Content
85%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 highly actionable, well-sequenced orchestrator skill whose guidance is almost entirely non-obvious and concrete, with proper propose-first and error-recovery checkpoints. Its main weakness is structure: the long branched Session Mode is inlined rather than split into a reference file, and a few explanatory asides could be trimmed.
Suggestions
Move the Session Mode details (discovery probe code, message-parsing rules, per-session JSON schema, aggregation logic) into a references/session-mode.md and keep a short gated pointer in SKILL.md, since plain /reflect skips that branch.
Trim the cache-logic 4-step list and the confidence-score bands to one line each — the behavior is obvious from a single sentence ("load cached <session-id>.json if present, else analyze and save").
Delete the redundant goal restatement ("The goal is to identify real repeated friction...") which repeats the opening paragraph.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and almost entirely original, non-obvious guidance (OpenCode v1/v2 schema detection, cache paths, report templates), but a few parts could be trimmed: the obvious 4-step cache-logic list, the confidence-score bands, and the intro line restating the goal. Minor over-explanation, not enough to drop to anchor 3. | 4 / 5 |
Actionability | Fully executable guidance: copy-paste-ready SQLite probe scripts with a justified temp-file pattern, exact message-parsing rules per row type, literal cache directory, per-session JSON output schema, and complete report/error message templates covering common cases. | 5 / 5 |
Workflow Clarity | The core workflow is a clearly sequenced 5-step process with an explicit "Propose Before Changing" validation checkpoint, ordered evidence sources, confidence thresholds for skipping weak candidates, and an Error Handling section with a skip-continue-report recovery loop for the batch session analysis. | 5 / 5 |
Progressive Disclosure | Sections are well-organized and easy to navigate, but the ~150-line branched Session Mode (probe code, parsing rules, JSON schema, aggregation logic) is inlined in a 365-line single-file SKILL.md with no bundle files at all — content that clearly belongs in a one-level-deep reference file is inline, matching anchor 3 rather than 4. | 3 / 5 |
Total | 17 / 20 Passed |