agentsclimarketplace

Verify repository

Skill lilabrooks/lila-agent-skills/skills/verify-repository

Reusable Agent Skills for Codex and Claude Code, focused on safe Git workflows and repository QA.

Install
npx -y skills add lilabrooks/lila-agent-skills --skill verify-repository

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

3 things to look at

  • 15 days oldThe repository was created 15 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Perform evidence-based repository readiness and final-QA audits. Use when the agent is asked to assess whether a repository is ready for agent or production use, run final QA, verify nothing broke, audit before a commit or release, inspect coverage and quality gates, validate git and ignore hygiene, check generated-project workflows, or clean verification artifacts.

SKILL.md

8.6 KB, as published. Nobody here has run it

Verify Repository

Use the repository's own commands as the source of truth, then add checks that its normal gate does not cover. Preserve user work and distinguish product failures from local-environment damage.

Rules

  • Read all applicable AGENTS.md and override instructions before acting.
  • Capture git status --short --branch before running tools.
  • Treat existing tracked and untracked changes as user-owned unless their origin is known. Do not edit, delete, stage, or commit them implicitly.
  • Keep staging, committing, pushing, and pull-request creation outside this skill. When the user combines verification with publishing, complete the verification boundary and hand the tested state to the GitHub publishing skill.
  • Prefer read-only inspection for an assessment request. Implement repairs only when the user asks for changes or the active request clearly includes them.
  • Run repository-owned gates instead of recreating their lint, test, typing, coverage, build, or documentation logic in this skill.
  • Never promise that a repository has no bugs. Report that no failures were found within the tested scope and list that scope.
  • Request approval when a meaningful check needs network access, external writes, destructive cleanup, or credentials.

Workflow

1. Establish the baseline

  1. Locate the repository root and applicable agent instructions.

  2. Inspect the current branch, worktree state, remotes, recent commit, manifests, lockfiles, CI workflows, release configuration, and documented commands.

  3. Identify unrelated changes before selecting checks.

  4. Run the bundled read-only audit:

    python3 /path/to/verify-repository/scripts/repo_audit.py --root .
    

    Add repository-specific expectations when evidence supports them:

    python3 /path/to/verify-repository/scripts/repo_audit.py --root . \
      --expect-ignored .venv \
      --expect-trackable uv.lock
    

2. Discover the authoritative gate

Check AGENTS.md, README setup instructions, Makefiles or task runners, dependency manifests, test configuration, and CI. Select the smallest command that represents the repository's normal handoff gate. Examples include make check, uv run make check, npm test, cargo test, or a documented equivalent.

Verify that local commands and CI agree on:

  • supported runtimes;
  • dependency groups and lockfile policy;
  • lint, formatting, typing, test, and documentation checks;
  • coverage mode and threshold;
  • build, packaging, and installed-entry-point checks.

3. Run checks in increasing cost order

  1. Validate diffs, syntax, formatting, and configuration parsing.
  2. Run focused tests for changed behavior.
  3. Run the authoritative repository gate.
  4. Run build, package installation, and installed-entry-point smoke tests when applicable.
  5. Exercise generators or templates in a temporary directory. Include spaces or brackets in paths when the interface accepts arbitrary targets. Avoid # in Python virtual-environment paths unless support is explicit because console-script shebang handling is not portable there.
  6. Choose the temporary snapshot mode deliberately when workflows depend on Git. Use a source archive without .git to verify graceful skips or clear unsupported-mode errors. Initialize and commit the same snapshot before exercising generators that read HEAD, call git archive, or otherwise require a checkout. Report the two results separately.
  7. Run every declared runtime version when the change is runtime-sensitive or the user requests final release confidence.
  8. Validate workflow YAML, shell syntax, and ShellCheck when those files change and the relevant tools are already available.

Use network-enabled setup phases for dependency installation in sandboxed cloud agent environments. Agent-phase internet access may be unavailable. Keep offline checks independent from optional external providers and credentials.

4. Audit agent and repository hygiene

Confirm, when applicable:

  • AGENTS.md contains current setup, verification, and safety instructions;
  • dependency installation is placed in the environment setup phase when it requires network access;
  • .gitignore covers demonstrated local artifacts without hiding lockfiles, examples, agent instructions, hooks, or repository configuration;
  • tracked files do not match ignore rules;
  • coverage and build outputs are absent from the final worktree;
  • branch and remote state match the requested handoff;
  • generated projects inherit valid configuration after token replacement or renaming;
  • changelog comparison links use a real canonical remote rather than a guessed URL;
  • macOS hidden flags, sync-conflict copies, or .icloud placeholders are not breaking virtual environments or tool state.

Use git check-ignore --no-index -v PATH with concrete examples. Avoid adding ignore entries for undocumented conventions merely because a filename sounds tool-specific.

5. Handle failures without masking them

Classify each failure as one of:

  • repository defect;
  • changed behavior needing a regression test;
  • missing dependency or unavailable network;
  • damaged local environment;
  • unrelated user-owned file interfering with a broad gate;
  • expected skip or unsupported platform.

When sync-conflict copies appear during verification:

  1. Record the affected paths before cleanup.
  2. Remove only known disposable caches such as .mypy_cache, .pytest_cache, and .ruff_cache, and only when the task authorizes cleanup.
  3. Rerun the affected check once.
  4. Run the audit utility immediately after the rerun.
  5. If conflict copies reappear, stop the cleanup loop and classify the checkout as under active sync churn. Report final QA as blocked by the environment and recommend moving the checkout outside the synced folder.
  6. Do not automatically delete conflict copies in source, tests, .git, other non-cache paths, or virtual environments.

Preserve the failing evidence. If an unrelated file blocks the broad gate, leave it untouched, test the intended tracked state in a temporary copy, and report both results. A clean temporary result does not make the raw worktree clean.

6. Hand off publishing requests

When the same user request also includes a commit, push, branch publication, or pull request:

  1. Finish verification and authorized artifact cleanup first.
  2. Record the verified HEAD, intended changed paths, verification command, result, and final git status --short --branch.
  3. Hand ordinary publishing requests to a host-native GitHub publishing workflow when one is available. Use $github-publish-changes only when the user selected it or its specialized basic-commit, exact-identity, mixed-worktree, fork, force-lease, or CI-monitoring behavior is required. Use the installed GitHub publisher when no host-native workflow is available.
  4. Leave all staging, commit, push, and pull-request actions to that publishing skill. Do not duplicate them here.
  5. Require the publishing skill to rerun affected checks when hooks, generated files, conflict copies, or any other change makes the tree differ from the verified state.

7. Clean and report

Remove only artifacts created by the verification run, and only when cleanup is requested or clearly part of the task. Preserve virtual environments unless the user authorizes recreation. Re-run git status after cleanup without invoking tools that regenerate caches.

Report:

  • the authoritative gate and its result;
  • test counts, skips, coverage, runtime versions, and smoke-test outcomes;
  • repairs made and regression tests added;
  • remaining blockers, unrelated files, and untested surfaces;
  • current branch, commit, and worktree state.

Keep the final result concise and lead with readiness or the blocking finding.

Bundled utility

scripts/repo_audit.py is read-only. It reports git/worktree hygiene, tracked files matching ignore rules, local branch state, generated build or coverage outputs, hidden .pth files, sync-conflict copies, and iCloud placeholders. Routine ignored test caches do not produce findings. The utility accepts repeated --expect-ignored and --expect-trackable arguments for project-specific ignore-policy checks. Treat its warnings as evidence to assess, not universal policy violations.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.