Content
81%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is an exemplary orchestrator-style skill: lean and prescriptive, with an explicit phase order, hard validation gates (regression test fails-before/passes-after, suite result recorded), well-defined escalation and blocked states, and a clean split between the overview and `references/phases.md`. The residual gaps are minor — a few sentences that restate constraints, and a two-level reference chain instead of one level.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense, rule-per-sentence prose that assumes competence — it never explains what a browser, git, or a dev server is — with concrete templates like `mktemp -d "${TMPDIR:-/tmp}/ce-dogfood-XXXXXX"`. A few passages could be trimmed (the host-specific invocation-rendering rules, and restatements like "The task list is session-scoped, but the report on disk is what a later run or a teammate resumes from"), so it sits at efficient-with-minor-trimmings rather than every-token-earns-its-place. | 4 / 5 |
Actionability | Guidance is largely executable: exact paths (`.compound-engineering/config.yaml`, `<root>/dogfood-reports/<YYYY-MM-DD>-<branch-slug>-dogfood.md`), a copy-paste temp-dir command, an explicit status vocabulary (`Pass`, `Fixed`, `Skipped`, `Blocked (needs human verify)`), and named delegation targets (`ce-worktree`, `ce-debug`, `ce-commit`, `ce-compound`). It stops short of score 5 because the actual execution steps for the common cases are deferred to `references/phases.md` rather than covered in the body itself. | 4 / 5 |
Workflow Clarity | The sequence is explicit and invariant ("Scope -> analyze the diff -> map the flows -> derive the matrix -> serve -> execute -> fix loop -> report. The order is the invariant"), with explicit validation ("A fix is not done until a regression test fails before it and passes after"), a checkpoint rule ("Create the report ... as soon as the matrix exists ... update it after each scenario"), the automated suite run recorded, and defined error-recovery paths (escalation to "Decisions for a human", terminal Blocked states). This matches the anchor with explicit validation steps and feedback loops. | 5 / 5 |
Progressive Disclosure | The 60-line body is a genuine overview that clearly signals its references ("Read `references/phases.md` before Phase 0 ... the run cannot be executed correctly from the phase list below") and the report template, and every referenced path resolves to a real bundle file. It misses score 5 because the bundle is two levels deep — SKILL.md -> phases.md -> agent-browser-driver.md / test-matrix-taxonomy.md / packs-resolve.py — rather than the ideal one-level-deep structure. | 4 / 5 |
Total | 17 / 20 Passed |