Conciseness

The body is dense and table-driven with concrete commands and no tutorial-style explanation of concepts Claude already knows; concept material (e.g. "Git's commit history is a DAG") is correctly offloaded to references. Not a 5: a 10-item trigger-phrase list duplicates the description's job, and lines like "Equivalent to AgentHub's 'leaves' query" add little.

Actionability

Concrete, mostly copy-paste-ready guidance throughout: `python scripts/result_ranker.py --session {id} --eval-cmd "pytest bench.py --json" --metric p50_ms --direction lower`, `git merge --no-ff`, `git tag hub/archive/{session}/agent-{i}`, a complete agent dispatch prompt, and a full board post format. Not a 5: slash commands like `/hub-init` and `/agenthub:spawn` are never shown with example arguments, and one status command appears without the `python scripts/` prefix.

Workflow Clarity

The INIT → DISPATCH → MONITOR → EVALUATE → MERGE lifecycle is clearly sequenced with numbered phases, a state-transition table, and a proactive-triggers table providing failure feedback loops (all crashed → retry with different constraints; stuck session → check board, consider timeout). Not a 5: there is no post-merge verification (e.g. run tests on the merged result), and the "No improvement over baseline" trigger references a baseline that the Init step never explicitly establishes.

Progressive Disclosure

The body is an overview with well-signaled one-level-deep references that all resolve to real bundle files (agent-templates.md, coordination-strategies.md, dag-patterns.md) plus a scripts table covering 5 of the 6 scripts. Not a 5: the "More in this skill" list ends with a command (`python3 scripts/dry_run.py --help`) masquerading as a loadable reference, and dry_run.py is missing from the Scripts table — minor navigation confusion.

Specificity

"Run several agents on one task in parallel", "spawns agents in git worktrees, ranks and merges the winner" lists multiple concrete actions covering the full spawn → rank → merge lifecycle, plus use domains ("faster code, copy variants, research angles"). Not a 4: coverage of the core lifecycle is comprehensive rather than having minor gaps.

Completeness

Both parts are explicit: the what ("Run several agents on one task in parallel and keep the best result... spawns agents in git worktrees, ranks and merges the winner") and the when ("Use whenever the user wants multiple attempts compared or best of N"). Not a 5: the when-clause lists only two trigger conditions where the anchor-5 example covers a fuller set of concrete trigger variants.

Trigger Term Quality

"Use whenever the user wants multiple attempts compared or best of N" gives good natural trigger phrases users would actually say. Not a 5: common variations the body itself lists ("fan-out", "tournament", "A/B test", "try multiple approaches") are absent from the description.

Distinctiveness Conflict Risk

The niche is clear — "keep the best result", "best of N", "ranks and merges the winner" — and unlikely to trigger for unrelated skills. Not a 5: the opening "Run several agents on one task in parallel" is generic fan-out language that overlaps with general multi-agent/subagent orchestration skills.