Content
77%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, domain-rich instruction skill with a clear five-step workflow, explicit validation, and a user-escalation feedback loop. The main weakness is progressive disclosure: the bundle's script and audit reference are never referenced from the body, and the annually-changing conference table and repeated hard rules could be consolidated or moved to a reference file.
Suggestions
Add a short section pointing to the existing bundle files — e.g., 'Use scripts/main.py to count words/characters and detect section headings' and 'For verification steps see references/audit-reference.md' — so SKILL.md navigates to them; they are currently orphaned and discoverable only by listing the bundle.
Consolidate the 'never fabricate' rule, currently stated in four places (Out-of-scope, Compression Rules item 5, Final Check, Hard Rules), into one Hard Rules entry that the other sections reference by name.
Add a compact worked example — an original 400-word abstract adapted to a 250-word structured ASCO/ASGCT format with its word count note — to make the compression rules and output format concrete.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with domain knowledge Claude doesn't reliably have (per-conference limits, the 15/25/40/20 word-proportion targets, abbreviation norms), but the 'never fabricate' rule is repeated in four places (Out-of-scope, Compression Rules, Final Check, Hard Rules) and the line-31 blockquote self-description ('Conference Abstract Adaptor reformats and condenses...') is filler — minor instances that could be trimmed, fitting anchor 4 rather than the every-token-earns-its-place anchor 5. | 4 / 5 |
Actionability | Guidance is concrete and executable for an instruction-only skill: numeric section proportions, an ordered cut priority (preserve/compress/cut-first/never-cut), an exact output format ('Word count: X / [limit]'), and a copy-paste conflict-resolution message. It stops short of anchor 5 because there is no worked before/after example covering a common case (e.g., a full abstract adapted to a 250-word structured one). | 4 / 5 |
Workflow Clarity | The five-step sequence (Assess → Adapt → Compress → Count & Verify → Final Check) has explicit validation checkpoints (word/character count verification, a final checklist) and a genuine feedback loop: if required elements still exceed the limit after aggressive cutting, it escalates to the user with a specific message instead of silently cutting. This matches the anchor-5 pattern; the destructive/batch cap does not apply. | 5 / 5 |
Progressive Disclosure | The body itself is well-sectioned, but both bundle files are completely un-signaled: scripts/main.py and references/audit-reference.md exist yet are never mentioned or linked from SKILL.md, leaving them orphaned with no navigation. Per the guideline to score against the actual bundle structure, absent references are more than a 'minor organization gap' (anchor 4), fitting 'references present but not clearly signaled' (anchor 3). | 3 / 5 |
Total | 16 / 20 Passed |