Content
78%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 well-structured, self-contained analytical add-on: concrete Smell/Move guidance per pattern, a deliberate approach table, explicit ordering rules, and a quality-gates checklist that functions as validation. The main gaps are the absence of a worked output example and a feedback loop for degraded inputs, plus minor verbosity in the input-coverage caveat.
Suggestions
Add a short worked example of the default output — e.g. a 2–3 row sample of the initiatives × verified stage × clarity × pattern-flags table with one cross-cutting finding and a stated risk-balance split.
Merge the thin-input caveat in 'How to Run Me' into a single sentence; the two consecutive sentences both say that findings degrade when per-initiative evidence is unreachable.
Add a feedback loop for the common failure mode: what to do when base sniff-test reads are missing or stale (e.g. fall back to a quick single-item sniff before consuming the reads, then re-check the quality gates).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and earns its tokens — each pattern gets a compact 'Smell:'/'Move:' pair and the scan approaches sit in a table — with only minor trimmable spots, e.g. 'With thin input I'll work from whatever board snapshot you give me and flag low coverage — board-level findings degrade fast when the per-initiative evidence isn't reachable' largely repeats the preceding sentence. It is not a 5 because a few sentences could be tightened, and it is clearly above a 3 since nothing explains concepts Claude already knows. | 4 / 5 |
Actionability | Guidance is concrete and executable for an analytical skill — each pattern names detection cues and moves ('fuzzy boundaries between a parent initiative and its "refinement"'; 're-cut to the riskiest assumption, or graduate it honestly to a committed Execute/Build'), and output modes are specified. It is not a 5 because there is no worked example of the default output (e.g. a sample initiatives × stage × flags table or a concrete risk-balance split instance). | 4 / 5 |
Workflow Clarity | The sequence is clear (prerequisite base reads → point at the system of record → state approach and altitude → scan patterns → choose output mode) with explicit checkpoints — 'A stale-status sweep comes before any Explore/Discovery-vs-Build&Deliver split, or the split is fiction' — and a Quality Gates checklist. It is not a 5 because there is no error-recovery feedback loop (e.g. what to do when base reads are missing), though the operation is read-only analysis so no destructive-operation cap applies. | 4 / 5 |
Progressive Disclosure | The skill is a single self-contained file with well-organized sections and one clearly signaled, one-level-deep reference ('Base skill: `sniff-test.md` and its references'). No bundle files exist and each pattern is only 1–3 lines, so nothing belongs in a separate file; navigation via section headers is easy. | 5 / 5 |
Total | 17 / 20 Passed |