Conciseness

The body is lean and operational: exact bash invocations (e.g., 'node "$HOME/.claude/skills/ck/commands/init.mjs"'), a compact data-layout tree, exact JSON schemas, and one-line rules. It explains nothing Claude already knows and every token is instructional, matching anchor 5 ('Lean and efficient; assumes Claude's competence'). Not anchor 4 — there is no over-explanation left to trim; the repeated 'Display output verbatim' lines are necessary per-command instructions.

Actionability

Guidance is fully executable and copy-paste ready: concrete node commands for all seven commands, piping patterns ('echo '<json>' | node ".../save.mjs"'), an exact JSON schema ('{"summary":"...","leftOff":"...","nextSteps":["..."],...}'), a confirmation-draft template, and a complete settings.json hook registration block. Matches anchor 5 ('Fully executable; copy-paste ready... specific examples cover the common cases'); not anchor 4, since no common case is missing.

Workflow Clarity

Multi-step flows are explicitly sequenced with validation checkpoints: init (run script → present confirmation draft → 'Wait for user approval' → pipe confirmed JSON), save (analyze → 'Wait for confirmation' → pipe), forget ('This will permanently delete context... Are you sure? (yes/no)' before the destructive delete), and migrate ('--dry-run' first; 'Originals are backed up... nothing is deleted'). Feedback loops are present ('If a script exits with code 1, display its stdout as an error message'; malformed projects.json → 'offer to reset it to {}'), so the destructive/batch cap-at-3 rule does not apply. Matches anchor 5 with explicit validation and error-recovery handling.

Progressive Disclosure

The body is a well-sectioned overview (Data Layout, per-command sections, SessionStart Hook, Rules) with trivial header-based navigation; no bundle files exist, and no material that belongs in a separate file is inlined — the JSON schemas and hook snippet are each only a few lines placed at point of use, while the bulk detail correctly lives outside SKILL.md as executable scripts. Matches anchor 5's 'easy navigation' and appropriate placement; not anchor 4, as there are no organization gaps or misplacements.

Specificity

The description enumerates seven concrete commands ('init, save, resume, info, list, forget, and v1-to-v2 migrate') plus a concrete hook behavior ('a SessionStart hook that injects a compact project brief'), matching the anchor-5 example of multiple specific concrete actions with comprehensive coverage — not anchor 4, since there are no gaps in the operation coverage.

Completeness

It explicitly answers both: what ('Persistent per-project memory for Claude Code (Context Keeper) driven by deterministic Node.js /ck commands... plus a SessionStart hook') and when ('Use when context must survive across sessions, saving session state..., resuming..., or picking up a project without re-explaining it'). The 'when' clause is fully explicit with concrete trigger phrases, matching anchor 5 rather than anchor 4 where the 'when' could be more specific.

Trigger Term Quality

Natural user phrasings with synonym coverage are present: 'context must survive across sessions', 'saving session state with next steps and decisions', 'resuming where a previous session left off', 'picking up a project without re-explaining it', 'Persistent per-project memory'. This is comprehensive natural-term coverage for the domain, matching anchor 5 rather than anchor 4 (no common natural variant of the trigger intent is missing).

Distinctiveness Conflict Risk

A clear niche is stated (per-project context persistence in Claude Code via named /ck commands and Context Keeper branding) with triggers specific to session-continuity intent, giving minimal conflict risk — matching anchor 5 ('Clear niche with distinct triggers') rather than anchor 4, since it would not plausibly trigger for unrelated skills.