Generate dependency graphs, audit pyproject.toml declarations against version policy, explore unused dependency APIs that could simplify code, and modernize code against the changelogs of upgraded dependencies.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.
Generate dependency graphs, audit pyproject.toml declarations against version policy, explore unused dependency APIs that could simplify code, and modernize code against the changelogs of upgraded dependencies.
compatibility
Designed for Claude Code. Recommended model: Opus.
You help users understand and maintain their project's dependencies. This skill has four modes: graph (visualize the resolved dependency tree), review (audit pyproject.toml declarations against version policy), explore (find unused dependency APIs that could simplify existing code), and (refactor code to adopt features from recently-upgraded dependencies, applying the changes).
modernize
Determine invocation method
If the context above shows CANONICAL_REPO, use uv run repomatic.
Otherwise, use uvx -- repomatic.
Gate the uvx form with the supply-chain cooldown: uvx --exclude-newer '1 week' --exclude-newer-package repomatic=P0D -- repomatic. The window matches [tool.repomatic] minimum-release-age; repomatic itself is exempt because a fresh release must stay installable, while its dependency tree stays gated. See claude.md § Cooldown on every install.
Mode selection
No arguments: Run both graph and review all.
graph: Generate and analyze the dependency graph only.
explore [<package>]: Search for unused dependency APIs that could simplify existing code.
modernize [<package>]: Refactor code to adopt new features from upgraded dependencies, applying and test-gating each change.
If $ARGUMENTS starts with --level or a number, treat it as graph mode with those arguments.
Graph mode
Mechanical layer
The _release-engine.yaml workflow's update-dep-graph job regenerates the dependency graph on release commits only, to avoid noise from transitive dependency changes. This mode is useful for interactive analysis — understanding the graph, spotting concerns, or generating it before pushing.
Argument handling
Pass remaining arguments through to <cmd> update-dep-graph.
If no extra arguments, run <cmd> update-dep-graph with no arguments.
After running
Display the Mermaid output.
Analyze the graph: count total dependencies, flag deep dependency chains, identify packages with high fan-in (many dependents) or fan-out (many dependencies).
Highlight any notable patterns or potential concerns (e.g., single points of failure, overly deep transitive chains).
Review mode
Mechanical layer
<cmd> lint-deps decides everything readable from pyproject.toml alone: upper bounds, missing specifiers, unsorted lists, type stubs outside the typing group, floors with no comment above them, and floor comments running past [tool.repomatic] lint-deps.comment-word-threshold words (40 by default). Run it first and let it own those rows. CI runs it from the release lane's lint-deps job (_release-build.yaml, reached through release.yaml), which fires on every push but annotates with --no-fatal until a release commit makes the shippability findings fatal. These policy rows never block either way, so no CI gate will catch them for you.
What is left is the part no parser settles, and it is the reason this mode exists: whether a floor is justified by the APIs the code actually calls, whether a comment is stale or weak, and whether a rationale contradicts its conditional marker. Spend the review there.
Scope selection
all (default when no sub-argument): Run all checks below.
runtime: Review [project].dependencies only.
dev: Review [dependency-groups] and [project.optional-dependencies] only.
policy: Print the policy summary without auditing.
Version specifier policy
These conventions are derived from the pyproject.toml files across all kdeldycke/* repositories. User-facing documentation of the same content is in docs/dependencies.md.
Runtime dependencies ([project].dependencies)
Use >= (not ~= or ==). Relaxed lower bounds give packagers freedom to release security hotfixes without waiting for an upstream bump. Upper bounds are forbidden per https://iscinumpy.dev/post/bound-version-constraints/.
Every version bound needs a comment tying the floor to a concrete code dependency. The comment goes on the line above the dependency and states which feature, method, or API from that version the project actually uses. Prefer referencing the call site or module that depends on it:
# wcmatch 10.0 changed globbing semantics; our sync_gitignore() relies on# the new symlink-aware matching behavior.
"wcmatch>=10",
A good floor comment answers: "if someone installed an older version, what would break and where?" If you cannot point to a concrete usage, the floor may be unnecessarily high.
