Content
85%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.
A tight, well-structured instruction skill: the survey sources, fix-owner taxonomy, approval gate, and check-verification loop are all concrete and explicitly sequenced. The only improvements are cosmetic — dedupating one stated-twice rule and showing one example finding line.
Suggestions
State the 'invented findings are the same violation as skipping the survey' rule once (Evidence first) and reference it from the Headless section rather than repeating it verbatim.
Include one example finding line in the proposal format (e.g., 'ralph loop retried lint step 7× → deterministic check → add pre-commit rule') so the `finding → surface → intended change` shape is unambiguous.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes Claude's competence (no concept explanations), but contains minor trims: the rule "an invented finding is the same violation as skipping the survey" is stated twice (Evidence first and Headless refit sections), and rhetorical flourishes like "starts in a better yard" and "decoration, not protection" add tokens beyond the bare rule. This matches 'efficient; minor instances of over-explanation that could be trimmed' rather than the every-token-earns-its-place anchor at 5. | 4 / 5 |
Actionability | Concrete artifacts are named throughout — `trace_timeline` / `trace_summary`, `omc session friction report`, `.omc/logs/`, `.omc/notepads/*/issues.md`, `.omc/refit/pending-proposals.md`, `docs/standards/`, `.mcp.json` — plus a concrete tool call (Skill tool with `agent-doc-discipline`). Minor gaps keep it below fully-executable: the trace instruments' invocation form is unstated, and 'pass its verification checklist' gestures at content it does not show; an example finding line would make the output format unambiguous. | 4 / 5 |
Workflow Clarity | The process is explicitly sequenced with validation checkpoints: 'Read before asking the user anything' for the survey, 'Stop for user approval' as an explicit gate, and a full feedback loop for landed checks — 'run it clean once, then demonstrate it failing on a deliberately introduced violation, then revert the violation' with a bail-out rule ('A check that cannot be shown failing goes back to the proposal'). Headless mode re-states the same contract with its boundary. This matches the anchor 'clear sequence with explicit validation steps; feedback loops for error recovery'. | 5 / 5 |
Progressive Disclosure | The skill has no references/, scripts/, or assets/ directories, and the body (~49 lines) is under 50 lines with no need for external files — the rubric's simple-skill exception applies. It is organized into five clearly-headed sections (Evidence first, Four fix-owner surfaces, Proposal and landing, Headless refit, Output), and the only external pointers (`agent-doc-discipline`, `configure-notifications`) are one level deep and clearly signaled. | 5 / 5 |
Total | 18 / 20 Passed |