30 commands · Claude Code · Codex CLI · Cursor IDE · Pi

Which command helps me right now?

Pick what you want to do. Each command below says when to use it, what you get, how to type it and what happens next.

I want to …

Clarify an idea/brainstorm, /grill
Start a project/bootstrap, /plan new
Plan a feature/plan feature
Build/session, /go, /close
Maintain/session housekeeping
Investigate/discovery, /debug
Test/test, /ux-grill
Look back/eval, /plan retro
Learn/evolve, /reconcile
Run many repos/portfolio, /dispatcher

Rule of thumb

Which session to start
Do you know exactly what to build?	`/session feature`
Not sure what to build yet?	`/plan feature` first
Big or unclear change?	`/session deep`
Tidy up?	`/session housekeeping`

How to type a command in your coding tool

How each coding tool runs a command
Claude Code	Type `/name` in the chat. In headless `claude -p`, use the long form, for example `/session-orchestrator:session deep`.
Codex CLI	Pick the skill in the Codex chat: `$session-orchestrator:name`. It is not a shell command. Plain `/go` and `/close` do not work there, and Codex never starts `go` or `close` on its own.
Cursor IDE	Type `/name`. Agents run one after another.
Pi	Type `/name`. Each command is a prompt template.

Start here

Six commands are enough for your first sessions. Learn these and ignore the rest until you need them.

`/bootstrap` Start here

When

You use session-orchestrator in a repo for the first time.

You get

A CLAUDE.md (AGENTS.md on Codex and Pi) with a Session Config and the minimum repo structure, committed. You see the created files and the commit hash.

Type it

Claude Code: /bootstrap [--upgrade <tier>]
Codex: $session-orchestrator:bootstrap
Cursor: /bootstrap
Pi: /bootstrap

Touches

Commits asks you first 7 to 9 setup questions; /bootstrap --fast --no-interview skips them and uses sensible defaults

Needs

A git repository.

Then

Run /session feature for your first session.

Source: skills/bootstrap/SKILL.md

`/session` Start here

When

You sit down to work on a repo and want to know what to do next.

You get

It reads your repository (git state, issues, CI, environment) and proposes what to work on. Nothing changes until you agree.

Modes

housekeeping tidy up and maintain
feature build one clear feature (best for a first session)
deep big or unclear change: look first, then build (the default)
ultradeep an optional, longer profile of deep for a named research question; not a fourth type

Type it

Claude Code: /session [housekeeping|feature|deep|ultradeep] [task]
Codex: $session-orchestrator:session deep
Cursor: /session deep
Pi: /session

Touches

Changes files asks you first only session state (STATE.md, session lock); code waits for /go

Needs

/bootstrap done. With no session history, a bare /session first asks you for task and mode.

Then

Agree to the plan with /go.

Source: commands/session.md

`/go` Start here

When

You read the session plan and agree with it.

You get

The agreed work, done in rounds (waves). Each round is checked with your tests, and you get a progress summary.

Type it

Claude Code: /go [instructions]
Codex: $session-orchestrator:go
Cursor: /go
Pi: /go

Touches

Changes files edits code per the plan; with no plan it does nothing

Needs

A plan from /session in this conversation, or this session's own paused STATE.md.

Then

Run /close when the rounds are done.

Source: skills/go/SKILL.md

`/close` Start here

When

The work of this session is done and you want to wrap it up.

You get

The full check, a clean commit, issues for what is left, and a session summary.

Type it

Claude Code: /close
Codex: $session-orchestrator:close
Cursor: /close
Pi: /close

Touches

Commits also opens issues and pushes if configured; refuses when STATE.md belongs to another session

Needs

An active session of your own. Arguments are ignored.

Then

Start the next session with /session.

Source: skills/close/SKILL.md

`/plan` Start here

When

You want to plan a new project or a feature, or look back at a finished phase.

You get

A researched question round, then a PRD or retro document and prioritized issues.

Modes

new a new project: also creates the repo and commits
feature one feature: PRD and issues
retro look back at a finished phase

Type it

Claude Code: /plan [new|feature|retro]
Codex: $session-orchestrator:plan feature
Cursor: /plan feature
Pi: /plan

Touches

Opens issues asks you first writes the PRD or retro; new also creates a repo and commits

Needs

/bootstrap done. /plan new also needs plan-baseline-path in your Session Config; feature and retro run without it.

