Commands
great_cto has 21 slash commands, and a normal day needs three of them. Type them in Claude Code, or use the same three from the terminal — on Claude Code or Codex. On Codex there are no slash commands; the pipeline runs through the CLI (Codex host guide).
Every command has a one-screen card: /help. Old names from before 3.40 still map: /help renamed, or the table below.
Every day
| Command | In the terminal | What you get |
|---|---|---|
/start "…" | great-cto run "…" | A new product or a task in an existing project, taken through the pipeline. Three decisions stay yours: what to build, how, and whether it ships |
/inbox | great-cto status | Only the decisions that need you: open gates, blockers, P0s |
/resume | great-cto resume | Where you left off, continued — only what you already approved; a pending decision still waits |
Add --host codex to the terminal commands to run them on Codex.
/start
Describe the product; the pipeline does the rest and stops three times for you: what gets built (gate:product), how (gate:arch), and whether it ships (gate:ship). One line in PROJECT.md — approval-level: ship-only — keeps only the last.
| Use | Does |
|---|---|
/start "a booking app for a dental practice" | Brief → architecture → plan → build with tests → review → security → deploy |
/start in a repo with code but no great_cto config | Takes the audit path first: detected stack, gaps with file:line evidence, a task per gap, PROJECT.md — then offers to start on your description |
/start audit [focus] · /audit | The audit path on its own, any time (/audit is the same command) |
/resume
Reads the latest notes, open tasks, decisions and git state, and tells you where you left off in one screen. If commits landed since the note, it says so at the top and re-runs only cheap read-only proofs before acting.
/inbox
Open gates, blocked tasks, P0s and pending approvals from the last 24 hours (/inbox 72 for three days) — the list of things that need a human, nothing else.
When you need it
| Command | What you get |
|---|---|
/save | A session note: what was done, how each "done" was verified, where the work stands, the next step |
/digest | What shipped, what broke, what it cost — weekly, per feature |
/review | A code review of a branch with evidence for every finding — or a domain compliance review |
/spec | Discovery → PRD → build spec, before any code |
/poc | A timeboxed yes/no on a risky idea; promote takes a winner through the audits it skipped |
/release | Store notes, a user-facing changelog, stale docs and landing copy flagged |
/trace | Requirement → task → test chain for one item or a whole feature |
/crystallize | This session's lessons, incident patterns and repeated procedures turned into reusable knowledge |
/recall | What this project already knows about a word — sessions and documents |
/sec | Security posture, threat model, SBOM, incident workflow, secret rotation |
/ownership | Who owns a path and who is on call |
/rfc | A cross-team decision proposed, discussed and closed — accepted ones become ADRs |
/exception | A signed, expiring record for a deliberate gate bypass, instead of --no-verify |
/doctor | A health check of great_cto itself; --fix applies the safe fixes |
/board | The local board at localhost:3141: decisions, ledger, fleet, harness |
/agent | An agent managed like an employee: review, evals, prompt evolution, retirement |
/help | The command card |
/save
Writes .great_cto/logs/session-<date>-<slug>.md: what was done, a Verify: command for each done item, decisions, what is pending, and the run state (branch@sha, dirty files, servers still listening). /save commit also commits and pushes the note.
/digest
| Use | Does |
|---|---|
/digest · /digest 30 | Weekly engineering digest: velocity, incidents, tech debt, decisions, open gates, one recommendation |
/digest board | The same as a board report |
/digest cost [7 | feature <slug> | agent <name>] | LLM spend: run-rate, cost per deploy and per shipped feature, router savings |
/digest sessions [30] | How your own sessions spend: long sessions, cache rebuilds, which habits cost most |
/digest slo [service] | SLO burn rate, multi-window — budget exhaustion before it happens |
/digest gov [--since 30d] | Whether the gates work: block rate, overrides, time in gate |
/review
| Use | Does |
|---|---|
/review [branch] | 12 review angles, then skeptical triage (three rounds and an arbiter) so every finding has evidence; opens or closes gate:code |
/review --deep | Triages every P0/P1 angle, not just security and reliability |
/review --domain <name> | The matching compliance reviewer writes a threat model and raises its gate: tax · legal · hr-ai · api · accounting · rcm · msp · procurement · voice |
/spec
/spec discover → /spec prd → /spec (build): a discovery plan, then a PRD, then requirements.md + design.md + tasks.md. /spec retrofit writes the spec for code that already exists.
