Content
70%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 excels at actionability and workflow clarity — commands are copy-paste executable, the seven-step pipeline is strictly ordered with structured error codes, fix hints, and validation at every checkpoint. Its weaknesses are token efficiency (heavy duplication and rationale-laden gotcha paragraphs across ~425 lines) and progressive disclosure (everything, including a 70-row flag reference, is inlined in a single monolithic file).
Suggestions
Move the full CLI flag table, the nf-core 4.1.0 compatibility policy, and the detailed gotchas into one-level-deep reference files (e.g. references/cli.md, references/compatibility.md, references/gotchas.md), keeping SKILL.md as a lean overview with pointers.
De-duplicate policy explanations that currently appear in multiple sections — igenomes_ignore (Algorithm, flag table, gotcha, Safety), remote inputs (Path Policy, Safety), and conda/CellRanger restrictions (Algorithm, gotcha, flags) — stating each rule once in its canonical section.
Tighten gotcha entries to the actionable rule plus the error code, trimming internal implementation rationale (zero-padding comparisons, subprocess timeout constants, sibling-wrapper parity notes) that an executing agent does not need.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is noticeably verbose: policy explanations are duplicated across sections (igenomes_ignore appears in Algorithm step 4, the flag table, a gotcha, and Safety; remote-input policy in both the Path Policy section and Safety; conda/CellRanger rules in three places), and several multi-hundred-word gotcha paragraphs carry internal implementation rationale (version-tuple zero-padding rationale, subprocess timeout constants, sibling-wrapper parity notes, upstream nf-schema warning forensics) that a executing agent does not need. While nothing explains generic concepts Claude already knows, several padded sections could be cut by half or more, matching anchor 2 rather than 3. | 2 / 5 |
Actionability | The CLI Reference provides fully executable, copy-paste-ready commands covering the common cases (standard run, --check preflight, --demo, runner invocation, fasta+gtf, prebuilt index, RNA velocity, CMO multiplexing, air-gapped iGenomes mirror), plus a complete flag table with types, defaults, and preset-conditional requirements. This matches the anchor for fully executable guidance covering common cases. | 5 / 5 |
Workflow Clarity | The workflow is a "strictly ordered 7-step pipeline" where "a failure at any step raises a structured SkillError with an error_code and a fix hint; no subsequent step runs" — explicit validation checkpoints (preflight gates before Nextflow, post-run required-output validation, checksum-verified --resume) with error-code/fix feedback loops for recovery, matching the top anchor. | 5 / 5 |
Progressive Disclosure | The document is well-sectioned with clear headers, but it is monolithic: the ~70-row flag table, the compatibility policy, and roughly 150 lines of gotchas are all inlined content that clearly belongs in separate reference files, and no bundle files (references/, scripts/, assets/) exist. This matches anchor 3 (good structure, but content that should be separate is inline) rather than 4, since no content is split out at all. | 3 / 5 |
Total | 15 / 20 Passed |