odin-research

Use when a redacted, trimmed agent transcript must be appended to a GitHub PR or issue body with human approval and preview. Not for automated or model-initiated insertion without explicit human review.

Use when a verified non-trivial fix lands. Automatically captures a durable learning doc to docs/solutions/ or a CONCEPTS.md entry, or determines nothing qualifies. Also handles refresh when solution docs may have drifted. Not for unverified fixes.

Use when the user wants the current jargon weather of a domain described without advocacy. Surveys circulating terms, their signaling freight, and their usage trajectory from verifiable external sources. Not for choosing a positioning move — use buzzword-hijack.

Use when the user asks "where to help", "contribution opportunities", or "find a good first issue". Returns two to five data-backed recommendations each naming a file, a reason, and an executable first step.

Use when the learner is ready to apply cleared concepts in a real project. Scopes one capstone sized to those concepts, writes capstone-<slug>.md, yields to the learner to build it, then judges the returned work item by item on a later visit. Not for exercises or quizzes — use drill.

Use when the human returns after a gap, cannot follow the project, or asks what happened. Returns a screen-length briefing grounded in live state. Also handles pre-decision orientation when the mental model is stale. Not for initial onboarding — use onboard.

Use when the user asks to summarize competitor changelogs or analyze recent competitor releases into a deduplicated report opened as a PR. Not for researching a specific feature across competitors — use competitor-feature-research.

Use when the user asks to research a specific feature across competitor products and publish a cited report. Not for summarizing changelogs — use competitor-changelog.

Use when the user explicitly asks to save, curate, or consolidate what was learned, or closes a meaningful knowledge-work session. Curates at most three typed, approved knowledge records from existing lessons into docs/knowledge/, checks for duplicates and contradictions, and confirms by naming retrieval tags. No remote, publish, deploy, or irreversible mutation.

Use when the user asks to rank deployment platforms and stacks against their product with quantitative trade-offs. Enumerates candidates, gathers raw metrics, normalizes them to a common scale, applies user weighting, and returns a ranked list with per-axis scores. Read-only; no source or remote mutation.

Use when a long session has accumulated stale or conflicting context or when the user asks to refresh it. Rechecks context against source and drops stale items so the acting context is provably current and minimal. No remote, credential, publish, deploy, or irreversible mutation.

Use when asked to mine prior chats on a scheduled or watcher tick and maintain project memory. Reads transcripts and the memory index, extracts decisions and conventions, filters against the baseline, and writes deduplicated high-signal updates or reports no update. No remote, credential, publish, deploy, or irreversible mutation.

Use when asked to create a multi-session learning plan. Returns a milestoned plan, practice, and review rubric. Don't use for tasks that require source or remote-system changes.

Use when Survey and Job Culture Index profiles need analysis for stress, burnout, disengagement, or flight-risk signals. Calculates energy utilization, trait shifts, and a defined risk level from the paired charts, with nonclinical recommendations. Not for clinical diagnosis — use a qualified clinician.

Use when two colleagues' working friction needs trait-based explanation, accommodations, process changes, and escalation boundaries. Also handles manager-report friction when both profiles exist. Not for performance adjudication — use the organization's formal process.

Use when a predicted Culture Index profile needs comparison with role requirements, team composition, and manager profile to inform a hiring decision. Maps trait matches and confidence to numeric scores, produces a recommendation on a defined ladder, and names verification areas. Not for transcript prediction — use culture-interview-profile-prediction.

Use when asked to predict Culture Index traits from an interview transcript before a survey exists. Also handles uncertain traits when evidence is sparse or contradictory. Not for interpreting completed survey results.

Use when a manager needs profile-specific communication, one-on-one, motivation, and energy guidance for a direct report. Returns trait-gap adjustments, tailored cadence, and energy-risk actions. Not for general pair or team compatibility analysis.

Use when a signed new hire's Culture Index profile and team profiles need a first-90-days plan. Returns buddy choice, friction map, manager briefing, communication preferences, 30/60/90 actions, and success indicators. Not for manager coaching — use culture-manager-coaching.

Use when customer feedback, NPS, churn, email feedback, call transcripts, or voice-of-the-customer analysis needs a report over a time window. Produces themes, pain points, competitive mentions, feature demand, exact counts, week-over-week comparison, cited links, and verbatim quotes in a report and PR. Requires human authorization for credentials, publication, and PR creation.

Use when a user wants to record why one world won over the others. The decision and its rationale are recorded in the decision diary. Don't use for remote, credential, publish, deploy, or irreversible changes.

Use when a current decision needs pressure-testing until the rationale is clear to a skeptic. The result states the decision in plain terms or names the exact gap that prevents a defensible explanation. Don't use for tasks that require source or remote-system changes.

Use when dogfooding a developer-facing product or workflow to produce an evidence-backed eight-dimension DX scorecard with measured time to hello world and prioritized fixes. Not for visual UI audit — use web-design-review.

Use when reviewing a prose plan, spec, PRD, requirements doc, design doc, or brainstorm, or invoking /doc-review. Returns tiered findings with verbatim evidence without editing the document. Not for collaborative drafting — use doc-coauthoring. No remote or irreversible changes.

Use when making an architectural decision, changing a public API, or shipping a feature needing rationale. Produces ADRs, inline gotchas, README/API docs, and changelog entries. Not for structured doc authoring — use docs-writing; not for PR-based doc sync — use docs-update.

