Conciseness

The operational core (drift signal lists, recovery protocol, exit protocol, handoff template) is concrete and earns its tokens, but the extended surfing metaphor ("Paddling in", "The peak", "The shoulder", "The close-out"), the motivational Principles section ("Ride the peak, not the whole ocean"), and repeated anchor listings (in Activation, the Full/Partial/Standalone bullets, and Project Context Files) are trimmable padding across ~420 lines. It fits anchor 3 — mostly efficient with some unnecessary explanation that could be tightened — rather than anchor 2 because the bulk is specific non-obvious guidance, not generic concepts Claude already knows.

Actionability

The body provides executable, copy-paste-ready artifacts: the `entire status 2>/dev/null` detection command, a complete handoff file template with exact naming pattern, the exact hooks JSON for `.claude/settings.json`, a verbatim user-notification message, and the `ls .context-surfing/handoff-*.md` fallback check. It falls short of anchor 5 because the cold-context reviewer brief in Recovery Protocol Step 2 is described rather than supplied as ready-to-send text, and directives like "maintain a running status in your output" lack a concrete format.

Workflow Clarity

The lifecycle is explicitly sequenced (activation → riding with pre-commit anchor checks → drift detection → recovery protocol → exit protocol → handoff/next session → close) with genuine validation checkpoints and feedback loops: the pre-commit anchor check gates every side-effecting action, re-anchor limits cap recovery attempts ("at most one local re-anchor and one cold-context check"), escalation paths branch on reviewer verdicts, and the handoff-consumed marker is set "only after re-entry succeeds". The handoff template doubles as a checklist, matching anchor 5's clear sequence, explicit validation steps, and error-recovery loops.

Progressive Disclosure

Section headers are clear and the one bundle file (`scripts/handoff-checker.sh`) is referenced correctly with a real path, but the ~420-line body inlines substantial material that could live in separate reference files — the cross-skill relationship/cadence rules, the pipeline interop tables, and detailed Entire CLI integration notes — with no use of a references/ directory at all. This fits anchor 3 (some structure, but content that should be separate is inline) rather than anchor 4's appropriately-placed split.

Specificity

The description lists several distinct 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" — covering the skill's core behaviors with only minor gaps. It falls short of anchor 5 because "bounded external anchors" is jargon rather than a fully concrete action, and it exceeds anchor 3 since it names more than 1-2 actions.

Completeness

Both 'what' (monitor context health via bounded anchors, cold-context review, clean handoff) and 'when' ("during large, long-running, multi-session, or explicitly context-sensitive work") are explicitly present, reinforced by an anti-trigger ("Do not auto-activate merely because a medium task has an intent frame and plan"). It is not a 5 because the 'when' lacks concrete user-said trigger phrases and terms like "large" are undefined; it exceeds anchor 3 because 'when' is explicit, not merely weakly implied.

Trigger Term Quality

It includes some relevant keywords a user might naturally say — "long-running", "multi-session", "context window", "drift", "handoff" — but misses common variations and synonyms such as "context degradation", "context pressure", "compaction", or "session handoff". The trigger framing is skill-centric rather than user-utterance-centric, fitting anchor 3 rather than anchor 4's good keyword coverage.

Distinctiveness Conflict Risk

The context-quality-drift niche — monitoring the agent's own coherence and handing off on real drift — is mostly distinct from scope-monitoring or session-resumption skills, and the explicit anti-trigger actively reduces overlap with intent-framed-agent. Minor overlap risk remains with session-state/handoff-type skills, so it fits anchor 4 rather than the clear minimal-conflict niche of anchor 5.