Content
66%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 skill's process engineering is excellent: a legible phase pipeline, explicit user checkpoints, feedback loops, and a per-cycle checklist make the workflow hard to get wrong. Its weaknesses are density and repetition — the same core contracts are restated several times in heavy protocol vocabulary, inflating token cost — and the absence of any template or worked example for the WorkUnitRecord and navigation block, the two artifacts the whole skill exists to produce.
Suggestions
State the externalize-then-point contract once in Phase 7 and reference it from Rules and anti-patterns instead of restating it in full; this alone would remove several paragraphs.
Add a short template or worked example of a WorkUnitRecord and a Phase 7 navigation block so the skill's two output artifacts have a concrete shape, not just a field list.
Move the Types table and Rules list into a reference file (e.g. references/contract.md) and keep a lean phase-by-phase core in SKILL.md to reduce inline token load.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is noticeably padded for a ~240-line protocol: the externalize-then-point contract is restated at least five times (Phase 7 twice, Rule 10, Rule 14, the 'Re-authored handoff' anti-pattern, and the Phase 7 checklist item), and sentences like 'the issue is the record a unit is worked from while in flight; the ledger of what landed is the commit history, and the issue is not it' add rhetorical scaffolding without operational content. This matches anchor 2 (several unnecessary or padded sections). It is not 1 because it never explains concepts Claude already knows — everything is project-specific protocol — and not 3 because the redundancy is systemic across whole paragraphs rather than isolated tightenings. | 2 / 5 |
Actionability | Mostly concrete guidance: an exact `gh issue list` command with JSON fields, a load-axis table for classification, explicit split conditions, a deterministic anchor-issue tiebreak (earliest-created included issue), and a per-cycle checklist — this is executable process guidance for an instruction-only skill. It falls short of 5 because the two central artifacts, the WorkUnitRecord and the Phase 7 navigation block, are specified as field lists in prose but never shown as a template or example, leaving the exact output shape to inference. It is clearly above 3 because the guidance present is specific and directly executable, not high-level hints. | 4 / 5 |
Workflow Clarity | The pipeline diagram plus Phase 0-7 gives an unambiguous sequence; user-confirmation checkpoints are explicit at Phase 2 (surface grouping map, present 2-3 alternatives), Phase 6 (route choice gates Phase 7), and Phase 0 (checkpointed batches for large audits); the re-triage path is a defined feedback loop back to the relevant earlier phase; and the operational checklist verifies every phase per cycle. This matches anchor 5 (clear sequence, explicit validation steps, feedback loops, checklist). It is not 4 because no checkpoint is merely implicit — even the batch-intake path specifies surfacing progress between batches. | 5 / 5 |
Progressive Disclosure | The body has good structure: clear phase headers, a types table, rules, anti-patterns, and a checklist, with no nested or buried references — and no bundle files exist to misdirect, so the scoring reflects the on-file organization. It is not 5 because ~240 lines of dense protocol inline is more than an overview; the Types table and Rules sections are self-contained material that could live in reference files, and the guideline reserving a 5 for well-organized-sections applies to skills under 50 lines. It is comfortably above 3 since everything is clearly signaled and navigable. | 4 / 5 |
Total | 15 / 20 Passed |