Lean and dense — no explaining concepts Claude already knows, no padding; compressed phrasing like 'A claim is a sentence with a source' and tables/inline lists assume competence, every token earning its place.

Executable, copy-paste-ready commands throughout — 'uv run {project-root}/_bmad/scripts/memlog.py', 'uv run scripts/recon_kit.py slug', and a complete headless JSON output block — concrete rather than abstract.

Clear sequenced activation steps, an intents table routing to references, the plan-gate hard stop, and landing-time verification checkpoints with feedback loops (re-validate on failure) for the destructive/batch research operations.

Overview-only SKILL.md with well-signaled one-level-deep references (draft/process/run/lifecycle/selection/synthesis/verification/finalize); all cited paths resolve to real bundle files, and the overview appropriately defers detail to those references.

Lists multiple specific concrete actions — 'draft a deep-research prompt', 'process a finished research report — file it, distill a succinct cited summary', 'run the research here through web fan-out' — plus named type packs and a select decision shape.

Clearly answers both 'what' (three services with type packs and custom types) and 'when' with an explicit 'Use when ...' trigger clause listing concrete activation phrases.

Strong natural-phrase coverage a user would actually say: 'deep recon', 'research this', 'draft a research prompt', 'process this research report', 'market research', 'domain research', 'technical research', 'competitor research', 'literature review', 'help me choose between'.

A clear niche (decision-grade research via three routing modes) with distinct triggers like 'deep recon' and 'process this research report' that are unlikely to collide with other skills.