Conciseness

The body is mostly operational (commands, not concept explanations), but it can be tightened: the ~30-line valid_root/plugin-root resolution bash block is duplicated verbatim in Step 3 and Step 5, and Step 0 spends ~40 lines on narration-rule contract with violation/normal examples that could be compressed. This fits the anchor 'mostly efficient but includes some unnecessary explanation or could be tightened' rather than the level-4 anchor, where over-explanation would be only minor.

Actionability

Every step ships a fully executable bash block with explicit exit-code handling, retry limits, error markers (ERROR/WARN/ABORT/CHERRY_PICK_CONFLICT), and fallbacks (e.g. TASK_SUMMARY fallback message, --no-verify rationale). This is copy-paste-ready guidance covering the common cases, matching the top anchor; the only placeholders (<task-description>, <task>) are inherent to a parameterized workflow.

Workflow Clarity

Steps 0-8 are clearly sequenced with validation checkpoints throughout: parallel pre-checks (Step 2), plugin-root/backend/model resolution guards (Step 3), dirty-changeset auto-commit/amend before review (Step 6), contract-gate scripts run in the worktree, Plans.md clean-precondition check before cherry-pick, and explicit failure/abort/retry branches (1 retry, then REQUEST_CHANGES or ABORT). This matches the top anchor with explicit validation steps, feedback loops, and error-recovery paths — critical given the destructive cherry-pick-to-main and force worktree removal.

Progressive Disclosure

Structure is good: the single bundle reference (references/cursor-cli-only.md, verified to exist) is clearly signaled twice with a ${CLAUDE_SKILL_DIR} link and one level deep, and the inline bash blocks belong inline as executable workflow. It falls short of the top anchor because the body is a ~400-line monolith where the duplicated resolution logic (used in two steps) arguably belongs in a shared script under scripts/, and the narration contract could live in a reference file.

Specificity

The description lists multiple specific concrete actions — "Delegate a single write task to Cursor Composer via cursor-companion.sh inside an isolated worktree, then Lead-review the diff and cherry-pick" — naming the exact tool (cursor-companion.sh) and the full pipeline (worktree isolation, Lead diff review, cherry-pick). This matches the anchor for comprehensive, specific concrete actions; it is not the level-4 anchor because there is no meaningful coverage gap in the action list.

Completeness

Both "what" (delegate single write task via cursor-companion.sh in an isolated worktree, Lead-review, cherry-pick) and "when" (explicit 'Use when...' clause with concrete trigger phrases) are clearly and explicitly stated. The additional "Do NOT load for" exclusions strengthen the when-clause further; nothing is missing or only implied.

Trigger Term Quality

Trigger phrases are natural and comprehensive with synonyms: "Use when user invokes cursor:do, says delegate to cursor, have composer write it, refactor with cursor, hand a file edit to Composer." It goes beyond the level-4 anchor ("a few natural terms missing") by covering the command name plus multiple phrasings a user would actually say.

Distinctiveness Conflict Risk

The niche is clear (single-task write delegation to Cursor Composer) and conflict risk is actively minimized by explicit negative triggers — "Do NOT load for: planning, code review only, read-only investigation, or multi-task team runs (use breezing --cursor or cursor:ask instead)" — which route near-miss intents to the correct sibling skills. This is the minimal-conflict anchor, above the 'minor overlap risk' level-4 anchor.