Conciseness

The operational core (drift signal lists, recovery protocol, handoff template, hook setup) is efficient and non-obvious, but padded sections remain: the surf "Mental Model", the "Principles" aphorisms ("No wipeouts. No zombie sessions."), pipeline position stated three times (lifecycle diagram, task-type table, numbered pipeline list), and the pre-commit anchor check fully described in both "The Monitoring Paradox" and "Riding the Wave". Not 2 because the padding is flavor and repetition rather than explanations of concepts Claude already knows, and the wave lexicon is load-bearing terminology used consistently throughout.

Actionability

Fully executable throughout: exact commands ("gh skill install pskoett/pskoett-skills context-surfing", "entire status 2>/dev/null", "ls .context-surfing/handoff-*.md"), a complete copy-paste JSON hook config, a full handoff file template with named sections, concrete file conventions (".context-surfing/handoff-[slug]-[timestamp].md", the ".consumed" marker), bounded retry limits ("at most one local re-anchor and one cold-context check"), and explicit fallback chains for every optional dependency (Entire CLI, context-monitor subagent). Judgment-dependent tasks like drift detection get concrete signal lists and scripts rather than vague direction.

Workflow Clarity

The multi-step process is clearly sequenced (activation with anchor loading → drift monitoring → weak-signal recovery protocol → strong-signal exit protocol → next-session resumption) with explicit validation checkpoints: the pre-commit anchor check gates every side-effecting action, the recovery protocol has decision branches (reconcile / cold-context check / escalate / exit), and "Mark the handoff consumed only after re-entry succeeds" closes the resumption feedback loop. The handoff template functions as a checklist for the most fragile step.

Progressive Disclosure

The one bundle reference is handled well — scripts/handoff-checker.sh is a real file, clearly signaled with install-path guidance for both install methods and the plugin variant — and external artifacts (intent-framed-agent, plugin/agents/context-monitor.md) are referenced with explicit fallbacks. However, the body is a ~430-line monolith: the 40-line handoff template, the cross-skill interoperability sections, and the task-classification table are inlined content that could live in separate reference files, leaving SKILL.md as a leaner operational core. That fits anchor 3 ("content that should be separate is inline") better than anchor 4.

Specificity

"Monitors context window health", "Uses bounded external anchors, one cold-context review when self-recovery is uncertain, and a clean handoff on real drift" lists several specific actions forming the skill's mechanism. Not 5 because "bounded external anchors" and "cold-context review" are internal jargon-tinged rather than fully concrete user-facing capabilities; not 3 because multiple distinct actions are named, not just 1-2.

Completeness

Both questions are explicitly answered: the "what" ("Monitors context window health... bounded external anchors... cold-context review... clean handoff") and an explicit "when" ("during large, long-running, multi-session, or explicitly context-sensitive work"), reinforced by a negative trigger ("Do not auto-activate merely because a medium task has an intent frame and plan") that sharpens activation. Matches the 5 anchor: concrete trigger conditions accompany both what and when.

Trigger Term Quality

"large, long-running, multi-session, or explicitly context-sensitive work" plus "context window health" gives good coverage of natural phrases a user would say when needing this skill. Not 5 because common synonyms and variations ("context drift", "context is degrading", "session handoff", "zombie session") are absent; clearly above 3 since several natural trigger phrases are present.

Distinctiveness Conflict Risk

"Monitors context window health during large, long-running, multi-session, or explicitly context-sensitive work" carves a clear niche distinct from planning, scope-monitoring, and verification skills. The negative trigger ("Do not auto-activate merely because a medium task has an intent frame and plan") actively disambiguates from sibling pipeline skills, minimizing conflict risk.