Then

Optionally /grill the PRD, then /session feature to build it.

Source: skills/plan/SKILL.md

`/evolve` Start here

When

After a few sessions, you want to learn from what went well or badly.

You get

Suggested lessons from your past sessions. You decide which ones to keep.

Modes

analyze suggest lessons (default)
review edit stored lessons
list show stored lessons
dialectic update USER.md and AGENT.md, only with --apply

Type it

Claude Code: /evolve analyze
Codex: $session-orchestrator:evolve analyze
Cursor: /evolve analyze
Pi: /evolve

Touches

Changes files asks you first stores a lesson only after you confirm it

Needs

/bootstrap done, persistence: true in your Session Config, some session history.

Then

Run /reconcile to turn lessons into rules.

Source: skills/evolve/SKILL.md

Clarify an idea

Turn a fuzzy idea into something you can plan.

`/brainstorm`

When

You have a feature idea, but its scope or UX is still fuzzy.

You get

A short question dialogue (3 to 5 rounds) and an approved design spec in docs/specs/.

Type it

Claude Code: /brainstorm [topic-slug]
Codex: $session-orchestrator:brainstorm export-to-csv
Cursor: /brainstorm export-to-csv
Pi: /brainstorm

Touches

Changes files asks you first writes only the spec; no code work until you approve the design

Needs

/bootstrap done.

Then

/plan feature with the spec, or /write-executable-plan.

Source: skills/brainstorm/SKILL.md

`/grill`

When

A plan or PRD feels settled and you want it attacked before anyone builds it.

You get

Hard questions, one decision at a time, grounded in your code. Contradictions and weak assumptions show up.

Type it

Claude Code: /grill [file-or-topic]
Codex: $session-orchestrator:grill docs/prd/<file>.md
Cursor: /grill <file>
Pi: /grill

Touches

Changes files asks you first mostly conversation; optionally writes a grill note in docs/specs/

Needs

A plan in the conversation, or a file to grill.

Then

/plan feature to write down the hardened plan.

Source: skills/grill/SKILL.md

Start a project

Set up a repo, check the setup, or move a product into its own repo.

Also: /bootstrap and /plan new in Start here.

`/spinout`

When

A product or a cleaned copy should move out of this repo into its own repo.

You get

A guided extraction in 5 steps: target, sanitize check, copy with fresh git, freeze marker, remotes.

Type it

Claude Code: /spinout [--type venture|snapshot] [--dry-run]
Codex: $session-orchestrator:spinout
Cursor: /spinout
Pi: /spinout

Touches

Commits asks you first creates a new repo and leaves a freeze marker in the source; --dry-run only prints the plan

Needs

/bootstrap done. Know the destination path and account.

Then

Run /bootstrap in the new repo.

Source: skills/spinout/SKILL.md

`/harness-audit`

When

You want to check that session-orchestrator is set up correctly in this repo.

You get

A score (healthy, warn or critical) and every category below 7 of 10.

Type it

Claude Code: /harness-audit
Codex: $session-orchestrator:harness-audit
Cursor: /harness-audit
Pi: /harness-audit

Touches

Changes files appends one record to .orchestrator/metrics/audit.jsonl

Needs

Nothing. Arguments are ignored.

Then

Fix weak categories, for example with /bootstrap --upgrade.

Source: skills/harness-audit/SKILL.md

Plan a feature

Write down what to build before it is built.

Also: /plan feature in Start here.

`/write-executable-plan`

When

You have an approved PRD or design spec and want steps any agent can follow.

You get

A plan in docs/plans/ with the files per task, complete code and exact check commands.

Type it

Claude Code: /write-executable-plan
Codex: $session-orchestrator:write-executable-plan
Cursor: /write-executable-plan
Pi: /write-executable-plan

Touches

Changes files one plan file; rejects placeholders such as TBD or TODO

Needs

/bootstrap done, and an approved PRD (from /plan feature) or spec (from /brainstorm).

Then

Build it in /session.

Source: skills/write-executable-plan/SKILL.md

Build

Do the work in checked rounds.

Most of the time: /session, /go, /close in Start here.

`/autopilot`

When

You want several full sessions to run on their own while you are away.

You get

Repeated start, plan, build and close loops. 10 kill-switches stop it; every run is logged in autopilot.jsonl.

