Multi-persona analysis. Use when the user wants a persona-driven deep dive, second opinion, adversarial pressure on a held position, side-by-side perspectives, or a synthesized panel take.
Use when the operator states a held technical belief they're about to act on — an assertion about reality already shaping a decision ("RAG is sufficient", "the latency issue is network-bound", "we don't need observability yet") — and wants adversarial pressure before building on top of it. Symptoms — operator phrases it in the indicative ("X is Y", "we don't need Z") not as a question; resources (a roadmap line, a sprint, headcount, a rewrite) are about to depend on the belief being true; operator uses words like "obviously", "clearly", "enough", "we don't need".
Use when operator has run episodes and feels persistent friction with how the harness thinks, or when pointing the harness at a new domain where the default agent prompts and phase order seem off. Use when the operator wants to tailor the harness shape (agent prompts, scout PHASES, skill wording) to how their main persona and upstream personas actually reason — without touching signal/episode infrastructure. Symptoms — the same kind of objection keeps appearing in review.md across unrelated increments; operator says "this agent asks for the wrong things"; harness was ported from another project and hasn't been adapted.
Use when goal.md exists but no active plan is set in config — a stated goal needs to be broken into connected increments before the scout can walk them. Also use when re-planning an existing goal (produces a new goal-plan-N, leaves goal.md and prior plans intact). Symptoms — scout.py prints `no-goal-plan`; operator has a goal but no increment list; operator wants to try a different breakdown of an existing goal.
Use when `goal.md` and `goal.harness.json` both exist and the operator is ready to walk increments. Explains the scout's 4-phase episodic loop, the four work agents, and — critically — when to use which panel intent for upstream review. Default review pattern is `review` (parallel_with_main, upstream personas). This skill also documents when to reach for `challenge` / `explore` / `ask` instead.
Use when the panel harness has no goal.md yet — operator needs to turn a vague sense of "I should work on something" into a specific goal tied to their actual context. The harness models any system against the operator's persona; the subject can be technical (a refactor, an architecture decision, a research question) or not. Symptoms — scout.py prints `no-goal-statement`; operator describes wanting to start but can't name the specific thing.
Use when the operator is making a decision about an artifact, feature, or change that will sit inside a system with people, time, or relationships — and they want to surface emergent / cross-cutting / second-order effects that neither upstream (author-side) nor downstream (consumer-side) reads would catch. Symptoms — operator says "what am I missing?" / "what might happen that we didn't plan for?" / "something feels off and I can't name it"; the decision has multi-month unfold potential; the artifact touches an ecosystem, a team, or user behaviors that might reshape around it. NOT for intrinsic-quality reviews, NOT for audience-reception reads, NOT for belief stress-tests.