Content
75%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 strong, highly actionable multi-step workflow: concrete executable commands for scanning, delta computation, extraction, distillation, manifest updates, and QMD refresh, with real error paths (QMD failure handling, dry-run migration, helper fallbacks) and good bundle structure. The main weakness is redundancy — the pre-extraction preference is repeated three times, desktop find commands twice, and JSONL format details are duplicated between the body and references/claude-data-format.md — which costs tokens without adding guidance.
Suggestions
De-duplicate the raw-JSONL format material: keep only the parsing essentials in Step 3 (filter to user/assistant, extract text blocks, use cwd) and move the full line/message schemas and event-type detail to references/claude-data-format.md, which already covers them.
State the pre-extracted-file preference once (in the Pre-extraction section) and have the Sampling Heuristic and Step 3 refer back to it, instead of restating the preference and the 50–200× reduction rationale in three places.
Consolidate the desktop-session find commands (currently given in full in both the Source 2 section and Step 1) into a single location and reference it from the survey step.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly purposeful procedural detail, but the pre-extracted-file preference is stated three times ("always prefer the extracted file" in Pre-extraction, "Prefer pre-extracted files" in the sampling heuristic, and the resolution-order block opening Step 3), the desktop-session find commands are duplicated between Source 2 and Step 1, and full JSONL message schemas are inlined that largely duplicate references/claude-data-format.md. That redundancy is more than the minor trimming of level 4, but nowhere near the pervasive padding of level 2 — most sections earn their tokens. | 3 / 5 |
Actionability | Nearly every section gives copy-paste-ready commands: "python3 …/scripts/manifest.py delta … --scan …", "python3 …/scripts/extract-jsonl.py --since …", "obsidian-wiki memory sync CLAUDE_HISTORY_INGEST projects=<…> … --takeaways …", "${QMD_CLI:-qmd} update", plus a concrete topic-routing table, frontmatter templates, and explicit fallbacks when a helper is unavailable ("if it's unavailable, apply the same resolution inline"). Common cases (first run, incremental, missing helper, QMD unset) are all covered. | 5 / 5 |
Workflow Clarity | Steps 1–6 are clearly sequenced with meaningful checkpoints for a batch operation: manifest delta classification (new/modified/unchanged), the desktop-session pre-check ("If that prints nothing, skip this entire section"), "migrate … --dry-run" before repair, sampled-vs-skipped reporting, and "If QMD refresh fails, do not roll back the vault changes; report the QMD status separately". Not level 5 because there is no validate→fix→retry loop over the written wiki pages themselves; not capped at 3 because verification steps are present rather than missing, exceeding level 4's single "verify output" checkpoint in breadth. | 4 / 5 |
Progressive Disclosure | Good structure with one clearly signaled, one-level-deep reference ("See references/claude-data-format.md for more details on the data structures" — verified to exist and to be flat, no nested references), and the SKILL.md reads as an overview-plus-procedure. Not level 5: roughly 100 lines of data-layout/format detail (directory trees, JSONL line schemas, audit-log field lists) are inlined even though the reference file covers much of the same ground, and the referenced scripts/manifest.py and scripts/extract-jsonl.py live in the parent repo rather than this bundle (acknowledged with inline fallbacks, but still an organization gap). | 4 / 5 |
Total | 16 / 20 Passed |