Conciseness

The body is six lean lines with zero padding or explanation of concepts Claude already knows — 'Read `.claude/skills/harvest/SKILL.md` and execute it exactly as written' plus one substitution rule. Every token earns its place, matching the 'lean and efficient' anchor exactly; it cannot be 4+verbose because there is no over-explanation anywhere.

Actionability

The only guidance is an indirection: 'Read `.claude/skills/harvest/SKILL.md` and execute it exactly as written' and 'follow `.agents/rules/cog.md`' — no actual steps for capturing learnings, staging, or proposing patches appear in this body, and the referenced files are not part of the bundle (`.claude/skills/harvest/SKILL.md` is self-referential, implying recursion). This matches the 'minimal concrete guidance; missing the specific steps to execute' anchor and mirrors the bad example of a pointer-only skill; not 3 because there are no partially-executable steps at all — the work is entirely delegated elsewhere.

Workflow Clarity

The sequence is roughly 'read playbook → execute it → then follow `.agents/rules/cog.md`' with one conditional substitution rule — a rough order with major gaps: no definition of what executing entails, no checkpoints, and no validation despite the skill involving staging durable knowledge for promotion (the description's approval gate never appears in the body). Fits 'rough sequence present but many gaps; validation absent'; not 3 because even the step list itself is incomplete rather than merely un-validated.

Progressive Disclosure

No bundle files exist (no references/, scripts/, assets/) yet the body's entire substance lives in external files — `.claude/skills/harvest/SKILL.md`, `.agents/rules/cog.md`, `.agents/agents/<name>.md` — none of which are shipped, and the primary reference is self-referential. The skill is effectively a pointer shell with buried references, matching 'minimal structure; references are buried'; not 1 because there is only one level of indirection stated, not a demonstrated 2+ level chain; not 3 because there is no in-body overview content at all to organize.

Specificity

Three concrete actions are named — 'Capture durable session learnings', 'stage for human promotion to 05-knowledge/lizard', and 'propose skill/CLAUDE.md patches' — which matches the 'several specific actions, minor gaps' anchor. Not a 5 because 'capture durable session learnings' is somewhat abstract and no output formats or mechanics are described; not a 3 because three distinct actions with a concrete target directory exceed '1-2 concrete actions'.

Completeness

The 'what' is explicit (capture learnings, stage for promotion, propose patches) and the 'when' is explicit via a dedicated trigger clause: 'Triggered by /harvest, SessionEnd hook staging, or nightly enhance', plus the boundary statement 'Never writes durable knowledge without your approval'. This matches the anchor requiring both what and when with concrete trigger phrases; not 4 because the trigger guidance is fully explicit, not weakly implied.

Trigger Term Quality

Relevant keywords exist ('/harvest', 'session learnings', 'skill/CLAUDE.md patches', 'SessionEnd hook staging', 'nightly enhance') but the natural user phrasings are largely internal jargon — '05-knowledge/lizard', 'SessionEnd hook staging', 'nightly enhance' are not terms a user would naturally say. This fits 'some relevant keywords but missing common variations or synonyms'; not 4 because there are no synonyms or natural-language variants of the core trigger concept.

Distinctiveness Conflict Risk

The description carves a fairly distinct niche (session-knowledge harvesting with named destinations like '05-knowledge/lizard' and 'skill/CLAUDE.md patches') and triggers are specific (/harvest, SessionEnd hook). Fits 'mostly distinct; minor overlap risk' — it could overlap with memory/reflection or CLAUDE.md-editing skills; not 5 because 'session learnings' and 'CLAUDE.md patches' are broad enough that adjacent skills (memory, init) could plausibly compete for the same requests.