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.
A tight, disciplined procedure: lean imperative text, a fully sequenced 10-step workflow with a user-confirmation gate and stop conditions, and appropriate deferral of detail to the parent llm-wiki skill. The main weaknesses are the unresolved `<plugin>` path placeholder and the absence of a post-write verification step for what is a batch, 5-15-page write operation.
Suggestions
Resolve the `<plugin>` placeholder once (e.g., name the actual install path of the engineering/llm-wiki plugin, or state how to discover it) so the three script commands are fully copy-paste executable.
Add a post-write verification step for the batch update — e.g., re-run update_index.py and confirm every touched page got its `updated:` bump and `sources:` increment before reporting back — to close the feedback loop on a 5-15-page write operation.
Inline one minimal example of a source-summary page (frontmatter + one body section) or a `> ⚠️ Contradiction:` callout so the write steps can be executed without consulting the external template.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and imperative throughout — "raw/ is immutable. Never edit files there. Read only.", "Minimum 5 file touches per ingest.", "Update `updated:` frontmatter on every page you touch" — with zero padding and no explanation of concepts Claude already knows. The only near-motivational line ("Don't skimp — the wiki's value comes from cross-references") carries a real behavioral instruction. Fits the 'every token earns its place' anchor; a 4 would require over-explanation to trim, which is absent. | 5 / 5 |
Actionability | Mostly executable: concrete commands ("python <plugin>/scripts/ingest_source.py --vault . --source <path> --json", "update_index.py", "append_log.py --op ingest --title ... --detail ..."), concrete paths ("wiki/sources/<slug>.md"), and concrete frontmatter fields ("title, category: source, summary, source_path, ingested, updated"). Falls short of the 5 anchor because the `<plugin>` base path is never resolved anywhere in the body, and the summary-page template and contradiction-callout format are deferred without an inline example — 'minor gaps' per the 4 anchor, not the vague/pseudocode levels below. | 4 / 5 |
Workflow Clarity | A clear 10-step sequence with an explicit pre-write checkpoint ("Wait for the user to confirm or redirect before writing"), merge detection via the `--json` brief, and a Red flags stop-list (outside raw/, exact duplicates, >5 contradictions). Because this is a batch operation touching 5-15 pages, the feedback-loop guideline applies, and post-write verification is missing — no step re-checks the regenerated index or the touched pages after the batch write. That is a 'minor validation gap' (4 anchor) rather than the absent-validation case (capped at 3), since pre-write validation and user gating are explicitly present. | 4 / 5 |
Progressive Disclosure | Good structure: the body is a well-sectioned overview (Role, Inputs, Workflow, Rules, Red flags) that defers detail to a clearly signaled, one-level-deep reference — "Follow `engineering/llm-wiki/skills/llm-wiki/references/ingest-workflow.md` ... Summary:" — with templates referenced from the llm-wiki skill, and no nested references. Not a 5 because the skill bundles nothing locally while all referenced scripts/templates live behind the undefined `<plugin>` path, a minor organization/navigation gap; not a 3 because nothing that belongs in a separate file is inlined and references are prominent, not buried. | 4 / 5 |
Total | 17 / 20 Passed |