Content
82%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.
The body is highly actionable, with seven complete executable SQL patterns, a coherent find-to-resume workflow, explicit pre-resume validation in the Examples, and protective Anti-Patterns. It is efficient and well organized, with only minor room for improvement: factoring out the repeated last_checkpoint subquery and adding a fallback path for when session matching or resume fails.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly lean: SQL-first patterns with one-line intros, and a Context section that stays skill-specific ("session_store" SQLite, read-only access via the sql tool) rather than explaining generic concepts. It falls short of anchor 5 because the correlated "last_checkpoint" subquery is repeated verbatim in patterns 1, 2, and 4, and the interruption-cause enumeration ("terminal crashes, network drops, machine restarts, or accidental window closes") could be trimmed; it is well above anchor 3 since there is no padding or teaching of things Claude already knows. | 4 / 5 |
Actionability | Every pattern ships complete, executable SQL — time-window filters ("datetime('now', '-24 hours')"), a concrete FTS5 MATCH string with synonym expansion ("'auth OR login OR token OR JWT'"), correlated subqueries, and a NOT IN automated-session filter — plus the bash resume command ("copilot --resume SESSION_ID") and a "gh issue list --label "status:in-progress"" cross-check. This matches anchor 5: copy-paste ready commands covering the common cases, with placeholders clearly marked. | 5 / 5 |
Workflow Clarity | The patterns form a logical find-to-resume sequence (find recent → filter automated noise → search → inspect full context → detect orphaned work → resume), and both Examples include an explicit pre-resume validation checkpoint ("Check its last checkpoint — was the code committed? Was the PR created?"), reinforced by Anti-Patterns ("Resuming sessions that completed successfully"). It matches anchor 4 rather than 5 because there are no feedback loops for error recovery (e.g., what to do when no session matches or a resume fails); the destructive/batch cap at 3 does not apply since the queries are read-only. | 4 / 5 |
Progressive Disclosure | The single file is well structured with clear sections (Context → seven numbered Patterns → Examples → Anti-Patterns) and no nested or buried references; no bundle files exist, so there are no broken or dead reference paths. It matches anchor 4 rather than 5 because at ~140 lines all content is inline with no one-level-deep reference split — several repeated query variants could live in a separate reference file to keep the overview leaner — and it exceeds the simple-skill under-50-line threshold that would otherwise allow a 5 on organization alone. | 4 / 5 |
Total | 17 / 20 Passed |