Content
77%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 a strong, highly actionable operational spec: exact commands, validation checkpoints, and a well-sequenced fix-review loop with edge cases handled. Its weaknesses are density and monolithic structure — the stall-detection rules and examples read as reference material inlined into the main flow of an already-long single file.
Suggestions
Move the stall/streak signaling rules (the two long paragraphs on file-set and severity streaks) into a references/ file (e.g. references/stall-signals.md) and keep a two-line summary plus the advisory text in SKILL.md.
Tighten the duplicated --since snippet by defining it once and referring back to it for the re-review step, and condense the repeated edge-case clauses in the streak rules into a short bullet list.
Consider moving the two worked examples to a references/examples.md, keeping only one compact example inline to illustrate the loop shape.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly operational specifics with no explanation of concepts Claude already knows, but it could be tightened: the stall/streak-tracking rules (lines ~226–263) are convoluted and restate edge cases redundantly, and the full --since heredoc snippet is duplicated verbatim. This lands between 'efficient with minor trimmable spots' (4) and clearly 'could be tightened' (3), and the density of the streak section pushes it to 3. | 3 / 5 |
Actionability | Guidance is fully executable: copy-paste-ready bash with safe heredoc patterns, exact roborev CLI invocations with flags, exit-code semantics for --wait, job-ID extraction instructions ('Enqueued job <id> for ...' line), and concrete worked examples covering the common branch and --since flows. Only inherent placeholders (<commit>, <job_id>) remain, which is appropriate. | 5 / 5 |
Workflow Clarity | The multi-step process is clearly sequenced (sections 1–4 with 3a–3d) with explicit validation checkpoints: branch match check, --since resolution and ancestor verification, test runs after fixes, pass/fail gating on the full-scope re-review, and iteration-limit handling. The skill is itself a feedback loop with error-recovery guidance (daemon errors, hook-review cleanup), matching the top anchor. | 5 / 5 |
Progressive Disclosure | There are no bundle files at all (no references/, scripts/, or assets/), so everything — including the ~38-line self-contained stall-signaling rule spec and the ~35-line examples section — is inlined in a 313-line SKILL.md. Sections are well-organized with clear headers, but reference-grade material is inline in the main flow with no external split, matching 'some structure but content that should be separate is inline'. | 3 / 5 |
Total | 16 / 20 Passed |