CtrlK
BlogDocsLog inGet started
Tessl Logo

code-review

Review pull requests in the Trackdub public core (trackdubllc/Trackdub). Use for Copilot code review, PR review, architecture checks, dependency-direction violations, model/license safety, fake readiness, runtime planner routing, and honest test evidence. Prefer this skill whenever reviewing diffs, pull requests, or suggesting review comments in this repository.

76

Quality

95%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

SecuritybySnyk

Low

Low-risk findings worth noting

SKILL.md
Quality
Evals
Security

Trackdub public core code review

You are reviewing changes in the Apache-2.0 public core (trackdubllc/Trackdub): engine, SDK, CLI, pipeline, inference, and shared libraries.

Private Avalonia desktop shell and gated licensing trust-ring work belong in trackdubllc/Trackdub-gated, not here.

Required context

Before commenting, read and apply:

  1. Root REVIEW.md (authoritative reviewer checklist)
  2. AGENTS.md (dependency direction, model governance, testing)
  3. .github/pull_request_template.md (expected PR body shape)

When the PR mentions a TS-* issue, load that context before judging scope:

  1. Prefer available MCP tools (GitHub issue links, Linear if configured).
  2. If Linear (or other) MCP is unavailable, fall back to the PR body, linked GitHub issues/PRs, and any Linear URL already in the description. Do not invent Linear status. Note when TS-* context could not be verified, then continue using the supplied PR evidence.

Review priorities

Review in this order:

  1. Correctness
  2. Scope control
  3. Architecture and dependency direction
  4. Model/license safety
  5. Runtime route visibility
  6. Test evidence
  7. Docs and milestone alignment

Automatic review stops

Treat these as request-changes findings (not nits):

  • Violates Domain-depends-on-nothing or other dependency-direction rules in AGENTS.md
  • Adds provider or model routing outside IRuntimePlanner / router paths
  • Adds translation behavior based on model-name aliases instead of manifest metadata (engine_family, capabilities, language_coverage), except where the current planner or router already treats aliases as soft hints (see REVIEW.md)
  • Uses a model before manifest, license, and commercial-safe handling exist
  • Treats provider registration, download, stage run, skip, or success as the same readiness state (never fake readiness)
  • Adds a new stage name without updating StageNames and its guard tests
  • Lands Avalonia shell or gated trust-ring product policy in this public core
  • Claims validation the PR body does not support with exact commands

Checklist focus

Architecture

  • Trackdub.Domain depends on nothing
  • Inference stays in Trackdub.Inference / Trackdub.Inference.Onnx
  • Persistence stays in Trackdub.Infrastructure (or existing Media boundaries)
  • CLI entry points stay in Trackdub.Cli; reusable automation stays in Trackdub.Sdk
  • Model wrappers and engine adapters do not mutate project state

Pipeline and readiness

  • Skip/fail paths preserve original artifacts and log exact skip reasons
  • Prefer immutable execution snapshots
  • StageRunRecord.Start(...) uses StageNames.* only (no inline stage-name literals)
  • Provider registered ≠ model downloaded ≠ stage ran ≠ stage skipped ≠ stage succeeded

Models and licensing

  • New packages are centrally versioned in Directory.Packages.props
  • New models update src/Trackdub.Inference/Runtime/ModelManifest/bundled-models.manifest.json
  • Unknown license is unsafe; commercial-safe impact must be explicit
  • Update docs/legal/THIRD_PARTY_NOTICES.md (or equivalent) when requires_attribution: true

Tests

  • Tests at the correct layer; list exact commands run
  • Name and justify skipped tests
  • Pipeline changes cover success, disabled/skipped, missing-prerequisite, and failure
  • Application tests prefer fakes from tests/Trackdub.TestDoubles/
  • New seams/guards ship enforcing tests in the same PR

Comment style

Write concrete, falsifiable comments:

  • Point to the exact file or behavior
  • State the risk
  • State what evidence is missing
  • Suggest the narrowest acceptable correction

Examples:

  • "This makes Domain depend on Infrastructure, which breaks the repo boundary rules."
  • "The PR says CPU fallback is unchanged, but this planner path now silently reroutes."
  • "This adds a model entry point without manifest and license-policy coverage."
  • "The tests listed do not exercise the changed runtime path."
  • "This trust-ring change belongs in Trackdub-gated, not the public core."

Merge bar

A change is merge-ready only when scope is bounded, architecture rules hold, model/license/runtime impacts are explicit, validation is honest, and docs match shipped behavior.

Repository
trackdubllc/Trackdub
Last updated
First committed

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.