Content
67%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, appropriately progressive meta-skill: dense imperative rules, a sequenced workflow with validation and iteration checkpoints, and disciplined offloading to a real references/ bundle. Its weaknesses are internal redundancy (the same size/trigger thresholds stated 2–3 times), a garbled line in the workflow, and three orphaned reference files that the References section never surfaces.
Suggestions
Deduplicate the size/threshold rules: state the >10-line code-block, ≤100-line, and ≥90% trigger-rate rules once (e.g., in Content Quality) and remove their repeats in Anti-Patterns and the Quality Checklist — the Quality Checklist currently lists the ≥90% trigger requirement twice.
Fix the broken step 2 text ("what it , when it trigger, expected output format?") so the Capture-intent step reads as a complete, grammatical instruction.
Add 'load when' entries to the References section for the three orphaned files — benchmark.md, rubric.md, and tessl-best-practices.md — so all bundle content is discoverable from SKILL.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and imperative with no AI-splaining of known concepts, but it repeats its own rules: the >10-line code-block rule appears in Content Quality, Anti-Patterns ("No Long Code Blocks"), and the Quality Checklist; the ≤100-line limit appears three times; and the ≥90% trigger-rate requirement appears twice within the same checklist ("Activation ≥ 90%" and "Trigger rate ≥90%"). Line 31 is also a broken sentence ("what it , when it trigger"), so the body clearly could be tightened — matching the 3 anchor rather than the minor-trims-only 4 anchor. | 3 / 5 |
Actionability | For an instruction-only skill the guidance is mostly executable: a snapshot command (`cp -r <skill-dir> <workspace>/skill-snapshot/`), hard thresholds (≥90% accuracy, ≤100 lines, >10-line extraction), and per-step pointers to real reference files. It stops short of 5 because several workflow steps ("spawn parallel subagents", "run trigger eval queries") defer their exact execution to references rather than including the command or a copy-paste example inline. | 4 / 5 |
Workflow Clarity | The 15-step workflow has a clear sequence with explicit validation checkpoints (Test with a with-skill/without-skill baseline, Evaluate on pass rate/tokens/time, Iterate on feedback, target ≥90% trigger accuracy) plus a pre-edit Snapshot safety step — near the 5 anchor. It lands at 4 because the new-skill and existing-skill tracks are merged into one continuous 1–15 numbering with awkward indentation, and validation mechanics are delegated to eval-workflow.md rather than stated as explicit validate/fix/retry steps at the point of use. | 4 / 5 |
Progressive Disclosure | The 83-line body stays under its own 100-line budget and closes with a well-signaled References section where each of 8 linked files carries a "load when <condition>" cue; all 8 paths resolve to real files one level deep. Scoring against the actual bundle, three reference files (benchmark.md, rubric.md, tessl-best-practices.md) are never linked or mentioned from the body — orphaned content that keeps navigation from the easy 5-anchor standard. | 4 / 5 |
Total | 15 / 20 Passed |