Conciseness

Lean and imperative throughout — exact paths, templates, and formats with almost no explanation of concepts Claude already knows. Not 5: the dated incident parenthetical ('(2026-09-06: ask-honest-gap withheld the answer...)') and some repetition of the no-training-data-fallback rule across sections could be trimmed; per the guidelines, the specific date penalizes unless placed in a deprecation/notes section.

Actionability

Fully executable guidance: exact resolution order for the memex root, exact index path and Glob fallback, concrete citation examples with real REF syntax, exact answer-page frontmatter, slug derivation rules with an example, collision handling ('append -2, -3'), exact user prompt text for promotion, and exact log.md entry templates. Copy-paste ready for the common cases.

Workflow Clarity

Eight clearly numbered steps with explicit error-recovery paths (no memex found → stop message; missing index.md → Glob fallback; gaps → status: gap filing; slug collision → -2 suffix) and guardrails (never overwrite, read-only on raw/). Not 5: there is no final verification checkpoint confirming the filed answer, log, and index entries are consistent (e.g. running/mentioning the audit script is only suggested in one fallback branch), a minor validation gap.

Progressive Disclosure

No bundle files exist and none are referenced, so the skill is scored as self-contained: well-organized sections ('Memex Resolution', numbered 'Ask Steps', 'Honest-Gaps Rule', 'Constraints') covering a single coherent procedure. Not 5: at ~127 lines it exceeds the under-50-lines simple-skill exception, and a few cohesive chunks (the promotion procedure, the honest-gap filing details) could plausibly live in a one-level-deep reference file to slim the main file.

Specificity

Quotes several concrete actions in third person — 'Reads the index, follows one SALP-REF hop, files a cited answer back' — plus a boundary ('Not for source code'). Not 5: 'SALP-REF hop' is project jargon whose meaning is only implicit here, and capabilities like citation format and honest-gap reporting are only partially covered. Not 3: it lists multiple specific actions, not just 1-2.

Completeness

Explicitly answers both: what ('Reads the index, follows one SALP-REF hop, files a cited answer back') and when ('Use when the user asks a question to be answered from the project's memex... rather than training data or the live web'), with a concrete negative boundary ('Not for source code'). Clearly matches the anchor-5 example's structure of capability list + explicit 'Use when' clause.

Trigger Term Quality

Relevant keywords present — 'asks a question', 'memex', 'training data', 'the live web' — but common natural variations users would say are missing: 'look this up in my notes', 'search my knowledge base', 'ask the memex' (the latter appears in the body, not the description). Not 4: the missing variations are ones users would plausibly say, not just a few.

Distinctiveness Conflict Risk

Clear niche ('project's memex (long-term knowledge bank for AI agents)') with distinct triggers and an explicit exclusion ('Not for source code', 'rather than training data or the live web') that separates it from general Q&A, web search, and code skills. Minimal conflict risk.