Content
90%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 lean, highly actionable skill body: exact thresholds, runnable commands, and a well-guarded calendar procedure with approval and fallback checkpoints. The two weaker spots are the absence of a single ordered end-to-end flow and a Papers section that references a nonexistent directory, plus an orphaned library.md.
Suggestions
Fix the Papers section: `references/papers/` and `INDEX.md` do not exist in this bundle — point instead to the 'Full-text reading notes' section of `references/evidence-map.md` and to `references/library.md`, which already explain that this skill reuses the other skills' papers.
Reference `references/library.md` somewhere in the body (e.g. from the Papers section) so the paper library is discoverable instead of orphaned.
Add one short ordered flow at the top (run `next_session.py` → apply the decision table → ask/apply the calendar update → `--record` after an accepted progression → emit the handoff block) so the end-to-end sequence does not have to be assembled from separate sections.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and operational throughout — a decision table with exact multipliers ("shorter (x0.7, or x0.85 for long sessions)"), run commands, a numbered calendar procedure, and a handoff template. It explains nothing Claude already knows and contains no padded sections, matching the 'every token earns its place' anchor. | 5 / 5 |
Actionability | Fully executable guidance: copy-paste commands with all flags ("--days 3", "--write-fit", "--record"), exact numeric thresholds ("at least 85 % of the last 14 days of plan done, this week up 10 % or less, at least 7 days since the last progression"), concrete payload fields ("notificationLevel: NONE"), and a filled-in handoff block format. It is not a 4 because the commands and thresholds leave no gaps for the common cases. | 5 / 5 |
Workflow Clarity | The calendar sub-workflow is numbered with real checkpoints ("ask first" on approval, "check the user's own events for clashes", "If the connector is not authorised, give the change as text", re-run with --record after acceptance). It falls short of 5 because the end-to-end sequence (run script → decide → apply calendar → record → handoff) is distributed across topic sections rather than given as one ordered flow, and there is no re-validation step after an update is applied. | 4 / 5 |
Progressive Disclosure | Good structure with a clearly signaled, one-level-deep reference ("rules and evidence: `references/evidence-map.md`", which exists and holds the bulk detail). Not a 5 because the Papers section points to `references/papers/` and `INDEX.md`, which do not exist in the bundle, and `references/library.md` is never referenced from the body — minor organization gaps rather than buried or inline-bloated content, so above anchor 3. | 4 / 5 |
Total | 18 / 20 Passed |