agentleFS
WeaverseWeaverse/.agents/skills/dependency-upgrades/SKILL.md

dependency-upgrades

Skill0 starsChanged 6 days agoMITView on GitHub

Verify npm/pnpm upgrades and JavaScript toolchain replacements beyond install success: peer-policy masking, hook rewrites, linter parity and artifact compatibility. Use when upgrading dependencies, fixing advisories or replacing lint, format, compiler or build tools.

4 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

8 rules. 0 are common, 7 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 4 edited versions we read keep these.

  • 7

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Prefer a narrow transitive/lockfile change where the parent already admits the patch.

    line 19
  2. Distinctive:

    Check the advertised Node range against every resolved dependency's engines.

    line 21
  3. Distinctive:

    Keep GitHub alerts, package-manager advisories and unique packages separate.

    line 27
  4. Distinctive:

    When no patched version exists, inspect the parent upgrade or removal path rather than forcing incompatibility.

    line 31
  5. Distinctive:

    Import sorting changes module side-effect order.

    line 43
  6. Distinctive:

    Keep framework route/code generation before typecheck.

    line 45
  7. Distinctive:

    After runner/filter changes, compare executed file sets and full test-name multisets, not just counts.

    line 51
  8. Very specific:

    Before handing off manual testing, install the frozen graph and regenerate local clients/types/icons/bundles without schema pushes.

    line 63

“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Dependency and toolchain upgrades

Resolution is not compatibility

  • npm ci with legacy-peer-deps=true does not establish peer compatibility. Inspect effective npm config and .npmrc files. If claiming strict compatibility, run an isolated frozen install with --legacy-peer-deps=false --strict-peer-deps=true; do not silently change the repository's persistent policy.
  • Lifecycle scripts can rewrite shared Git hooks in worktrees. Record hook state before an install and compare afterwards. Use an authorized script-free install plus explicit generation when appropriate; do not mistake missing generated artifacts for an application defect or weaken hooks to get a green result.
  • Query registry metadata for the exact release. latest can be a prerelease; aliases and calendar versions need semantic review. Respect release-age policy and report held versions. Compare lockfile integrity for advisories.
  • Prefer a narrow transitive/lockfile change where the parent already admits the patch. A major override needs testing at the actual consuming boundary, not just an import. Never dismiss an advisory to clear a count.
  • Check the advertised Node range against every resolved dependency's engines. A lower bound also promises future majors that dependencies may exclude; update runtime selectors, package metadata and docs together.
  • Resolve duplicate framework/SDK copies: registration through one copy may not reach dispatch through another.

Audit evidence

  • Keep GitHub alerts, package-manager advisories and unique packages separate. Zero open alerts does not prove the frozen lockfile is clean; equal totals do not prove the same findings. A local fix does not close a default-branch alert. Record audit exit separately from any successful collection wrapper.
  • "Unreachable" requires tracing the real consumer, including export/SSR paths that accept stored content. When no patched version exists, inspect the parent upgrade or removal path rather than forcing incompatibility.

Toolchain parity

  • Review migrator edits to manifests, catalogs, overrides, scripts, hooks and source. Cached task runners can strip environment overrides; test the real child with a harmless sentinel rather than trusting the wrapper.
  • Print resolved linter configuration for source and test files. Trigger each previously enforced rule as a positive control, especially hooks/accessibility and type-aware rules that can silently be disabled.
  • Translated suppressions need the new diagnostic span, not just a renamed rule ID. Prove both the intended exception and a fresh violation. A warning-level autofix may still rewrite source in a supposedly check-only lane.
  • To claim no formatter churn, format the baseline in a disposable copy and count changed paths. A green check of already-reformatted candidate files proves only a fixed point.
  • Import sorting changes module side-effect order; moving a JSX prop across a spread changes precedence. Review execution/AST order, not only token equality. Exercise the real fix/staged-file hook path.
  • Keep framework route/code generation before typecheck. API/declaration consumers still using the old compiler make a compiler migration incomplete even if the main typecheck is green.

Tests and deliverables

  • A suite named unit can launch a browser. Read fixtures before executing it and retain browser permission gates. After runner/filter changes, compare executed file sets and full test-name multisets, not just counts.
  • For wire changes, test the unchanged deployed client's parser. Upgrading both test SDK and server can make them agree on a format old clients reject.
  • For package builders, compare ESM/CJS loading, export targets, declarations, sourcemaps, tarball paths and API reports. Equal counts are not parity. Recompute after the final bundler edit; overload and union changes need semantic review even when nearby quote/order changes are cosmetic.
  • Cached Turbo/Nx logs are replayed evidence, not fresh execution. Pass force flags to the task runner itself, not through -- to its child. Container install/prune changes need the real production image target; a host build cannot prove that stage.
  • Transitive packages can break a pinned parent's imports. Inspect the resolved graph and known-good lockfile before proposing an unrelated major upgrade; test scoped overrides in isolation.

Before handing off manual testing, install the frozen graph and regenerate local clients/types/icons/bundles without schema pushes. State branch, SHA, runtime, executed versus cached gates and unverified image/browser work.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.