Type it

Claude Code: /autopilot [--headless] [--max-sessions=N] [--max-hours=H] [--dry-run]
Codex: $session-orchestrator:autopilot
Cursor: /autopilot
Pi: /autopilot

Touches

Commits edits and commits like /go and /close; --dry-run does not execute

Needs

Start it at the top level, never inside a running session. --headless needs claude -p login.

Then

Read autopilot.jsonl and the session summaries.

Source: skills/autopilot/SKILL.md

Maintain

Keep the code, the memory and the setup in shape.

Also: /session housekeeping for a short tidy-up pass.

`/architecture`

When

The code feels tangled and you want ideas that make it easier to test and navigate.

You get

Proposed refactors that turn shallow modules into deep ones.

Type it

Claude Code: /architecture
Codex: $session-orchestrator:architecture
Cursor: /architecture
Pi: /architecture

Touches

Reads only proposes, edits nothing

Needs

Works best with CONTEXT.md and docs/adr/. Without CONTEXT.md it first asks about your domain.

Then

Turn a proposal into /plan feature or a session task.

Source: skills/architecture/SKILL.md

`/memory-cleanup`

When

Claude's memory notes for this repo are stale, contradictory or too long.

You get

A cleaned memory folder with MEMORY.md under 200 lines, or a proposal to apply later.

Type it

Claude Code: /memory-cleanup [--dry-run | --apply-pending]
Codex: $session-orchestrator:memory-cleanup
Cursor: /memory-cleanup
Pi: /memory-cleanup

Touches

Changes files asks you first edits ~/.claude/projects/<repo>/memory/; --dry-run only writes a proposal

Needs

An existing memory folder.

Then

Keep working. Repeat every 5 or more sessions.

Source: skills/memory-cleanup/SKILL.md

`/repo-audit`

When

You want to know whether a repo meets the baseline: config, CI, tests, security, docs.

You get

A checklist report over 9 categories: pass, fail or warn.

Type it

Claude Code: /repo-audit
Codex: $session-orchestrator:repo-audit
Cursor: /repo-audit
Pi: /repo-audit

Touches

Changes files report only, never fixes; writes a JSON file to .orchestrator/metrics/

Needs

Nothing. Arguments are ignored.

Then

Fix failing categories in /session housekeeping.

Source: skills/repo-audit/SKILL.md

`/sunset-review`

When

Once a quarter, you want to find skills, agents or commands nobody uses any more.

You get

Every item sorted into Active, Investigate, Demote or Retire.

Type it

Claude Code: /sunset-review [--kind skill|agent|command] [--window-days N]
Codex: $session-orchestrator:sunset-review
Cursor: /sunset-review
Pi: /sunset-review

Touches

Changes files writes a report; never deletes anything

Needs

Usage records (default window 90 days).

Then

Decide on the Retire and Demote candidates yourself.

Source: skills/sunset-review/SKILL.md

Investigate

Find problems and their causes before you fix anything.

`/discovery`

When

You want a systematic hunt for quality problems and tech debt.

You get

A list of findings you go through one by one. Confirmed ones become issues.

Type it

Claude Code: /discovery [all|code|infra|ui|arch|session|audit|vault|feature]
Codex: $session-orchestrator:discovery code
Cursor: /discovery code
Pi: /discovery

Touches

Opens issues asks you first looks without changing; opens issues only for findings you confirm

Needs

/bootstrap done. --since and --full cannot be combined.

Then

Pick the new issues up in /session.

Source: skills/discovery/SKILL.md

`/debug`

When

A test fails, the build breaks, or something behaves unexpectedly.

You get

A root-cause investigation in 4 steps, written down before any fix is proposed.

Type it

Claude Code: /debug [bug-or-issue]
Codex: $session-orchestrator:debug "test timeout in CI"
Cursor: /debug <bug>
Pi: /debug

Touches

Changes files asks you first writes a debug note in .orchestrator/debug/ first

Needs

/bootstrap done. Without an argument it picks a likely bug and confirms it with you.

Then

Fix it in a session, pointing to the debug note.

Source: skills/debug/SKILL.md

Test

Check that the app works and that your tests are worth keeping.

`/test`

When

You want to check that a web or macOS app really works for a user.

You get

Driven test runs with screenshots, a UX verdict, and a report matched against your open issues.