A floor comment documents the floor as it stands, in one short paragraph. It is not a record of how the floor got there. Left alone it drifts that way on its own: each bump appends a paragraph about the newly required version, nothing is deleted, and the comment becomes a private changelog of the dependency with the declared version buried under the floors it replaced. When raising a floor, rewrite the comment rather than extending it: keep what the new version buys (API, fix, requires-python alignment, the call site consuming it, a CVE or upstream issue identifier), delete every superseded floor (git log -- pyproject.toml keeps that history), and move out what is not about this floor (usage detail belongs in the module using it, a comparison against an alternative package in an XXX pointer to the upstream ticket). lint-deps warns past 40 words, which is the ceiling to write to even where it is not run.
Security fixes are also a valid floor bump reason. A CVE or advisory in an older version justifies raising the floor even when the API is unchanged. The comment should cite the CVE or advisory:
Python version support is not a valid reason to bump a floor. The dependency resolver already picks the right version via requires-python metadata. If boltons>=20 works and boltons 25 merely adds Python 3.13 support, keep >=20 — the resolver handles it. Exception: when a dependency drops a Python version your project still supports (or your project drops one, aligning minimum requires-python), that alignment is a valid floor bump reason. The comment should state the version range alignment, not the Python support:
Use conditional markers for Python-version-gated deps. Example: "tomli>=2; python_version<'3.11'". When a dep has a version marker, the floor rationale must make sense for the Python versions where the dep is actually installed — not for versions excluded by the marker.
Alphabetical order within the list.
Development dependencies ([dependency-groups])
Prefer [dependency-groups] (uv standard) over [project.optional-dependencies] for test, typing, and docs groups.
>= is preferred for dev deps too, but ~= is acceptable when stricter pinning reduces CI randomness. If a package also appears in runtime deps, the dev entry must use the same specifier style. The relaxation is about specifier style (~= allowed), not about floor accuracy — dev dep floors still need to be grounded in actual API or compatibility requirements, not adoption timestamps.
Standard group names:test, typing, docs (lowercase, alphabetical).
Type stubs go in the typing group with stub-specific versions: "types-boltons>=25.0.0.20250822".
Alphabetical order within each group.
General rules
No upper bounds (<, <=, !=, ~= that implies an upper bound). The only exception is conditional markers like python_version<'3.11'.
Extras syntax is fine: "coverage[toml]>=7.11".
One dependency per line for readable diffs. Short groups that fit on one line are acceptable — the format-pyproject job normalizes layout automatically.
Audit procedure
Read the full pyproject.toml. lint-deps already reports the specifier style, missing comments, ordering, bare dependencies and misplaced type stubs, so read its output rather than re-deriving them. For each dependency entry, check what it cannot:
Check
What to flag
Weak comment
Comment cites Python version support instead of a concrete code dependency. Flag unless it documents a requires-python alignment or a Python version drop
Stale comment
Comment references a reason that no longer applies (the cited method was replaced, or the Python version left the support matrix)
Accreted comment
Comment narrates superseded floors beside the declared one. Rewrite it around the version in force; lint-deps catches only the long ones
Inflated floor
Floor higher than the oldest version providing the APIs actually used (see floor verification below)
Marker/rationale mismatch
Floor rationale contradicts the conditional marker (e.g., "Python 3.14 wheels" on a dep gated by python_version<'3.11' — that dep is never installed on 3.14)
Section style
[project.optional-dependencies] used where [dependency-groups] would be appropriate
Conditional markers
Missing Python version marker for backport packages
Stale cooldown exceptions
exclude-newer-package entries in [tool.uv] for packages that no longer need them (see below)
Floor verification
Comments and changelogs can lie; the codebase is the source of truth. For each dependency with a weak or suspicious comment, verify the floor against actual usage:
Grep for imports. Search the source tree for all imports from the package. List the specific APIs used (functions, classes, constants).
Determine the oldest version providing those APIs. Check when the API was introduced — changelogs, release notes, or pip index versions <pkg> to see what exists on PyPI.
