immune-brain

Use to plan an AI/LLM evaluation strategy; advisory planning only, no implementation or QA closure.

Use to investigate a bug or incident and gather evidence before a fix; investigation only, implementation stays with the executor.

Use to review documentation for consistency and accuracy against code changes; read-only advisory, no doc rewrites or workflow closure.

Use for a delegated read-only advisory review through one explicit lens (security, performance, etc.); requires a named lens, makes no edits or closure.

Use to map an unfamiliar repository's architecture before brainstorming or planning; read-only discovery, no implementation edits.

Use to report the deterministic workflow checkpoint snapshot for a validated plan; snapshot only, no execution, QA, review, or compounder authority.

Use to frame and clarify an ambiguous problem and its open questions before planning; framing only, no implementation or plan writing.

Use to review material code or behavior changes before completion; review and same-boundary follow-up handoff only, no direct edits.

Use after work is fully closed and reviewed to capture reusable learnings into docs/solutions/ and refresh memory; not for open or unclosed work.

Use to implement exactly one activated plan step's edits and record execution evidence; no plan changes, no QA closure, no scope expansion.

Use once to bootstrap Immune-Brain project files and directories (memory, specs, plans); not for planning, brainstorming, or code changes.

Use to run a validated plan to completion through autowork checkpoints, review gates, and compounder handoff; does not plan, edit, or make QA decisions directly.

Use when designing or redesigning app pages, screens, dashboards, forms, list/detail views, or frontend layouts before implementation.

Use for a multi-role advisory discussion of design options and trade-offs; discussion only, no implementation or plan writing.

Use to create or revise a spec and iteration plan from requirements; owns scope and plan decomposition, not implementation or step activation.

Use to repair PR review feedback and CI failures within the current PR scope; routes to code review and QA, not new planning.

Use to adversarially pressure-test scope, risks, and assumptions before planning; advisory only, does not own or write the plan.

Use to judge whether the active step closes from recorded evidence and record pass/rework/replan; judgment only, no edits.

Use to review UI, visual, accessibility, responsive, or design-contract changes; advisory only, no implementation or QA closure.

Use to drive the single active step of a validated plan to its next boundary; coordinates execution, QA, and handoff but does not write plans (imm-planner) or issue QA pass (imm-qa).

Use to coordinate brainstorm and planning into a validated plan before running; coordination only, no implementation edits.

Use to review prompt and tool/agent-contract changes; read-only advisory, no edits or workflow closure.

Use for an advisory release-readiness check before shipping; advisory only, no deployment authority.

Use only to fix explicitly delegated test files for an active step; no source-code or workflow-state changes.