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。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/PaulRBerg/agent-skills --skill node-deps-bumper命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
Assess recurrence risk for agent behavior using local Codex/Claude Code transcripts and recommend evidence-backed durable fixes to AGENTS.md or skills. Not for routine post-success skill-evolution or agent self-improvement reviews.
Create or polish repo agent context: README.md, AGENTS.md/CLAUDE.md, installed project skills, and other Markdown context docs.
Use for current official documentation about Codex, Codex CLI, or Claude Code behavior, configuration, prompting, skills, permissions, tools, surfaces, capabilities, troubleshooting, hooks, app-server, or hook trust; fetch the relevant official URL before answering.
| argument-hint | [--dry-run] [package ...] |
| disable-model-invocation | false |
| effort | medium |
| model | sonnet |
| name | node-deps-bumper |
| user-invocable | true |
| description | Use for dependency updates: bump npm/pnpm/yarn/bun packages, check outdated, or run taze. |
Use Taze to build one structured update plan, apply compatible ranged updates, and make major-version decisions as a batch.
Resolve the skill directory and save the helper plan from the target repository:
bash <skill-dir>/scripts/run-taze.sh --plan [--include package-a,package-b] > <taze-plan.json>
The JSON plan classifies every discovered update as apply, review-major, review, or skip-fixed. The helper
detects monorepos, includes locked versions during scans, and mirrors Bun minimum-release-age settings. If the
repository uses package-manager age gates or Bun catalogs, read
references/conditional-workflows.md for that active branch only.
If --dry-run was requested, present the plan and counts, then stop without changing manifests or lockfiles.
Select every ranged minor/patch update marked apply. Never auto-approve a major package by name. Present all
review-major and unknown updates in one decision batch with current version, target version, package role when
discoverable, and relevant migration/release notes. Apply only the majors the user selects.
If nothing is selected, report the no-op and stop. If the root manifest uses Bun catalogs, preview the exact selected catalog transitions from the accepted plan:
uv run <skill-dir>/scripts/update-bun-catalogs.py \
--root <repo> --plan <taze-plan.json> --include package-a,package-b
The preview is read-only. Missing catalog entries, conflicting plan rows, unsupported versions, or a catalog value that no longer matches the plan fail before writes. The helper does not select upgrades.
Before the first manifest or lockfile write, discover and run the repository's standard validation suite against the
existing dependency state. Prefer its advertised aggregate check; otherwise run every exposed dependency-resolution,
build, test, typecheck, lint, formatting-check, codegen-check, and repository-invariant command. Use frozen or
non-writing modes where available, and record the exact commands for the post-bump rerun. Attribute every failure
before deciding whether it blocks. Proceed when an unrelated pre-existing failure is reproducible, can be compared
after the bump, and does not prevent dependency resolution or the checks needed to detect regressions; do not fix it
as part of the bump. Dependency or peer-resolution conflicts, actionable unsafe behavior, or a baseline that cannot
provide trustworthy before/after signal block the bump. In that case, stop with
### ⛔ Dependency bump blocked — baseline unusable and report the exact prerequisite and diagnostics without asking
for redundant authorization. Informational notices such as unavoidable deprecations do not block.
Write all selected Taze updates in one command:
bash <skill-dir>/scripts/run-taze.sh --write --include package-a,package-b
For Bun catalogs, rerun update-bun-catalogs.py with the same plan and include set plus --write. It atomically
updates every matching default/named catalog occurrence and preserves each existing ^, ~, or empty prefix. Then
run ni so the repository's package manager updates its lockfile.
Inspect the manifest and lockfile diff. Rerun the exact baseline commands, plus the narrowest checks that exercise the updated dependencies and any required migrations. Treat every newly introduced error, type issue, check failure (including tests, builds, lint, formatting, codegen, and repository invariants), dependency or peer-resolution conflict, and actionable compatibility or safety warning as caused by the bump unless evidence shows otherwise.
Fix every issue caused by the bump, including required source or configuration migrations, while preserving intended
behavior. Do not suppress diagnostics, weaken validation, or change expected behavior merely to make checks pass.
After each fix, rerun the affected check, then rerun the complete recorded suite until it passes. If no clear safe
fix exists within the task's authority, stop with ### ⚠️ Dependency regression decision required. Present all such
issues in one table with the evidence, affected locations, fix and revert options, and likely effects. Do not report
completion until the user chooses, the fix is applied or the offending update is reverted, the lockfile is
regenerated, and the complete suite passes.
Present plans as ### 📦 Dependency plan with counts and a compact table:
| ID | Plan value | Decision | Package | Current → target | Type | Notes |
|---|
Use the plan's exact apply, review-major, review, and skip-fixed values alongside plain-language decisions.
Assign stable IDs to rows needing a choice so the user can answer once. Use ### 🔎 Dry run — no files written for a
preview and ### ✅ No selected updates for a no-op.
Finish applied work with ### 🏁 Dependencies updated, a tree of changed manifests/lockfiles, and
### 🧪 Verification. Use ### ⚠️ Remaining review only for non-blocking informational matters, never for an
unresolved issue caused by the bump. Keep helper JSON, package/version strings, commands, and diagnostics exact and
undecorated.
Completion requires an attributed, comparison-safe pre-write baseline, a reviewed plan, the retained selected updates, a regenerated lockfile, no new failure in the recorded suite or dependency-specific checks, and no unresolved issue caused by the bump. Dry-run completion requires the structured plan and zero writes.