Lower the floor when it exceeds the oldest compatible version. Prefer conservative minimums (the major version that introduced the API) over aggressive ones. Update both the version specifier and the comment.
Run uv lock after any floor change to verify the lock still resolves.
Special cases
Backport packages (like backports-strenum, tomli, exceptiongroup) exist solely to provide a stdlib class to older Python versions. Their entire API is the backported class itself, available in all versions. The floor is typically >=1 (or the first release) unless a specific bug fix is needed for the Python versions where the dep is actually installed.
Conditional deps with stale bug-fix floors. A dep gated by python_version<'3.11' that has a floor set for a bug affecting Python <3.8.6 — if the project's requires-python is >=3.10, that bug is irrelevant and the floor can be lowered.
pytest plugins with no special API beyond auto-registration (like pytest-randomly, pytest-github-actions-annotate-failures) have low effective floors — their basic functionality has been stable across major versions. Set the floor at the major version introducing the current plugin interface, not at the latest release.
Floor bumps to adopt new APIs
A floor bump is justified when a newer version of an existing dependency provides an API that replaces hand-rolled code in the project. This is the flip side of floor verification: instead of checking whether the floor is too high, check whether it could be raised to unlock a simplification.
A valid simplification bump must:
Replace existing code, not add new features. The goal is less code, not more capability.
Be a net reduction in complexity. Swapping a one-line comprehension for a library call is not a win.
Use the public API of the dependency. Private/undocumented attributes do not count.
Update the floor comment to reference the new API and the code it replaces.
When explore mode identifies a candidate, the review output should include it as an Info-level suggestion with the current code, the replacement, and the version that introduced the API.
Red flag patterns in comments
These comment patterns typically signal a floor set at adoption or auto-bump time, not at an API boundary:
"First version we used" / "first version when we last changed the requirement" — the floor is an artifact of when the dep was added or last bumped by a dependency bot, not a deliberate API minimum.
"First version to support Python 3.X" — unless it documents a requires-python drop alignment or a concrete build failure (missing wheels that cause install failures on that Python version), this is not a valid floor reason.
The ~= -> >= conversion pipeline. A common inflation path: (a) dep added as ~=X.Y (latest at time), (b) a dependency bot bumps to ~=X.Z, (c) a bulk "relax requirements" commit converts all ~= to >=. Each step inflates the floor without API validation. Check git log for this pattern when a floor looks suspiciously high.
exclude-newer-package cooldown audit
The [tool.uv] section may contain exclude-newer-package entries that exempt specific packages from the global exclude-newer cooldown window. They arrive from two places: written by hand (the package is published by the same maintainer, or is developed in-repo), or written by audit --fix, which reaches a CVE fix still inside the window through an entry rather than lifting exclude-newer for the whole tree.
Expiry is mechanical, so do not audit for it.sync-uv-lock owns the lifecycle of every entry and reports each move in its PR body: it rewrites a relative span ("0 day") into a fixed cutoff pinned to the locked version's upload time, which holds the package instead of letting it track latest, then prunes the entry outright once that held version ages past the global cooldown and the package rejoins normal resolution. A live fixed cutoff is therefore a freeze doing its job, not a leftover: proposing its deletion un-holds a package the freeze was deliberately pinning. A surviving relative span means the freeze has not run yet, not that a span is the steady state.
What is left for this audit is the part no schedule settles:
Is the package still a dependency? If it was removed from [project].dependencies and all [dependency-groups], the entry is dead weight that no prune will ever reach, since pruning keys off the locked version's age and the package is no longer in the lock.
Is a hand-written exemption still justified? A same-maintainer or in-repo package keeps its span indefinitely and correctly. An external package exempted during a migration that has since finished no longer needs one.
Does the comment explain the reason? Like version floors, a hand-written exemption should carry a comment naming why the cooldown does not apply to it.
Flag those as warnings. Never flag an entry merely for existing or for looking old.
