Conciseness

The bulk is dense, novel protocol (drift taxonomy, recovery/exit steps, handoff template) rather than concepts Claude already knows, but there are several padded passages: the extended ocean-wave metaphor ('Paddling in… The close-out'), motivational prose ('No wipeouts. No zombie sessions.', the closing Principles section), a duplicated anchor enumeration, and a stranded pipeline bullet list that repeats activation guidance. This sits between anchors 2 and 3 — the padding is roughly a quarter of the document, above the midpoint, so 3 rather than 2.

Actionability

Fully executable guidance throughout: copy-paste commands ('entire status 2>/dev/null', 'ls .context-surfing/handoff-*.md 2>/dev/null'), a complete settings.json hook configuration with exact script path (verified to exist in the bundle), an exact handoff filename pattern and full template, explicit re-anchor limits ('at most one local re-anchor and one cold-context check'), and a concrete read-only subagent brief. Standalone-mode fallbacks cover the common cases where optional dependencies are absent.

Workflow Clarity

The lifecycle is clearly sequenced (activation → drift detection → recovery protocol → exit protocol → next-session resume) with explicit validation checkpoints: the pre-commit anchor check gates every side-effecting action, the recovery protocol is a bounded feedback loop with escalation, and the handoff is 'marked consumed only after re-entry succeeds'. The handoff template doubles as a checklist. No destructive or batch operations apply the score-3 cap.

Progressive Disclosure

Good section structure with clear headers and a real, correctly pathed bundle script reference (scripts/handoff-checker.sh, confirmed present). Not score 5 because the SKILL.md is a ~420-line monolith rather than an overview pointing to reference files — the handoff template and drift-signal taxonomy are candidates for separate reference files — and there are minor organization gaps, notably the stranded 'Full pipeline / Partial pipeline / Standalone' bullet list that belongs in the Activation section.

Specificity

The description lists several concrete actions — 'Monitors context window health', 'Uses bounded external anchors, one cold-context review when self-recovery is uncertain, and a clean handoff on real drift' — which is more than the 1-2 actions of a score-3 anchor. It falls short of score 5 only because the actions are named but slightly abstract (e.g., 'bounded external anchors') rather than fully concrete, and voice is consistently third person so no penalty applies.

Completeness

Both what (monitor + anchors + cold-context review + clean handoff) and when ('during large, long-running, multi-session, or explicitly context-sensitive work', plus the negative boundary 'Do not auto-activate merely because a medium task has an intent frame and plan') are explicit, so the score-3 cap for missing trigger guidance does not apply. Not score 5 because the 'when' names task classes rather than concrete user trigger phrases, and 'large' is undefined without the sibling pipeline's taxonomy.

Trigger Term Quality

Good coverage of natural terms a user would say: 'context window health', 'long-running', 'multi-session', 'drift', 'handoff'. It misses a few common variations or synonyms (e.g., 'context degradation', 'session handoff', 'compaction', 'token budget'), keeping it below the comprehensive score-5 anchor but above score 3's 'missing common variations'.

Distinctiveness Conflict Risk

Clear niche (context-quality drift vs. scope drift) with an explicit negative boundary against auto-activation, giving minimal confusion with intent-framed-agent. Minor overlap risk remains with the closely related session-context skill around shared 'session'/'context' trigger language, which matches the score-4 anchor rather than score 5's 'minimal conflict risk'.