agentleFS
OrionicsLtdOrionicsLtd/claude-skills/skills/dependency-safety/SKILL.md

dependency-safety

Skill1 starsChanged 23 days agoMITView on GitHub

Guardrails for changing dependencies. Use this WHENEVER a change touches pyproject.toml, lock files, adds/upgrades/pins a package, or edits a Dockerfile or its base image: only touch a dependency when genuinely necessary, make the smallest reviewable change, and never bump versions or regenerate a lockfile casually.

3 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

7 rules. 0 are common, 5 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before touching a dependency, ask: is this necessary?

    line 16
  2. Distinctive:

    After any dependency change, read the lock diff and confirm only what you intended moved:

    line 34
  3. Distinctive:

    If a package you didn't touch changed version, stop and find out why — an accidental transitive bump is the most common silent regression.

    line 37
  4. Very specific:

    If a stray command rewrote the lock, restore it (git checkout <ref> -- <lock>) rather than committing an unintended re-resolve.

    line 42
  5. Distinctive:

    Prefer an explicit, minimal version constraint over "latest." Know what a change to a bound actually allows before you widen or bump it.

    line 47
  6. Very specific:

    If a resolver setting controls reproducibility (e.g.

    line 49
  7. Distinctive:

    If a dependency/runtime issue is already burning, resist shotgun fixes (bump the package *and* swap the base *and* re-lock at once).

    line 55

“Common” is measured: kept by at least 6 in 10 of the 3 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 safety

Changes are minimal — only what is necessary.

Default to no change

Before touching a dependency, ask: is this necessary? Only add, upgrade, or pin one for a concrete reason — a bug you need fixed, a security advisory, a feature the task actually requires. If the task works with what's already there, change nothing.

For a security advisory, follow security-advisories: it confirms the advisory applies and picks the lowest version that fixes it.

  • Don't add a package for something small — prefer the standard library or an existing dependency over a new package for a few lines of functionality.
  • Don't upgrade unrelated packages — a bump that isn't part of the task is scope creep and a review burden. Leave it out.
  • Don't regenerate the lockfile casually — uv lock/uv sync (or npm install, poetry update, cargo update) can silently re-resolve dozens of transitive deps. Regenerate only when you mean to, then review the diff.

Review the lockfile diff

After any dependency change, read the lock diff and confirm only what you intended moved:

  • If a package you didn't touch changed version, stop and find out why — an accidental transitive bump is the most common silent regression.
  • Scrutinize compiled / C-extension packages (numpy, pandas, pyarrow, cryptography, grpcio, pillow, …). They carry native ABI constraints and are the most likely to break a runtime that looked fine at build time.
  • If a stray command rewrote the lock, restore it (git checkout <ref> -- <lock>) rather than committing an unintended re-resolve.

Pin with intent

  • Prefer an explicit, minimal version constraint over "latest." Know what a change to a bound actually allows before you widen or bump it.
  • If a resolver setting controls reproducibility (e.g. uv's exclude-newer), keep it valid for the tool version in use — a value an older tool can't parse may be silently ignored, letting the resolver pull the newest of everything.

When debugging, change one thing at a time

If a dependency/runtime issue is already burning, resist shotgun fixes (bump the package and swap the base and re-lock at once). Change one variable, verify, then the next — otherwise you won't know what actually fixed it, and you risk trading one silent drift for another.

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.