Cross-repo reference
When the context shows DOWNSTREAM, also compare the dependency list against the canonical repomaticpyproject.toml, fetched at the version this repo has adopted rather than at the tip of main: take the tag from the uses: pins in .github/workflows/, then run gh api "repos/kdeldycke/repomatic/contents/pyproject.toml?ref=vX.Y.Z" --jq '.content' | base64 -d. An unpinned fetch resolves to main, whose floors may have moved for a release the downstream repo cannot use yet, turning unreleased work into a phantom "downstream is behind" finding. Use it to identify:
Shared dependencies where the downstream floor is lower than upstream (may be missing a needed bump).
Shared dev dependencies where upstream has moved to a newer group structure.
Output format
Produce:
Policy compliance summary: A table with one row per dependency: name, specifier, has comment (yes/no), issues found.
Warnings: Missing or stale comments, ordering issues, inflated floors, marker/rationale mismatches.
Info: Suggestions for floor adjustments based on API verification or cross-repo data.
Suggested fixes: For each error/warning, show the current line and the recommended replacement. For inflated floors, include the verified API minimum and a rewritten comment.
Explore mode
Search for unused APIs in existing dependencies that could replace hand-rolled code. This is purely analytical: it produces recommendations, not changes.
Scope selection
No argument: explore all runtime dependencies.
<package>: explore a single dependency (e.g., explore boltons).
Procedure
For each dependency in scope:
Catalog current usage. Grep the source tree (not tests) for all imports from the package. List every function, class, and constant actually used, with file locations.
Catalog available APIs. Using your knowledge of the library (and its docs if needed via WebFetch), list the public APIs the project does NOT currently use. Focus on utilities, helpers, and data structures: the kind of thing that replaces 3-10 lines of hand-rolled code.
Search for replacement candidates. For each unused API, grep the source tree for code patterns it could replace. Be specific about what constitutes a match:
Library API
Pattern to search for
boltons.iterutils.partition
Two complementary list comprehensions filtering the same iterable
boltons.iterutils.first
next(iter(...), None) or seq[0] if seq else None on non-generator sequences
boltons.iterutils.bucketize
Loops building a dict[K, list[V]] via setdefault(k, []).append(v)
boltons.iterutils.chunked
Manual slice loops (for i in range(0, len(seq), n))
boltons.dictutils.subdict
{k: v for k, v in d.items() if k in keys} or if k not in exclude
boltons.fileutils.atomic_save
Path.write_text() on files where partial writes would corrupt state
packaging.specifiers.SpecifierSet
Manual version-range checks with </>/in loops over Version objects
packaging.requirements.Requirement
Regex parsing of PEP 508 requirement strings
packaging.markers.Marker
Regex parsing of PEP 508 environment markers (only when the public API exposes the needed structure)
wcmatch.fnmatch / wcmatch.pathlib
stdlib fnmatch or pathlib.Path.glob() that would benefit from brace expansion, negation, or extended patterns
pyproject_metadata.StandardMetadata properties
Manual toml_dict.get("project", {}).get("field") when a StandardMetadata instance is already available
Verify each candidate. Read the actual code at each location. Discard false positives:
A one-line comprehension is not worth replacing with a function call.
next() on a generator is idiomatic Python; first() adds a dependency import for no clarity gain.
Path.write_text() is fine when the file is immediately committed by CI or when partial writes are harmless.
TableFormat.GITHUB hardcoded in functions that generate markdown for PR bodies is correct: the --table-format CLI option is for terminal output only.
Stdlib glob.glob() with recursive=True is fine when no extended glob features (brace expansion, negation) are needed.
packaging.markers.Marker only helps if the public API exposes the structure you need. Its primary public method is evaluate(), not AST inspection.
Check version requirements. For each surviving candidate, determine which version of the dependency introduced the API. Compare against the current floor. If a bump is needed, it must satisfy the floor bump criteria.
Output format
Produce a table of findings:
Dependency
Unused API
Location
Current code (summary)
Replacement
Version needed
Verdict
boltons
subdict
metadata.py:2232
dict comprehension filtering by key set
subdict(metadata, keys)
any (available since 16.x)
Skip: one-liner, no clarity gain
pyproject-metadata
StandardMetadata.keywords
cli.py:2733
toml.get("project", {}).get("keywords")
metadata.pyproject.keywords
current floor sufficient
Adopt: parsed object already available
Only recommend changes where the replacement is a genuine simplification: fewer lines, better error handling, or elimination of a manual reimplementation.
