Content
71%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, information-dense workflow document with clear phase sequencing, concrete tool usage, and commendable anti-fabrication guardrails. Its main defect is that the entire reference layer it discloses to (six .md files holding code examples, the report template, checklists, and manufacturing detail) is absent from the bundle, so the progressive-disclosure structure is largely dangling.
Suggestions
Add the six referenced files (QUICK_START.md, WORKFLOW_DETAILS.md, REPORT_TEMPLATE.md, MANUFACTURING.md, EXAMPLES.md, CHECKLISTS.md) to the bundle, or remove the dangling pointers and inline the essential content — currently only scripts/developability.py resolves, so every 'See X for code examples' pointer dead-ends.
Tighten the Reagent Verification and Phase 7 codon-usage sections: reduce each tool to a one-line summary with its signature (e.g. `AntibodyRegistry_get_by_rrid(rrid="AB_2298772")`) and move the live-verification narratives to a reference file.
Add explicit error-recovery guidance at fragile steps, e.g. what to do when ESMFold pLDDT in CDR regions is low (fall back to SAbDab experimental structures) or when SOAP tools fail, to complete the feedback loop the Phase 0 verification starts.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and operational throughout — it gives raw Vernier positions, IMGT CDR ranges, scoring weights, and PTM motifs without explaining basics Claude already knows. Not 5: the Reagent Verification section (lines ~310–313) carries multi-line live-verification narratives (e.g. the SciCrunch paragraph recounting 12 synonyms and a proper_citation string) and the Phase 7 codon-usage step is a run-on paragraph that could be trimmed; not 3: the padding is localized, not pervasive. | 4 / 5 |
Actionability | Gives executable specifics: `scripts/developability.py --seq <VH/VL>` with described outputs, `CodonUsage_get_optimal_codons(taxid=...)` with the exact taxid resolution caveat, worked tool calls like `AntibodyRegistry_search(query="GFAP")`, numeric tier cutoffs, and explicit output filenames. Not 5: the per-phase code examples are all deferred to `WORKFLOW_DETAILS.md`, which does not exist in the bundle, so the promised copy-paste layer is missing; not 3: inline guidance is concrete and mostly executable as written. | 4 / 5 |
Workflow Clarity | The 8-phase pipeline is clearly sequenced (overview diagram plus per-phase Goal/Key steps/Output), with Phase 0 tool verification and explicit anti-fabrication checkpoints ("Do not fabricate Tm/titer numbers", "Do not invent KD ranges", "report only if actually run"). Not 5: there are no explicit error-recovery feedback loops (e.g. what to do when pLDDT is low or a SOAP call fails), and the referenced `CHECKLISTS.md` validation material is absent from the bundle. | 4 / 5 |
Progressive Disclosure | Scored against the actual bundle: the Reference Files table and per-phase "See X" pointers are well-signaled and one level deep, but 6 of the 7 referenced paths (`QUICK_START.md`, `WORKFLOW_DETAILS.md`, `REPORT_TEMPLATE.md`, `MANUFACTURING.md`, `EXAMPLES.md`, `CHECKLISTS.md`) do not exist — only `scripts/developability.py` resolves. Navigation leads nowhere for the detail layer, leaving SKILL.md effectively monolithic with dangling references. Not 2: the structure and signaling themselves are good, not buried or inlined; not 4: 'most content appropriately placed' fails when the bulk of the detail content is missing. | 3 / 5 |
Total | 15 / 20 Passed |