Content
92%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.
An exemplary operational skill body: executable commands throughout, a clear phased workflow with an explicit verification step, and clean one-level-deep progressive disclosure verified against the actual bundle. The main improvement opportunity is removing the redundancy between the Activation section and PHASE 0 steps 2-3.
Suggestions
Trim the Activation section: it duplicates PHASE 0 steps 2-3 (multi-query expansion, --include-subagents, explorer-style lanes, narrowing by platform/cwd/model/time) — either collapse it into a one-line pointer to PHASE 0 or merge the unique memory-recall phrasings into step 2.
Move the OpenCode implementation paragraph in Filters (opencode db path, SQLite queries, CLI fallback, messages/parts scan avoidance) into references/opencode.md, keeping only the operational implication (e.g., 'OpenCode reads its DB first; pass --root only for nonstandard stores').
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and operational — tables, ready-to-run commands, and terse notes like "prints JSON for stdout and `jq`" with no basic-concept explanations. It is not a 5 because the Activation section restates PHASE 0 steps 2-3 (query expansion, explorer-style lanes, --include-subagents) almost verbatim, and the OpenCode SQLite/DB paragraph is implementation detail that could be trimmed to a pointer. | 4 / 5 |
Actionability | Guidance is fully executable: a copy-paste bash block covering list/find/read with real flag combinations, a router table mapping each platform to an exact reference path, `detail_hint` commands for the next step, and a troubleshooting table with exact flags and paths. This matches the 5 anchor (copy-paste ready, covers common cases); a 4 would have minor gaps in the concrete examples. | 5 / 5 |
Workflow Clarity | PHASE 0 gives a clear 5-step sequence — route by platform, expand queries and run the finder, split parallel lanes, read `detail_hint` details, verify against raw transcripts — with an explicit validation step ("the raw `path` remains the source of truth") and a troubleshooting section as feedback loops. The operations are read-only, so the destructive/batch validation cap does not apply; it exceeds the 4 anchor because checkpoints are explicit rather than implied. | 5 / 5 |
Progressive Disclosure | The body is a well-organized overview: a router table clearly signals one-level-deep references (references/codex.md, claude.md, senpi.md, opencode.md, all-platforms.md — all verified to exist, with no nested references inside them), the finder script is bundled, and platform storage detail is appropriately split into per-platform files. Navigation is unambiguous, matching the 5 anchor rather than the 4 anchor's 'minor organization gaps'. | 5 / 5 |
Total | 19 / 20 Passed |