Khaos-Brain

Run one repository-managed automatic bounded KB Dream evidence pass. Use only for explicit Dream maintenance or the scheduled KB Dream automation, not Sleep consolidation, ordinary retrieval, or trusted-card maintenance.

Run the repository-managed Khaos Brain organization contribution pass. Use only when a user or automation explicitly asks to export local shareable KB cards into a validated organization repository; no-op in personal mode or unvalidated organization settings.

Run the repository-managed Khaos Brain organization maintenance pass. Use only when a user or automation explicitly asks to inspect, review, or maintain a validated organization KB repository and this machine has opted into organization maintenance; this is the organization-level Sleep-like maintenance flow, not ordinary local KB Sleep.

Run the repository-managed automatic incremental KB Sleep pass. Use only for explicit Sleep maintenance or the scheduled KB Sleep automation, not ordinary retrieval or active-task write-back.

Apply an explicitly prepared, recovery-oriented Chaos Brain software update. Use for a user/UI-authorized update or the repository-managed system update automation; never decide on the user's behalf that an unprepared update is wanted.

Use only when explicitly requested or delegated by flowguard-development-process-flow's agent_workflow mode to rehearse a multi-skill, tool, plugin, or external-action workflow from a fresh inventory; generic workflow planning enters flowguard-development-process-flow first.

Use when an existing FlowGuard model and code map should drive behavior-preserving architecture contraction, including merging duplicate handlers or modules, collapsing adapters, removing dead branches, or preparing a StructureMesh refactor.

Use for external behavior registration, bidirectional source coverage, exactly one primary owner model, change-mode accounting, internal Primary Path Authority handoff, or broad done/release/archive/publish confidence.

Use when a FlowGuard model should drive pre-code module and function boundaries, FunctionBlock ownership, field/state/side-effect owners, facade design, adapter boundaries, or validation structure before production edits.

Use to generate canonical bad cases, Cartesian combinations, oracles, coverage shards, and receipts from a declared finite boundary, family seed, payload, transition, model mesh, coverage universe, or observed-problem backfeed.

Use for staged development, multi-skill work, freshness, sync, release, or final process claims.

Use before non-trivial discussion, proposal, feature, bug, refactor, UI, test, prompt, skill, workflow, or process work in an existing modeled system to identify current ownership and duplicate-boundary risk.

Use when a change adds, removes, renames, migrates, replaces, externalizes, preserves, or audits fields, schema keys, config flags, prompt fields, payload columns, persisted attributes, defaults, aliases, or fallbacks.

Use when a FlowGuard project has three or more local models, an oversized model, stale child evidence, parent/child partitioning, target split derivation, child reattachment, affected siblings, or whole-flow mesh closure risk.

Use when post-FlowGuard runtime, test, replay, log, manual, production, or UI evidence exposes a missed behavior class, or a repair needs generalized closure.

Use when model obligations, owner external CodeContracts, source audits, transition cells, boundary observations, payload cases, closure targets, or ordinary test evidence need direct current comparison.

Use when a locally green FlowGuard model needs topology-grounded future-use hazard review for broad claims, business paths, old/new disposition, side effects, terminals, loops, external boundaries, or parent/child compression.

Use only when explicitly requested or delegated by flowguard-development-process-flow's plan_detailing mode to compile a rough non-trivial plan into structured artifacts, steps, receipts, validations, failure/rework branches, and claim boundaries.

Use when an existing large script, module, package, command, public API, facade, config surface, or plugin entrypoint split needs model-derived ownership, dependency, compatibility, parity, and release gates.

Use when tests, checks, transition cells, payload cases, or evidence are large, slow, layered, stale, skipped, backgrounded, release-only, or require parent/child ownership and freshness proof.

Use for UI content admission, observed surface, behavior, structure/text, operability, evidence, or transitions.

Open the local Khaos Brain desktop card browser for human review. Use when the user asks to open, show, view, inspect, or create a human-facing entry for the Khaos Brain UI, desktop app, card browser, or Windows shortcut. Do not use this for AI KB retrieval or feedback; use the predictive KB retrieval workflow for that.

Retrieve relevant entries from the local predictive knowledge base as a lightweight preflight for repository work. Use route-first retrieval: infer the task direction, then search by domain path and cross-index before relying on flat keyword matching. In Codex, prefer a scout sidecar before non-trivial work and a recorder sidecar after work when sub-agents are available. Treat "no relevant entry" as useful signal, not a reason to skip the scan up front.

FlowGuard kernel for ordinary behavior/state modeling, unclear route selection, cross-route coordination, and kernel-owned internal gates. Use when no direct FlowGuard satellite clearly owns the risk or several routes must be coordinated; use the matching satellite directly when ownership is clear.

Review Khaos Brain organization KB maintenance proposals, including card-and-Skill bundles, shared Skill safety, evidence quality, privacy boundaries, and GitHub auto-merge readiness. Use as a review lens for approving, rejecting, merging, splitting, or promoting organization KB candidates or organization Skill registry changes.