Conciseness

The 840-line body duplicates content heavily: context-detection and banner instructions appear twice (the EXECUTION CONTRACT's Steps 1-2 and again verbatim under "# Discover Workflow - Discovery Phase"), the contract steps are re-summarized a third time in "Implementation Instructions", the Dev/Knowledge banner templates are near-identical pairs, plus ASCII art and generic TaskCreate boilerplate. This matches anchor 2 ("Noticeably verbose; several unnecessary explanations or padded sections") rather than 1, since the padding is duplication, not explanations of concepts Claude already knows.

Actionability

Guidance is mostly executable: exact script paths (orchestrate.sh probe-single, build-fleet.sh, state-manager.sh), a complete Agent prompt template, an intensity-to-fleet-size table, and per-step error handling. Not 5: the legacy "How It Works" section instructs a direct "orchestrate.sh discover" call that the contract explicitly prohibits, and placeholders like ${PROMPT} and ${codex_status} are never defined, leaving minor gaps and one contradiction.

Workflow Clarity

The execution contract lays out a clearly sequenced, blocking Step 1-7 flow with explicit validation checkpoints (research-verify gate, synthesis-file existence checks), a minimum-2-results threshold, graceful degradation rules, and per-step error handling with feedback loops. Not 5: the appended legacy section presents a second, conflicting workflow (single orchestrate.sh call, background_task examples) that contradicts the contract and muddies which sequence governs; validation itself is thorough, so 3 would undervalue it.

Progressive Disclosure

This is an 840-line monolithic SKILL.md with no references/, scripts/, or assets/ bundle; banner templates, presentation templates, security-framing details, example transcripts, and state-update code that clearly belong in separate files are all inlined. The two files it does cite ("skills/blocks/codex-host-adapter.md" and "skill-security-framing.md") are dangling — not present in the bundle. Matches anchor 2 ("content that clearly belongs in separate files is inlined; or references are buried") — not 3, since hundreds of lines of separable material are inlined and the one signaled reference is unverifiable.

Specificity

"Multi-AI research" names the domain, but the only action stated is the generic "research using available external providers" — no concrete actions (dispatching probes to external CLIs, building an agent fleet, synthesizing multi-perspective findings) appear. This matches anchor 2 ("Names the domain but actions are minimal or generic") rather than 3, which requires 1-2 concrete actions, and not 1 because the domain is specifically named.

Completeness

The "what" is stated ("Multi-AI research using available external providers"), but the "when" is only weakly implied by the "(Double Diamond Discover phase)" parenthetical with no "Use when..." clause or equivalent explicit trigger guidance, capping this at 3 per the rubric guidelines. Not 4, because the when-condition is never made explicit.

Trigger Term Quality

"research" is a natural term a user would say, and "Multi-AI" / "external providers" are relevant domain keywords, but common variations and synonyms are missing (multi-provider, provider names, compare, investigate, web research). "Double Diamond Discover phase" is framework jargon users would not naturally say, matching anchor 3 ("Some relevant keywords but missing common variations or synonyms") — not 4, since several natural terms users would actually speak are absent.

Distinctiveness Conflict Risk

The "Double Diamond Discover phase" parenthetical carves a clear niche distinct from sibling workflow phases (define/tangle/ink), with only minor overlap risk against generic research skills since "research" alone is broad. Not 5 because no distinct trigger phrases disambiguate it at invocation time, and not 3 because the multi-AI plus phase framing is more specific than anchor-3-level overlap.