Filtering noise
Most dependencies are already well-used. Expect the majority of candidates to be discarded during verification. A run that produces zero recommendations is a valid outcome: it means the codebase is already leveraging its dependencies effectively.
Common false-positive patterns to reject early:
Swapping idioms for library calls.next(iter(x)) → first(x) adds an import for no clarity gain.
Adding atomicity where none is needed.atomic_save on files that are immediately git-committed or overwritten by CI.
Unifying glob implementations. stdlib glob and wcmatch.glob serve different purposes; not every glob.glob() call needs extended syntax.
Replacing regex with structured parsing when the regex is simpler. re.match(r"extra\s*==\s*'([^']+)'", marker) is more direct than navigating a Marker object's internal structure.
Modernize mode
explore finds simplifications hiding in any installed dependency and only reports them. modernize is narrower and active: it works from the dependencies that changed version recently, reads what those versions added, and applies the resulting simplifications, gated by the test suite.
[!WARNING]
This is the one mode that edits code on its own, and it acts on third-party changelogs it can misread. Every change must be behavior-preserving and verified against the local test suite before it stays. Run it where you can review the diff, and treat a failing test as a veto, never something to "fix" by loosening the test.
Scope selection
No argument: every dependency upgraded since the last release tag.
<package>: a single dependency (e.g., modernize click-extra).
Procedure
Find the version deltas. Determine which dependencies changed, and from and to which version, cheapest source first:
The most recent sync-uv-lock PR — its body lists every bump with its old and new version and a changelog or compare link per package. Find it with gh pr list --search 'head:sync-uv-lock' --state all --limit 1, then gh pr view <number> --json body.
Failing that, find the last release tag (git tag --sort=-v:refname | head -1) and diff the lockfile against it (git diff <tag> -- uv.lock), reading the version pairs.
For a single named package, read its locked version and the pyproject.toml floor.
Read each changelog for the delta. Fetch the release notes covering that version range from the link in the bump table (GitHub releases via gh api repos/{owner}/{repo}/releases, or WebFetch on the compare URL; the PyPI project page otherwise). Extract only what is new or changed in the range: added public APIs, fixed bugs our code works around, and deprecations of APIs we still call. Degrade gracefully: if WebFetch is unavailable and the package is not on GitHub, fall back to your own knowledge of the library's release history and lower your confidence accordingly.
Map deltas to our code. For each new or changed item, grep the source tree (not tests) for code it touches, reusing the candidate patterns and false-positive filters from Explore mode:
A new helper that replaces hand-rolled logic.
A bug fix that lets us delete a workaround — search comments for the package name, work around, TODO, and version-guarded branches.
A deprecation we still call, which must move to the replacement before the dependency removes it.
Apply one dependency at a time. Make the edits for a single package, keeping each behavior-preserving. If adopting an API needs a higher floor, raise it and rewrite the comment per Floor bumps to adopt new APIs, then run uv lock.
Verify before moving on. Run the project's tests, mypy, and ruff (the same fast local channel /babysit-ci and /repomatic-ship rely on). If anything fails, fix it within the same change or revert that package's edits. Only move to the next dependency once green. A change that cannot be made green is reverted, not forced.
Report. Summarize per dependency: the version delta, what was adopted or dropped, the files touched, and anything skipped with the reason. A dependency whose changelog offers nothing actionable is a valid no-op.
What not to touch
New capability. Adopting a feature to add behavior is out of scope: this mode only removes or replaces existing code.
Major-version migrations. A breaking upgrade that needs broad rework is a deliberate human project. Report it and stop, do not attempt it autonomously.
Speculative adoptions. If no current code is simplified, do nothing.
Next steps
Suggest the user run:
/repomatic-deps review all to audit version floors and specifier policy.
/repomatic-deps modernize after a sync-uv-lock PR lands, to fold the freshly-upgraded dependencies' new features into the code.
/repomatic-audit for a comprehensive alignment check beyond dependencies.