Type it

Claude Code: /test [scope|profile] [--dry-run]
Codex: $session-orchestrator:test smoke
Cursor: /test smoke
Pi: /test

Touches

Opens issues writes test artifacts and a report; --dry-run only prints the plan

Needs

A runnable target: Playwright for web, Peekaboo for macOS.

Then

Fix findings in a session or with /debug.

Source: skills/test/SKILL.md

`/test-audit`

When

You wonder whether a test is worth writing, or your test suite has grown too big.

You get

A verdict on a new test (gate), or a smaller suite where every delete or merge carries evidence.

Type it

Claude Code: /test-audit [gate|audit|campaign] [scope]
Codex: $session-orchestrator:test-audit audit tests/lib
Cursor: /test-audit audit <scope>
Pi: /test-audit

Touches

Changes files audit and campaign change tests under a coverage guard; gate only judges

Needs

A test suite. A missing scope is asked once.

Then

Run your full tests and commit with /close.

Source: skills/test-audit/SKILL.md

`/ux-grill`

When

A web app runs locally and you want its UX checked the same way every time.

You get

Measured checks per route and screen size, then a walk through each user journey with screenshots.

Type it

Claude Code: /ux-grill [loopback-url | manifest-path]
Codex: $session-orchestrator:ux-grill http://127.0.0.1:3100
Cursor: /ux-grill <url>
Pi: /ux-grill

Touches

Opens issues asks you first writes measurements; never commits or edits product code

Needs

The app running on a local URL, and agent-browser installed.

Then

Fix the issues, then run it again to compare.

Source: skills/ux-grill/SKILL.md

Look back

See how a session or a phase really went.

Also: /plan retro for a finished phase.

`/eval`

When

You want an honest score of how well the last session followed its process.

You get

A score per dimension against a fixed rubric (no single total), and an optional HTML report.

Type it

Claude Code: /eval [--session <id>] [--no-write]
Codex: $session-orchestrator:eval
Cursor: /eval
Pi: /eval

Touches

Changes files appends to .orchestrator/metrics/eval.jsonl unless --no-write

Needs

At least one finished session.

Then

Run /evolve to learn from the weak dimensions.

Source: skills/eval/SKILL.md

Learn

Turn lessons from past sessions into rules your agents read.

First: /evolve in Start here.

`/reconcile`

When

You have approved lessons and want the strong ones to become rules.

You get

Proposed rules in .claude/rules/, four at a time. Your agents read them next time.

Type it

Claude Code: /reconcile [--dry-run]
Codex: $session-orchestrator:reconcile
Cursor: /reconcile
Pi: /reconcile

Touches

Changes files asks you first writes only the rules you approve; --dry-run writes nothing

Needs

Stored lessons from /evolve.

Then

Commit the new rules with /close.

Source: skills/reconcile/SKILL.md

Run many repos

See all your repos at once and pick the next one.

`/portfolio`

When

You run many repos and want one overview of their issues, MRs and staleness.

You get

One dashboard file in your vault: 01-projects/_PORTFOLIO.md.

Type it

Claude Code: /portfolio [--dry-run] [--repo <name>]
Codex: $session-orchestrator:portfolio
Cursor: /portfolio
Pi: /portfolio

Touches

Changes files writes the dashboard; --dry-run prints the diff only

Needs

gitlab-portfolio.enabled: true and a vault folder in your Session Config; glab or gh logged in.

Then

Pick a repo, or ask /dispatcher.

Source: skills/portfolio/SKILL.md

`/dispatcher`

When

One repo is done and you want to know which repo is most worth working on next.

You get

A ranking of free repos and one recommendation. Once you confirm, it claims that repo and hands over to /session.

Type it

Claude Code: /dispatcher [--dry-run] [--repo <name>]
Codex: $session-orchestrator:dispatcher
Cursor: /dispatcher
Pi: /dispatcher

Touches

Changes files asks you first reads only until you confirm; then it takes that repo's session lock

Needs

Several repos under one folder.

Then

Run the suggested /session in the chosen repo.

Source: skills/dispatcher/SKILL.md

More tools

Smaller helpers for special cases.

`/persona-panel`

When

You want a text or file reviewed by several expert or buyer personas at once.

You get

One verdict per persona, merged into a final verdict.

