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 dense, highly actionable body with an exemplary prescriptive workflow, concrete commands, expected outputs, and an unusually strong Gotchas section of domain-specific failure modes. Its weaknesses are repetition of the same cautionary examples across sections and a broken progressive-disclosure structure: the repeatedly-cited GUIDE.md (plus examples/ and demo/) is missing from the bundle.
Suggestions
Ship GUIDE.md in the bundle (or inline the ten modes and the step-by-step walkthrough) — the body directs agents to read it three times, but the file does not exist; likewise verify examples/ and demo/ are present as referenced.
Consolidate the MAPQ-0/alt-contig material (Why This Exists, and two Gotchas entries) into one place, and merge the two DRAGEN chr4 caller-disagreement examples — the same story is told three and two times respectively.
Split the ~400-word Core Capabilities #8 summary paragraph into shorter bullets or move its curated-calls rules into a reference file, and drop the Trigger section that duplicates the frontmatter trigger_keywords.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly skill-specific and free of explanations of known concepts, but noticeably repetitious: the GRCh38 alt-contig MAPQ-0 story is told in full three times ("Why This Exists", "Check how the BAMs were aligned", "Losses called from MAPQ-filtered depth"), the DRAGEN chr4 repeat example appears twice ("Caller counts differ", "Caller disagreement is a question"), the Trigger section duplicates the frontmatter trigger_keywords, and Core Capabilities #8 is a ~400-word run-on paragraph. Not 4: the redundancy spans whole sections, beyond "minor instances that could be trimmed". | 3 / 5 |
Actionability | Copy-paste ready commands throughout: four executable CLI invocations in CLI Reference, a demo command with expected output ("7 captioned screenshots in about 10 seconds"), a diagnostic command ("samtools view -H tumor.bam | grep '^@PG'"), a full example report table, and concrete output-structure trees. Not 4: the common cases (single check, whole project, demo, cluster without IGV) are each covered by a specific runnable command. | 5 / 5 |
Workflow Clarity | The seven-step Workflow is clearly sequenced with prescriptive/flexible labels, an explicit validation gate in step 1 ("Stop with a clear message otherwise"), render verification with a retake loop in step 6, and run-level checks referenced from methodology.md. Not 5: the ten Modes section defers its prompts and commands entirely to GUIDE.md, which is not in the bundle, so those sub-workflows cannot actually be followed; "Agents: read GUIDE.md before running the skill" points nowhere. | 4 / 5 |
Progressive Disclosure | Good in principle — SKILL.md as overview, one-level-deep references to references/methodology.md and references/cli-options.md, both real files, clearly signaled with link syntax and read-when guidance — but scored against the actual bundle: GUIDE.md is referenced three times as required reading and does not exist, and examples/ and demo/ are also referenced but absent. Not 4: the most heavily promoted reference is dangling, which breaks navigation for the modes and the whole-project walkthrough; not 2: the two references that do exist are real, substantive, and well-signaled. | 3 / 5 |
Total | 15 / 20 Passed |