/poc
/poc "hypothesis" starts a timeboxed proof of concept that skips most of the production pipeline and forces ship / pivot / kill at expiry. /poc decide, /poc extend <days>, /poc status; /poc promote <slug> runs the audits the POC skipped before it can reach production.
/release
/release notes [version] (App Store / Play notes) · changelog [from..to] (user-facing) · docs (stale docs) · sync (landing copy that no longer matches).
/trace
/trace <task-id> shows why a task exists (upstream) and what it affects (downstream); /trace feature <slug> audits a whole feature's chain for gaps.
/crystallize
| Use | Does |
|---|---|
/crystallize learn [focus] | Extracts this session's lessons into .great_cto/lessons.md |
/crystallize · status | Incident patterns ready to promote |
/crystallize approve GP-NNNN · reject · rollback · prune | Promote a pattern into agent improvements — with an eval — or undo it |
/crystallize skill [name] | A procedure you keep walking agents through, captured as a skill |
/recall
/recall <word> searches session history and the project's documents. /recall ccr:<id> brings back the full original of context that compression left out.
/sec
/sec (posture) · threat <slug> (threat model) · sbom · incident "<what>" · rotate (secret rotation).
/ownership
/ownership map | show | set <path> <team> | verify — the ownership matrix, detected from git, with CODEOWNERS generated. /ownership oncall [who | handoff | schedule | escalate] — rotations, shift handoffs, escalation paths.
/rfc
/rfc new "<title>" · list · show <id> · comment <id> "<text>" · close <id> accept|reject — an accepted RFC becomes an ADR.
/exception
/exception create --gate <gate> --reason "<why>" [--days N] · list · check <gate>. The guards refuse --no-verify, --admin and weakened checks; this is the audited way to make a deliberate exception.
/doctor
Pipeline state, missing artefacts, hooks, the last run per agent, permission denials, an outdated PROJECT.md schema — and on Codex, whether great_cto's hooks are installed and reviewed. /doctor --fix applies the safe fixes.
/board
Opens http://localhost:3141 (starts it if needed): Decisions — every gate waiting on you, Ledger — cost and budgets, Fleet — agents, Harness — reviewer scores.
/agent
/agent review [<name>|all] · evals <name> · evolve <name> [--lesson "…"] (a candidate prompt must beat the current one on held-out evals) · retire <name>.
/help
/help — the card · /help renamed — old names · /help <old-name> — where it went.
In the terminal
| Command | Does |
|---|---|
npx great-cto init | Installs great_cto for Claude Code (--host codex for Codex) |
great-cto board | The local board |
great-cto upgrade | Companion plugins and great_cto in Codex; upgrade --self updates the CLI |
great-cto uninstall | Shows what an install wrote; --yes removes it, your data stays |
Renamed in 3.40
| Old | Now |
|---|---|
/audit | /start audit (still works) |
/discover · /prd | /spec discover · /spec prd |
/migrate | /doctor --fix |
/promote | /poc promote |
/oncall | /ownership oncall |
/learn · /skillify | /crystallize learn · /crystallize skill |
/ccr <id> | /recall ccr:<id> |
/cost · /burn · /gov-metrics | /digest cost · /digest slo · /digest gov |
/agent-review · /gen-evals · /prompt-evolve · /agent-retire | /agent review · evals · evolve · retire |
/tax-review /upl-check /aedt-bias-audit /api-contract-review /close-review /coding-audit /msp-review /procurement-review /voice-compliance | /review --domain tax · legal · hr-ai · api · accounting · rcm · msp · procurement · voice |
The generated, exhaustive reference — every mode and flag, read from the command files — is reference/commands.md.