Type it

Claude Code: /persona-panel <target> [--mode voting|hard-gate|summary]
Codex: $session-orchestrator:persona-panel docs/landing.md
Cursor: /persona-panel <target>
Pi: /persona-panel

Touches

Changes files writes a result file to .orchestrator/persona-panel/; --dry-run writes nothing

Needs

A target inside the project, and persona files in .claude/personas/.

Then

Revise the target based on the verdicts.

Source: skills/persona-panel/SKILL.md

`/eli5`

When

The last answer was too technical or too long.

You get

The same facts in plain words, in your language, with nothing dropped.

Type it

Claude Code: /eli5 [topic]
Codex: $session-orchestrator:eli5
Cursor: /eli5
Pi: /eli5

Touches

Reads only only rewrites text in the chat

Needs

Nothing.

Then

Act on the now clear answer.

Source: skills/eli5/SKILL.md

`/release`

When

You maintain this package and want to cut a new version.

You get

The release steps in their required order, with the conditions that stop a release.

Type it

Claude Code: /release [X.Y.Z]
Codex: $session-orchestrator:release 5.14.0
Cursor: /release
Pi: /release

Touches

Publishes publishes to npm; that cannot be undone

Needs

A clean main branch level with origin, a current GitHub mirror, a CHANGELOG entry, and all checks of release.mjs --check green.

Then

Verify with npm view session-orchestrator versions.

Source: skills/release/SKILL.md

`/tmux-layout`

When

During a long session you want STATE.md, CI and events visible side by side.

You get

A tmux layout with 4 panes to watch. The chat stays in your terminal.

Type it

Claude Code: /tmux-layout [--layout debug]
Codex: $session-orchestrator:tmux-layout
Cursor: /tmux-layout
Pi: /tmux-layout

Touches

Reads only opens panes for watching only

Needs

tmux installed. This skill is opt-in (status: proposed).

Then

Keep working in the chat.

Source: skills/tmux-layout/SKILL.md

`/templates-ack`

When

The templates-first check blocks a gh or glab create, but you already read the template.

You get

Issue or MR creation passes for the rest of this session.

Type it

Claude Code: /templates-ack [reason]
Codex: $session-orchestrator:templates-ack (has no effect on Codex)
Cursor: /templates-ack
Pi: /templates-ack

Touches

Changes files writes one acknowledgement file in .orchestrator/runtime/

Needs

A Claude Code session. Without one it is refused.

Then

Run the gh or glab create command again.

Source: commands/templates-ack.md

Happens in the background

You never type these. They run when a command needs them.

23 internal skills

Building blocks the commands above call, for example session-start for /session and wave-executor for /go.

claude-md-drift-check
convergence-monitoring
docs-orchestrator
ecosystem-health
frontmatter-guard
gitlab-ops
gitlab-portfolio
hook-development
mcp-builder
mode-selector
npm-publish
peekaboo-driver
playwright-driver
quality-gates
remote-offload
session-end
session-plan
session-start
test-runner
using-orchestrator
vault-mirror
vault-sync
wave-executor

14 agent roles

The agents a wave sends out: each role does one kind of work, such as writing code, writing tests or reviewing.

analyst
architect-reviewer
code-implementer
db-specialist
dialectic-deriver
docs-writer
eval-judge
qa-strategist
security-reviewer
session-reviewer
skill-applied-judge
test-writer
ui-developer
ux-evaluator

10 hook events

Moments in Claude Code where the guards and the session records run. Codex uses six of them; Pi maps five through a bridge.

SessionStart
SessionEnd
PreToolUse
PostToolUse
Stop
SubagentStop
PostToolUseFailure
PostToolBatch
SubagentStart
CwdChanged

2 MCP tools

Optional tools your coding tool can call. They are not slash commands.

session_config: Reads the Session Config from CLAUDE.md or AGENTS.md.
session_metrics: Reads the last five session records from .orchestrator/metrics/sessions.jsonl.

Which command helps me right now?

I want to …

How to type a command in your coding tool

Start here

/bootstrap Start here

/session Start here

/go Start here

/close Start here

/plan Start here

/evolve Start here

Clarify an idea

Start a project

Plan a feature

Build

Maintain

Investigate

Test

Look back

Learn

Run many repos

More tools

Happens in the background

`/bootstrap` Start here

`/session` Start here

`/go` Start here

`/close` Start here

`/plan` Start here

`/evolve` Start here