brain-dev

Ask a real production rootcause brain using `rc ask`, then verify the resulting run. Use inside a brain checkout when asked whether a brain change works on prod infra, when asked to simulate a customer support email, when asked for a direct raw investigation, or when asked to test a pushed `dev/*` brain ref without moving `main`. Captures the answer, run accounting, trace URL, and brain journal diff.

Update or check the installed rootcause brain skills kit and the rootcause `rc` CLI on a local brain checkout. Use when a user asks how to get the newest local-brain-work, brain-ask, brain-git-sync, rc-debug, rc-health, rc-fleet, brain-publish, or other shipped brain skills, asks whether Codex/Claude auto-updates them, asks to run install.sh again, wants to check the latest released kit version, or wants to check/update the rootcause CLI used by the skills.

Run the full local dreamcycle / dream-cycle best-practice flow for a rootcause project from a brain checkout using only the public rc CLI. Use when asked to learn from recent runs, feedback, sent-edit deltas, recurring patterns, journal signal, bad scores, or recurring triage mistakes; progressively inspect evidence with rc, decide whether lessons belong in brain files, persona settings, triage policy, or triage rules, verify with production runs, and publish/sync the result.

Safely reconcile and publish a brain repository's local Git work with origin/main. Use for `$brain dev: git sync`, `$brain-git-sync`, cross-computer Git synchronization, local/remote main divergence, or before brain-publish; do not use brain-publish for a pure Git-sync request.

Run the full local harvest cycle for a rootcause project from a brain checkout using only the public rc CLI — the sent-history analog of brain-dream-cycle. Use when asked to onboard a mailbox's past replies, bootstrap a brain from historical sent mail, harvest/mine/synthesize a mailbox's answered-email corpus, or seed brain knowledge from a production export. Trigger a production harvest export, download the cleaned Markdown corpus, prepare a deterministic opaque-ID manifest with coverage ledger, fan out per-topic subagents to distil patterns (never raw mail), critic the first draft, reduce, decide durable homes across brain files / persona / triage, run a privacy+contract lint, evaluate against held-out threads, gate on operator diff approval, publish, then delete the local corpus.

Publish, ship, deploy, or promote rootcause brain changes after safely reconciling Git with origin/main. Use for `$brain dev: publish`, `$brain-publish`, making brain edits live, exact-SHA server sync, stable/edge promotion, tenant/project publish, actions, or a RootCause support request. Do not use for a pure `$brain dev: git sync` request; use brain-git-sync.

Run a thorough local public-website scout from a rootcause brain checkout, then distil the captured first-party evidence into a progressive-disclosure brain. Use when asked to bootstrap, replace, or substantially improve a project brain from its public website; crawl many site pages; investigate sitemaps, agents.md, llms.txt, UCP, policies, support, or catalog content; or create brain knowledge without using the hosted ten-page website scout.

Local Brain Work for a rootcause project's BRAIN: map a brain checkout, run grounding scripts, run offline/live/docker test tiers, preview tenant projection, test hosted Python actions locally, route broad did-it-work/is-anything-broken prompts to focused rc skills, and check a brain change before pushing. Use inside a rootcause-brain checkout before production-loop validation; no private RootCause source required.

Direct guarded production access from a brain checkout through `rc dev console` and exact source-mirror refresh through `rc dev mirror`. Use when a developer or their coding agent already knows the exact production primitive to run: refresh a pushed source commit, SQL/schema lookup, dry-run-first rare SQL write, cataloged brain script discovery, log command planning, or preflight-first action execution. No rootcause-side LLM; the caller reasons locally and rootcause supplies scoped, audited hands.

Debug one rootcause production run, thread, or session from a brain checkout using public rc commands and the local run-dump renderer. Use when given a run UUID, thread/session id, failed run, fleet/health UUID, delivery question, or request for the full reasoning/tool trace. Analysis-first: inspect evidence, propose the smallest fix, and stop before editing brain files unless the user explicitly asks to implement.

Review recent production runs with `rc fleet runs`, including audits limited to human-scored reviews, discover exact cross-run action history with `rc fleet actions`, then mine systemic failures with `rc fleet patterns`. Use inside a brain checkout for run health, review-score audits, action arguments/statuses/run links over a date window, worst-offender triage, recurring failures, or when no single run is known yet.

Check whether a rootcause project is quietly unhealthy using `rc fleet health` for stale or failed source mirrors plus dead-lettered runs. Use inside a brain checkout for periodic sweeps, CI/cron-style gates, or "is anything broken?" questions. Exits non-zero when unhealthy and points flagged run UUIDs toward `rc-debug`.