Content
63%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 delivers highly actionable, read-only operational guidance with concrete commands, exact thresholds, formulas, fallbacks, and output templates — no known-concept padding. Its weaknesses are length-driven: duplicated template content, an oversized manifest example, and a monolithic three-mode structure that belongs in separate reference files rather than a single 30KB SKILL.md with no bundle.
Suggestions
Move Insights Mode and Equilibrium Mode into separate reference files (e.g., references/insights.md, references/equilibrium.md) and keep SKILL.md as a status/delta overview with clearly signaled one-level-deep links — the three-mode inline structure is the main progressive-disclosure gap.
Show the expected entry format of hot.md and log.md lines (or a one-line example each) so the synthesis/lint 'recent run' lookups in Step 4a are executable rather than inferred, and tighten the wiki-dedup audit invocation into a concrete command.
Trim token cost: cut the manifest JSON example to 2 representative entries, and render the Tier Suggestions formatting block once (either in section 8 or in the output template, not both).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~530-line body is dense with genuinely non-obvious operational detail (manifest schema, delta classification table, thresholds, formulas), so it is not padded with known-concept explanation — but it could be tightened: the ~40-line full manifest JSON example, the Tier Suggestions block rendered twice (section 8 and the output template), and long output templates all inflate tokens. This sits between 'efficient with minor trims' (4) and 'mostly efficient but could be tightened' (3), landing at 3 given the sheer volume of duplicated template content. | 3 / 5 |
Actionability | Copy-paste-ready commands ("obsidian-wiki graph-analyse \"$OBSIDIAN_VAULT_PATH\" --pretty --snapshot", "${QMD_CLI:-qmd} update"), exact thresholds (≥90 days, ≥5 incoming links, cohesion < 0.15, token warn default 100000), scoring formulas (+3/+2 re-ranking bonuses), and concrete output templates make this mostly executable. Not 5 because a few steps remain under-specified: the expected formats of hot.md/log.md entries for synthesis/lint lookups are never shown, and the wiki-dedup 'audit mode (no --merge, no --auto)' invocation is described rather than given as a command. | 4 / 5 |
Workflow Clarity | The sequence is clearly numbered (Before You Start → Manifest → Step 1 scan → Step 2 delta → Step 3 report → Step 3b footprint → Step 4 ranked next-steps) with mode triggers, skip conditions (vaults < 20 pages), and explicit fallbacks (manual wikilink parsing if obsidian-wiki absent; 'omit that row and say which player is missing rather than guessing'). The destructive/batch cap does not apply — the skill is read-only except _insights.md, and QMD failure handling ('do not roll back; report the QMD status separately') is an explicit checkpoint. Not 5 because Step 3b's 'After building the status summary, compute the token footprint' is ambiguous against the template that already embeds the footprint section, and validation of the delta table itself (e.g., what to do when hash and mtime disagree in an old entry) is only implicit. | 4 / 5 |
Progressive Disclosure | Cross-references are clearly signaled and one level deep ("See `llm-wiki/SKILL.md` → `.manifest.json`", "`scripts/manifest.py migrate <vault>`", WRITING.md), and in-file section structure is good — but no bundle files exist (no references/, scripts/, assets/), so three full modes (~140 lines of Insights Mode, ~80 of Equilibrium Mode) plus the complete report templates are inlined monolithically in a 30KB body. This matches 'some structure but content that should be separate is inline'; not 2 because in-file navigation via headers and mode triggers is genuinely clear. | 3 / 5 |
Total | 14 / 20 Passed |