
| description | Use to bring the whole repo back in sync after a batch of changes. Routes to every update-* skill in deterministic order and aggregates the resulting diff. |
Umbrella skill that dispatches to every update-* and sync-* skill
in the repo. Does no rewriting of its own — only schedules other
skills and aggregates their output.
Run the skills below in order. Skip a row if its target artifact does not exist yet.
| Order | Skill | Target |
|---|---|---|
| 1 | update-docs | docs/ ↔ source (tokens, schemas, scripts, search internals) |
| 2 | update-readme | README.md |
| 3 | sync-design | src/styles/ + src/components/ ↔ docs/DESIGN.md |
| 4 | sync-cross-browser | src/styles/ cross-browser parity (Safari is the master) |
| 5 | sync-oss-spec | Whole repo ↔ OSS_SPEC.md (final residual-drift sweep) |
Add a row every time a new update-* or sync-* skill is added —
the registry is the only source of truth for which sync skills exist.
Run order matters: update-docs first so prose in docs/ is
reconciled against the code before any downstream skill reads it;
update-readme next so the README picks up any docs-section renames
or content changes; sync-design runs after so it can react to
documentation drift exposed by the docs and README sweeps;
sync-cross-browser runs next so any property-level patches it
applies sit on top of a design-conformant tree (it never retunes
design tokens, only adds missing prefixes and standard pairs);
sync-oss-spec runs last as the residual-drift sweep that catches
any structural conformance gaps the per-artifact skills did not
touch (per OSS_SPEC.md §21.5).
The sibling .last-updated file records the hash of the last successful
sweep across all skills in the registry.
chore: sweep maintenance skills..agent/skills/maintenance/.last-updated with the new HEAD
commit hash.make fmt-checkmake lintmake buildAll three must pass before the sweep is considered successful.