Content
67%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-contracted, genuinely actionable instruction skill: concrete output shapes, exact memory paths, source labels, explicit stop conditions, and a done-when checklist. Its main costs are token redundancy — the claims-routing boilerplate and cross-skill links are repeated many times over — and heavy dependence on external files that are not present in this bundle for the details it delegates (propose-request format, handoff-summary shape).
Suggestions
State the claims-routing protocol once (e.g., in a short 'Claims routing' note or a referenced protocol file) and refer to it as 'per the claims-routing note' elsewhere — the phrase "via an authorized `operation: propose` request to `registry-events.py`" currently appears six times.
De-duplicate the cross-skill references: strategic-narrative-designer and audience-mapper are each linked 4–5 times across the Scope guard, Skill Contract, Reference Materials, and Next Best Skill; one Reference Materials entry plus the Next Best Skill section would cover it.
Inline the minimal concrete detail for the delegated operations — one example of an `operation: propose` request body for `registry-events.py` and the handoff-summary shape (or a short template) — so the skill is executable without the external files that are absent from this bundle.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is information-dense and never explains concepts Claude already knows (the JTBD forces are framed in one compact parenthetical, not tutored), but it is noticeably padded by repetition: the boilerplate "via an authorized `operation: propose` request to `registry-events.py`" appears verbatim six times, the opening paragraph largely restates the frontmatter description, and strategic-narrative-designer/audience-mapper links are each repeated four-to-five times across Scope guard, Skill Contract, Reference Materials, and Next Best Skill. This sits at the 3 anchor ('could be tightened') — the duplicated boilerplate and cross-references are more than the 4 anchor's 'minor instances', but not the pervasive padding of the 2 anchor. | 3 / 5 |
Actionability | Concrete, executable guidance throughout for an instruction-only skill: exact table columns ("objection · frequency-source · reframe candidate"), exact memory paths (`memory/narrative/audience-belief-mapper/YYYY-MM-DD-<topic>.md`), the Measured/User-provided/Estimated labeling scheme, and three copy-paste Quick Start invocations. It falls short of the 5 anchor because key executable details are delegated to files not shown — the propose-request format for `registry-events.py` and the handoff-summary shape ("Emit the standard shape from skill-contract.md") are invoked but never specified, leaving gaps a 5 would close. | 4 / 5 |
Workflow Clarity | The seven numbered Instructions are clearly sequenced (anchor → extract beliefs → objections table → four forces → verbatim language → claims sweep → assemble) with real checkpoints: step 1's stop-and-route gate ("if no persona evidence exists, stop and route to audience-mapper") and an explicit "Done when" checklist. It misses the 5 anchor because there are no per-step verification or fix-and-retry loops (e.g., how to verify a frequency count against the notes before it enters the table); it exceeds the 3 anchor because the gate and done-when checklist are explicit rather than implicit. | 4 / 5 |
Progressive Disclosure | Good structure in a single well-sectioned file, with a dedicated Reference Materials section that signals each external file's purpose (one level deep, clearly labeled). However, the bundle ships no reference/script/asset files — every link (`../../../references/tale-benchmark.md`, `../../../CONNECTORS.md`, `scripts/connectors/tavily.py`, sibling SKILL.mds) points outside the bundle and resolves to nothing here, and some inline material (the repeated claims-routing protocol) reads as content that belongs in one referenced file. This fits the 4 anchor ('references mostly clear; minor organization gaps') better than the 5 anchor's cleanly split, self-contained bundle. | 4 / 5 |
Total | 15 / 20 Passed |