CtrlK
BlogDocsLog inGet started
Tessl Logo

canonical/mason

Agent kit for working on canonical/chisel-releases. Cross-agent skills + scripts for authoring and reviewing chisel slice definition files.

81

Quality

85%

Does it follow best practices?

Impact

75%

Average score across 5 eval scenarios

SecuritybySnyk

Low

Low-risk findings worth noting

Overview
Quality
Evals
Security
Files

SKILL.mdskills/chisel-slice-reviewer/

name:
chisel-slice-reviewer
description:
Reviews chisel slice definition files (SDFs), branches or PRs in canonical/chisel-releases for correctness, dependencies, conventions, tests and forward-port requirements. Read-only; returns findings and a verdict.

chisel-slice-reviewer

Review the supplied SDF as it stands, or a branch/PR against its target base. State the scope and any assumption needed to interpret the request.

roots and boundaries

  • <skill-root> is the absolute directory containing this SKILL.md. Resolve kb/ and scripts/ there; keep the checkout as working directory.
  • Run python3 "<skill-root>/scripts/_orientation.py" first. If no checkout is present, ask where it is; do not clone or guess one.
  • Confirm that the files and manifest being checked match the review's revision and target release. Use read-only git and upstream reads; do not switch branches or edit the checkout. If the requested revision is not available to a helper, report that check as unavailable rather than checking different files.
  • Read kb/_verification.md for check status and review completion. Runtime evidence may come from CI for the reviewed revision; do not claim that static inspection or future CI proves runtime success.

1. run static checks

For a branch or PR represented by the working tree:

"<skill-root>/scripts/review-diff.py" --base <target-base-ref>

For a single SDF:

"<skill-root>/scripts/_check-slice.py" slices/<pkg>.yaml
"<skill-root>/scripts/_check-test.py" slices/<pkg>.yaml

Fold findings into the review under kb/_verification.md, then assess design and runtime evidence before choosing the overall verdict.

2. assess correctness and design

Read the references relevant to the changed slices:

  • kb/_slice-definition-format.md and kb/_chisel-formats.md: field semantics, manifest format and tool-version compatibility.
  • kb/_slice-conventions.md: grouping, naming, legal files, path style and preservation of published slice contents.
  • kb/_dependency-policy.md: target package inspection, runtime dependency selection, provider resolution and generated files.

Inspect actual target package contents where needed:

python3 "<skill-root>/scripts/_deb-list.py" <pkg> <arch>

Trace the retained dependencies and compare against the files actually supplied by their slices. Review broad globs, shared paths and mutation effects. For removed declarations, establish whether effective contents are preserved through dependencies; static path comparison cannot decide this.

Validate disputed rules against the live sources in kb/_upstream-sources.md. Distinguish parser errors, demonstrated runtime defects and repository conventions. Do not turn an unsupported stylistic preference into a hard gate.

3. assess tests and upstream checks

Use kb/_spread-tests.md for coverage, assertions and rootfs isolation. Review each affected package, including libraries, data and dependency packages. Explicit binary references alone do not prove execution or meaningful tests.

Inspect the target branch's CI configuration and the reusable workflows it calls, using kb/_upstream-sources.md. Compare available lint, cut and spread results with the reviewed revision and architectures. Mark missing evidence unavailable; do not change the checkout to obtain it.

Read kb/_chisel-branches.md and live contribution guidance linked from kb/_contributing.md to establish forward-port obligations and PR requirements. Check discovery completeness before claiming that all maintained releases are covered. Package disappearance requires archive evidence, not a denied lookup.

review report

Return findings with file, slice/path, evidence and proposed correction:

  • Blocking: demonstrated parse/runtime failures or other hard requirements.
  • Should-fix: justified convention, design or test deficiencies.
  • Nits: optional improvements.

List passed, failed and unavailable checks separately, then give the overall verdict under kb/_verification.md: approve, request-changes or incomplete. Include remaining forward ports and verification gaps. Respond to the user when invoked directly, or the parent agent when delegated.

README.md

tile.json