Content
63%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 well-structured, genuinely useful routing map: every skill is tied to a trigger situation, the main flow is an explicit decision tree with context-window hygiene rules, and guidance is given in concrete commands rather than abstractions. Weaknesses are moderate duplication between the main flow and Standalone sections, an unstated interaction model for the skill itself, and a prominently referenced PHASE-BOUNDARIES.md that is not present in the bundle.
Suggestions
State the skill's own interaction up front — e.g. that when invoked it should ask for (or infer) the user's situation, then recommend a route through this map — so the agent knows what to do, not just what the map contains.
Ship PHASE-BOUNDARIES.md alongside the skill (or inline the five-question tree), since the body leans on it for the one decision it calls the fuzziest.
Deduplicate the /grill-me vs /grill-with-docs comparison — explain the stateless/stateful split once and cross-reference it from the other section to trim roughly a paragraph of repeated reasoning.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The map is dense and consistently skill-focused, but the /grill-me vs /grill-with-docs distinction ("stateless", "leaves a paper trail", "strictly the better one") is explained in full twice — once in step 1 of the main flow and again under Standalone — and /wayfinder stacks redundant qualifiers ("it's slower and denser, so save it for exactly that, never a well-scoped feature"). Mostly efficient but with real duplication that could be tightened; not level 4 because the repetition is more than minor. | 3 / 5 |
Actionability | For an instruction-only router this is concrete: exact slash commands per situation, explicit branch questions ("can you settle every question in conversation?", "is this a multi-session build?"), file locations (".scratch/<feature>/issues/"), and a guardrail ("Before stating what a skill does or recommending a step be skipped, read that skill's SKILL.md"). It falls short of 5 because the skill's own behavior when invoked — how the routing conversation with the user proceeds — is never stated. | 4 / 5 |
Workflow Clarity | The main flow is a numbered 1–4 sequence with explicit yes/no branches and context checkpoints ("Keep steps 1–3 in one unbroken context window", compact at the nearest phase boundary near the smart zone). Validation loops are not applicable since this is routing, not a destructive or batch operation. It misses 5 because the phase-boundary decision is admitted as "the fuzziest decision in this whole map" and delegated to an external file rather than resolved inline. | 4 / 5 |
Progressive Disclosure | Sections are well organized (main flow, on-ramps, vocabulary, phase boundaries, standalone) and the single external reference is clearly signaled ("Read [PHASE-BOUNDARIES.md](PHASE-BOUNDARIES.md) for the ordered tree..."), but the bundle contains no such file — no references/, scripts/, or assets/ directories exist — so the one reference is dangling and navigation dead-ends. Good structure, but the broken/absent referenced file keeps this below 4. | 3 / 5 |
Total | 14 / 20 Passed |