Content
42%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 contains a genuinely useful domain core (input formats, verified CLI examples, output schema, scoring algorithm details) but it is buried in generic template boilerplate and self-referential filler ('See ## Usage above for related details'). Several documented items fail verification against the actual bundle (nonexistent flag, requirements.txt, and function names), and reference material is weakly signaled while inlining content that belongs in references/.
Suggestions
Fix the accuracy mismatches with the actual script: remove the nonexistent --wildtype-peptides example and requirements.txt reference, correct the 'cd 20260318/...' path, and update the 'Key functions' list to the real function names (_generate_variant_peptides, _predict_candidate, _assess_immunogenicity, _rank_candidates).
Cut the generic template sections (Risk Assessment, Security Checklist, Evaluation Criteria, Lifecycle Status, Response Template, Input Validation, Output Requirements) or collapse them to a few lines; they add tokens without task-specific value.
Name the actual reference files in the body ('See references/README.md for scoring thresholds, HLA motif data, and validation workflows; references/example_mutations.csv for a ready-to-run example input') and move the HLA support list and scoring formulas into references/ to slim SKILL.md into a true overview.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~450-line body contains substantial padding unrelated to the task: generic template boilerplate ('Scope-focused workflow aligned to...', 'Structured execution path designed to keep outputs consistent and reviewable'), plus Risk Assessment, Security Checklist, Evaluation Criteria, Lifecycle Status, Output Requirements, Response Template, and Input Validation sections that are skill-template filler rather than neoantigen-specific instruction. This fits 'Noticeably verbose; several unnecessary explanations or padded sections'; it is above level 1 because a meaningful core (input formats, CLI usage, output schema, scoring algorithms) is genuinely useful domain content. | 2 / 5 |
Actionability | There is real executable guidance — the CLI examples match the actual argparse flags in scripts/main.py (--hla, --vcf, --mutations, --variant-peptides, --peptide-length, --rank-cutoff, --output) and the Python API example matches the actual NeoantigenPredictor class. However, key documented items fail against the bundle: the third CLI example uses a --wildtype-peptides flag that does not exist in the script, 'pip install -r requirements.txt' references a nonexistent requirements.txt, the Example Usage 'cd 20260318/scientific-skills/...' path does not exist, and four of five 'Key functions' (extract_variant_peptides, predict_mhc_binding, calculate_foreignness, score_immunogenicity) do not exist in main.py. This fits 'Some concrete guidance but incomplete... missing key details', below level 4 whose 'minor gaps' standard these mismatches exceed. | 3 / 5 |
Workflow Clarity | A sequence exists (Workflow's 5 steps, the Example run plan, and a Quick Check that validates the script parses via py_compile before running), but the Workflow steps are abstract template language ('Confirm the user objective...', 'switch to the fallback path') rather than a concrete command sequence for the prediction task, and the connection between the workflow steps and the actual CLI invocations is left implicit. This matches 'Steps listed but validation gaps; sequence present but checkpoints missing or implicit', not level 4 which requires concrete sequenced commands with most checkpoints explicit. | 3 / 5 |
Progressive Disclosure | A real references/ bundle exists (README.md with algorithm papers, HLA motif data, amino acid tables, best practices; example_mutations.csv) at one level deep, and the body points to it. But the pointer is weakly signaled — 'Reference guidance: references/ contains supporting rules, prompts, or checklists' and 'See references/ directory' with generic bullets rather than naming the actual files — and large reference-type content (the full HLA support list, scoring algorithm formulas, output JSON schema) is inlined in SKILL.md instead of split out. This fits 'Some structure but could be better organized; references present but not clearly signaled; content that should be separate is inline'. | 3 / 5 |
Total | 11 / 20 Passed |