package-updater-update
Use when bringing every outdated dependency and toolchain pin current across one repository or a fleet.
- License 许可
- MIT
- 所属仓库
- tony/skills
- 最近更新
- 2026-11-08
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- package-updater-update
- metadata
- { "source": "plugins/package-updater/skills/update/SKILL.md", "argument-hint": "[--root <dir>] [--repo <path|slug>...] [--owner <name>...] [--audit-only] [--branch <name>] [--pr] [--issue github|linear] [--push]" }
- description
- Use when bringing every outdated dependency and toolchain pin current across one repository or a fleet.
- allowed-tools
- [ "Bash", "Read", "Grep", "Glob", "Edit", "Write", "Agent", "AskUserQuestion" ]
- disable-model-invocation
- true
02 资源文件 8
- 参考文档agents/openai.yaml308 B
- 参考文档references/commit-conventions.md8.5 KB
- 参考文档references/holds.md4.9 KB
- 脚本references/plan_rules.py6.0 KB
- 参考文档references/prose-rules.md4.5 KB
- 脚本references/repo_facts.py27.5 KB
- 参考文档references/repo-scope.md5.7 KB
- 参考文档references/stopped-run.md4.1 KB
03 SKILL.md 正文
Update packages
Bring the repositories in scope up to date: one researched commit per toolchain pin and per named package, a bulk refresh for the rest, and each follow-up as its own commit.
Run the phases of the package-updater-updating-packages skill across the scope; it
names the references and owns the phase structure, so the sweep and the
single-package path cannot diverge.
For one named package, use the package-updater-update-package skill. For
.tool-versions, .nvmrc, packageManager or engines, use
the package-updater-update-toolchain skill.
User arguments: $ARGUMENTS
Procedure
Follow the skill's phases. This command supplies the scope and the fan-out.
1. Scope
Set the scope per references/repo-scope.md, which owns the
flags, the fork rule and the ask-first rule. --owner keeps only
repositories whose remote belongs to the given accounts.
2. Inventory and discover
Run the skill's inventory and discovery phases across the whole scope before touching anything. The inventory gathers each repository's remote, default branch, manifests, lockfiles, toolchain pins and quality commands. A successful lookup may establish that a file is absent; a failed lookup leaves that part of the inventory unavailable. Stop changes to a repository whose required inventory could not be read.
Sort what you find into the commit tracks of
references/commit-conventions.md now. Pipe each --lock-diff or
--manifest-diff result from the skill's Phase 2 into references/plan_rules.py track (resolve the path from this skill's directory to an absolute one):
it puts every major in named, every in-range move in bulk, and lists
under by_hand each file or range the diff could not classify. plan_rules.py track --pin <file>... names the track of each pin file; package.json and
go.mod print mixed because their toolchain keys and dependency keys
land as different tracks.
Judgment stays with you: which in-range package earns its own commit under the "Named package bumps" conditions in that reference, and which packages form a release train. A move sorted wrong here produces a commit that cannot be reverted independently later.
3. Research, one chain per distinct move
Group by upgrade chain rather than by repository: twenty repositories taking the same uv release share one piece of research. Dispatch a subagent per chain where the host supports it. Verify every release URL before it reaches a commit body, and intersect each finding with the repository it lands in.
The bulk refresh needs no research and carries no claims.
4. Orchestration plan
Enter plan mode before the first edit. In Claude Code call EnterPlanMode;
in Cursor, Codex, or Gemini use /plan or Shift+Tab.
Where plan mode is unavailable, present the plan as text.
Present the plan the skill defines and wait for approval. The plan
lists each repository with the branch it lands on and its push target.
With --audit-only, stop here and report; nothing is written.
Exit plan mode before landing changes.
5. Land
Work through repositories one at a time, landing per Phase 4 of the
skill and publishing only as Phase 5 allows. Write commit messages and
the report per references/prose-rules.md.
