Conciseness

The body is dense and assumes Claude's competence — it explains only environment-specific CCB conventions, not general concepts. However, the 'pane text/evidence is not authority' rule is restated in steps 1, 2, 3, and again in Red Lines, and a few qualifiers ('Do not infer an all target', 'previews are not authority') could be consolidated, which fits the score-4 anchor (efficient with minor over-explanation to trim) rather than the lean score-5 anchor.

Actionability

Concrete executable commands throughout ('command ccb ping ccbd', 'ccb queue --detail "$AGENT"', exact tmux capture commands, the gh issue create example, and a concrete output template), but the repair commands 'command ccb repair ack ... / retry ... / resubmit ...' use unspecified ellipsis arguments rather than executable forms. That is a minor gap, matching the score-4 anchor rather than the fully copy-paste-ready score-5 anchor or the pseudocode-level score-3 anchor.

Workflow Clarity

Six clearly sequenced steps (establish authority → deep pane diagnosis → classify → bounded recovery → verify → incident bundle/authorization) with explicit validation checkpoints: step 5 re-runs the smallest relevant checks and re-captures the pane, gates require stopping when blocked, and destructive actions (restart, clear, resubmit, issue submission) each have explicit preconditions. This matches the score-5 anchor (clear sequence, explicit validation, feedback loops, checklists).

Progressive Disclosure

The skill is a self-contained SKILL.md (no references/, scripts/, or assets/ exist, so no references are broken) with well-organized sections (Scope, Workflow with six numbered subsections, Red Lines). At 194 lines, the pane-state taxonomy and the classification output format are inline material that could plausibly live in a one-level-deep reference file, which fits the score-4 anchor (good structure, most content appropriately placed, minor organization gaps) rather than the ideal split of the score-5 anchor.

Specificity

Lists multiple specific concrete actions with comprehensive coverage: 'Diagnose a named CCB agent', 'combining authoritative runtime and job lineage with deep read-only pane inspection', 'apply bounded recovery when evidence supports it', 'verify the result', and 'request authorization before submitting a redacted GitHub issue'. This matches the score-5 anchor (multiple specific concrete actions, comprehensive) and is not the level below, which expects only 'several specific actions; minor gaps'.

Completeness

Explicitly answers both questions: the 'what' is the full diagnose → recover → verify → authorized-issue-submission pipeline, and the 'when' is an explicit 'Use for ...' clause with concrete trigger phrases. This matches the score-5 anchor exactly; the score-4 anchor would require the 'when' to be less explicit or specific.

Trigger Term Quality

Triggers cover both explicit invocation aliases ('$ccb_diagnose agentname', '$ccb-diagnose agentname') and the natural failure phrases a user would actually say: 'stuck, disconnected, not continuing, not replying, or showing provider errors'. This is comprehensive coverage including synonyms, matching the score-5 anchor rather than the score-4 anchor, which expects 'a few natural terms missing'.

Distinctiveness Conflict Risk

'CCB agent' names a clear niche with distinct triggers (CCB-specific invocations and CCB-agent failure symptoms), so there is minimal conflict risk with generic maintenance or debugging skills. It matches the score-5 anchor ('clear niche with distinct triggers') and is clearly above the score-4 anchor ('minor overlap risk with closely related skills').