Use when pinning down domain terminology, maintaining the domain model, or when a term conflicts or needs sharpening. Resolves each term to one canonical owner, records it in CONTEXT.md with rejected synonyms, and keeps the glossary implementation-free.

Use when roadmap, plans, or docs may have drifted from code, or when restarting a stalled project. Returns an evidence-cited Reality Check Report with drift analysis and a prioritized reconstruction plan. Not for PR doc sync — use docs-update.

Use when a concept needs practising: the user asks for an exercise, quiz, recall, or gap probe. Runs one scaffolded practice move and records every attempt in PROGRESS.md. Not for explanations — use explain-concept.

Use when the user requests an engineering retrospective for a named period. Discovers repository and tracker telemetry, establishes source precedence, derives team breakdowns and evidence-backed habits, and writes a single report file. Not for an agent-environment retrospective — use agent-environment-retrospective; for a learning milestone — use learning-retrospective.

Use when reviewing an evaluation, benchmark, or scoring harness for leakage or contamination. Returns a read-only audit naming detected leakage patterns, where independent ground truth enters or fails to enter, and a fix for each finding.

Use when a concept needs making clear rather than practising: explain it simply, why does this exist, draw it, what is the difference, or simplify it for a beginner. Delivers a one-screen explanation from one chosen angle, grounded in sources. Not for scaffolded practice — use drill.

Use when asked to create a durable local explainer document for a concept, diff reference, idea, or work recap. Classifies the input, gathers grounding material, offers a quiz, drafts and structurally verifies the artifact in a scratch directory, and presents it. Not for one-screen explanations — use explain-concept.

Use when asked to cold-read an artifact with fresh zero-context eyes and cut whatever a stranger cannot follow. Dispatches blind sub-sessions that read only the inline artifact, compares their understanding against a private intent note, and reports severity-ordered defects. Not for criteria-review of requirements docs — use doc-review.

Use when a user wants an answer only from the breaking seat (stakeholders, dependencies, accounting, compatibility under destructive pressure). A breaking-perspective answer is emitted without blending. Not for any other from-* lens seat or multi-lens synthesis.

Use when the user wants an answer only from the business seat, to produce a business-perspective analysis of money, customers, and timing without blending other lenses. Not for any other from-* lens seat or multi-lens synthesis.

Use when a user wants an answer only from the career seat (effects on human trajectories). Emits a career-perspective analysis without blending other lenses. Not for any other from-* lens seat or multi-lens synthesis.

Use when asked to answer only from the codebase seat: what existing code tolerates or punishes. Emits a single-perspective analysis without blending. Not for rebuilding from primitives — use from-first-principle. Read-only; no source or remote mutation.

Use when a user wants an answer only from the human seat: what a person can love, trust, and tolerate. Emits a human-perspective analysis without blending. Not for rebuilding from primitives — use from-first-principle. Read-only; no source or remote mutation.

Use when asked to answer only from the impact seat: who and what actually moves. Returns an impact-perspective analysis without blending. Not for rebuilding from primitives — use from-first-principle. Read-only; no source or remote mutation.

Use when the user wants an answer only from the innovation seat: original technique, talent, and culture. Emits an innovation-perspective analysis without blending. Not for rebuilding from primitives — use from-first-principle. Read-only; no source or remote mutation.

Use when the user wants an answer only from the moat seat: building, keeping, and thickening defensibility. Emits a moat-perspective analysis without blending. Not for rebuilding from primitives — use from-first-principle. Read-only; no source or remote mutation.

Use when the user wants an answer only from the rent-seeking seat: extraction without building. Emits a rent-seeking-perspective analysis without blending. Not for rebuilding from primitives — use from-first-principle. Read-only; no source or remote mutation.

Use when a user wants an answer only from the skeptic seat: cold reasoning without project loyalty. Emits a skeptic-perspective analysis without blending. Not for rebuilding from primitives — use from-first-principle. Read-only; no source or remote mutation.

Use when the user wants an answer only from the stability seat: preservation of the working machine. Emits a stability-perspective analysis without blending. Not for rebuilding from primitives — use from-first-principle. Read-only; no source or remote mutation.

Use when asked to summarize feedback on the active PR. Returns severity-grouped feedback and an action list. Not for finding PRs to review — use gh-review-requests. Not for resolving feedback — use resolve-pr-feedback. Read-only.

Use when the user asks about recent engineering work or what the team is working on, or is preparing planning or roadmap material. Also handles an optional Slack summary when the user explicitly requests it. Not for remote mutation or any irreversible change.

Use when facing a bug, integration failure, dependency issue, unclear API usage, implementation blocker, or tool capability need and the user wants open-source solutions from GitHub. Searches issues, PRs, code, and repositories via `gh`, ranks by problem fit, and returns a reuse/adapt/avoid recommendation with a verification command. Not for source or remote-system changes.

Use when starting a codebase or service, scaffolding, migrating or refactoring, picking a language edition, runtime, framework, or dependency, or when the user asks for the latest, current, LTS, or modern way. Not for remote, credential, publish, deploy, or irreversible changes.

Use when a user resumes work and needs an orientation card without full human recall; returns a small card of shape, invariants, and smell locations. Not for packaging context for another agent or session — use handoff; never source or remote-system changes.

Use when substantive work reaches a session boundary, resumes after interruption, completes, or must move to another session, agent, harness, directory, or repository. Maintains ignored local continuity state and emits a bounded portable brief without launching its receiver.