Conciseness

Nearly every section carries skill-specific operational content (stream resolution regex, token budgets, harness-bug guards, anti-pattern fixes with GOOD/BAD bash) rather than concepts Claude already knows. It sits at anchor 4 rather than 5 due to trimmable redundancy: Phase 3/3b steps and the archive command are restated almost verbatim from references/reference.md, and the Usage Examples section largely re-illustrates what earlier sections already specify. It stays above 3 because the padding is duplication of genuinely useful material, not explanation of known concepts.

Actionability

Concrete, executable guidance dominates: the exact upsert invocation with sample arguments, the literal mkdir/mv archive command, the stream-name regex, and copy-paste GOOD/BAD bash snippets in Anti-Patterns. It falls short of anchor 5 because Phase 1 uses shorthand pseudo-commands ("Bash: ls .context/session/ + ls -t ...") that must be translated into two real calls, and how the status value is determined before archiving is left implicit; it is clearly above 3 since the core commands are literal and complete.

Workflow Clarity

The three-phase sequence (gather → analyze/synthesize → write/report, plus 3b archive) is clearly ordered with parallel-call guidance, and the destructive- overwrite risk is guarded by the 'NEVER overwrite without reading first' anti-pattern plus the merge-forward GOOD example, complemented by the quality self-check in reference.md. It holds at 4 rather than 5 because there is no explicit post-write verification (e.g., confirming the token budget or that the INDEX row landed), and status resolution before auto-archiving is implicit rather than a stated checkpoint.

Progressive Disclosure

SKILL.md is a well-organized overview (workflow, when/when-not, anti-patterns) with the heavy detail — file template, status mapping, upsert and archival rules — correctly split into references/reference.md, and the index-update logic into scripts/upsert-index.sh. Both are real files, referenced one level deep with a clear inline pointer ("See references/reference.md for...") and a bottom References table with a When-to-Use column; reference.md contains no second-level reference chain. This matches anchor 5.

Specificity

"Save session to CONTEXT-llm.md with conversation summary" names the concrete artifact and one main action, but stops there — it never lists what the summary contains (e.g., Next tasks, decisions, hot files) or the index update/archive side effects. This matches anchor 3 ("Names domain and 1-2 concrete actions, but not comprehensive"); it is above 2 because the target file and output format are explicit, and below 4 because it does not list several specific actions.

Completeness

Both halves are explicit: what — "Save session to CONTEXT-llm.md with conversation summary" — and when — "Use when saving work, checkpointing progress, preserving session state" — followed by concrete quoted trigger phrases. This matches the anchor-5 exemplar pattern (what + when + concrete triggers) exactly; nothing is left implicit.

Trigger Term Quality

Explicit natural triggers — "save context", "save session", "checkpoint", "save my progress" — are phrases a user would plausibly say, plus "saving work" and "preserving session state" as use-conditions. It sits below anchor 5 because common variations ("save my work", "preserve state", "session snapshot") are absent and no file/keyword like CONTEXT is offered as a trigger; above 3 because several genuinely natural terms are present rather than just domain keywords.

Distinctiveness Conflict Risk

The CONTEXT-llm.md artifact and "save context"/"checkpoint" triggers carve a clear niche, but "preserving session state" and "save session" could overlap with closely related skills in the same family (e.g., a handover-document or load-context skill). Fits anchor 4 ("Mostly distinct; minor overlap risk with closely related skills") — not 5 because the session-preservation purpose is broad enough to collide with sibling context skills.