odin-research-advanced

Use when the user asks for a council, second opinions, or parallel investigation. Spawns parallel subagents from distinct stances, returns a decision memo. Not for cross-examination of existing proposals — use cross-critique.

Use when independent proposals on a contested decision need structured cross-critique before choosing. Reuses original authors and synthesizes convergence, objections, and the surviving option. Not for parallel multi-stance investigation — use council.

Use when one reviewer is not enough because failure modes are heterogeneous, or a claim needs cross-lens pressure before it ships. Returns a convergence picture with per-lens verdicts and the single resolving question. Not for collapsing a decision field — use converge.

Use when the user asks to research a topic and produce a thorough sourced report. Produces a prose-first cited Markdown report with key findings, recommendations, and risks. Don't use for remote, credential, publish, deploy, or irreversible changes.

Use when a user suspects the current direction is tunnel-visioned or inherited its framing: spawn 2-5 independent reads, classify each, cluster by shared root, and surface divergent-incompatible reads first. Not for framing-bias checks on a question — use prism.

Use when asked to "interrogate" or run an adversarial multi-model review of a supplied code artifact. Returns an act/consider/note/dismiss verdict with an agreement map. Don't use for tasks that require source or remote-system changes.

Use when the user asks why a plan is wrong or says "poke holes in this". Returns one load-bearing objection and the cheapest experiment that would prove whether it matters. Don't use for tasks that require source or remote-system changes.

Use when one reviewer angle is insufficient or the user asks to prism an artifact or review it from different angles. Selects independent lenses, strips session framing, fans out fresh zero-context reads, and reports divergences before convergences with one decisive resolving question. Not for source or remote-system changes.

Use when invoking /product-signal-pulse with an optional lookback window to query configured product signals. Resolves config, runs a first-run interview if unconfigured, dispatches read-only queries to analytics, tracing, and payment sources, and writes a 30-40 line pulse report under docs/pulse-reports. Not for credential, publish, deploy, or irreversible changes.

Use when asked to investigate design rationale, regression causes, or thresholds via source playbooks. Produces a confidence-tiered cited narrative with unknowns retained. Not for why-style decision rationale from conversation context — use why.

Use when asked to analyze a screen recording, voice capture, or meeting notes artifact for product feedback. Routes to a quick bug report or extensive analysis with timestamped evidence artifacts and a brainstorm handoff. Not for credential, publish, deploy, or irreversible changes.

Use when researching a named library, framework, SDK, API, or service, or finding a migration guide. Produces a cited Markdown artifact written to disk with source-cited claims, confidence labels, and open questions. A primary-source-only mode restricts the ladder to official docs, API refs, and source code. Not for codebase-internal research — use scout.

Use when invoked as visual-fact-check with a document path. Verifies every factual claim against its cited sources, corrects errors in place, and appends a verification summary. Not for both-ways fact-checking of assertions — use verify-both-ways.