With --issue, follow "When an issue or card comes first" in the skill
for the order, the branch name and --body-file; the plan approved in
step 4 lists the filing. Write the issue or card body and the pull
request text per references/prose-rules.md. The issue body is one sentence naming
the sweep, then ## Outstanding with one plain bullet per repository
and move (package or tool, current version, target, track). The pull
request body is one sentence on what landed, then one bullet per track.
6. Verify, publish and report
Verify each repository before it is pushed or its worktree removed, per Phase 5 of the skill.
Rules
- Tracks and follow-ups are separate commits, as the commit-conventions reference says. Never bundle across them.
- A bump left knowingly red says so in its body and its follow-up lands in the same run.
- A fleet-wide claim is verified per repository before entering that repository's history.
Output
A scope with no manifest, lockfile or pin file halts per
references/stopped-run.md, also under --audit-only; it never
prints an audit hero or "0 outdated".
Otherwise open with a one-line hero (✓ N commits across M repos or, for
--audit-only, ⚠ Audit only: N packages outdated), then these
sections in order, each omitted when it has nothing to report:
## Audit: per repository, the ecosystems found and what is outdated in each, separated into what moves in-range and what needs a manifest edit. Note what was excluded as fork, unowned, or already current, and what was not checked because a lookup failed.## Gated: releases the cooldown currently hides, with the date each becomes visible.## Moves: one entry per distinct chain: what changed upstream, the links to every release in its span, and which repositories it reaches.## Commits: per repository, the commits made in landing order and whether they were pushed.## Follow-ups: config, snapshot and migration commits, and any bump that was red until its follow-up landed.## Verification: the quality checks run per repository and their real results, with any pre-existing failure named as pre-existing.## Handoff: stale GitHub Actions, ruff floors and Terraform versions found, each with the sibling command that handles it.## Holds: perreferences/holds.md, every hold across the tree with its recorded condition, which conditions are now met, and which holds have no recoverable reason.## Drift: pins disagreeing across a workspace, and holds applied unevenly across workspace members.
Reply in this shape, as plain text outside any code fence, with plain bullets, no bold (never - **Label:** ...), and no local paths:
✓ 2 commits across 1 repo
## Audit
- api: requests 2.31.0 to 2.32.3 in range, pytest 7.4.4 to 8.3.2 needs a manifest edit.
- web: left-pad excluded, already current.
## Moves
- requests 2.31.0 to 2.32.3: the headline change of each release, from the notes read in this run, with <release link per version>; reaches api only.
## Commits
- api: 9a3f1c2 py(deps) requests 2.31.0 -> 2.32.3, 4be0d71 py(deps[dev]) pytest 7.4.4 -> 8.3.2. Not pushed.
## Verification
- api: tests and lint pass.
- web: no change.
A halt follows "Plan stop and halt" in the package-updater-updating-packages skill.
Halt example, same plain-text form:
⚠ Stopped: nothing was changed. Ownership of the one repository under this root could not be confirmed.
To continue:
- Confirm you maintain it.
- Name the owner of the repository.
- Stop here.
End with the next-step options of Phase 6, for example: run the sibling command for the handoff items, drop a cooldown exemption whose block has lapsed, take a gated release once it ages out, or stop here.
Portability notes
$ARGUMENTS— the text the user passed when invoking this skill. If your host does not substitute it, read it as the user's request in the current turn, and ask when there is none.- Bundled files — every relative path in this skill points at a file shipped inside this skill directory. Read them from here, not from the host's plugin tree.
04 同仓库更多 Skills
serene-bliss
Use when asked to make a CLI --help, docs page, error message, or tooling UX more serene or calmer for developer experience; returns text, no edits.
review
Weave code review: run independent adversarial reviews in parallel, then synthesize findings
refine
Use when asked to refine a draft, iterate on code until cleaner, or polish a README through repeated critique and improvement.
prompt
Weave prompt: compare independent adversarial implementations, then adopt the single best one
plan
Weave planning: compare independent adversarial plans, then synthesize the best one
fix-review
Fix weave review findings: validate, add test coverage, fix, and commit each as atomic changes