Commands

Commands are skills in .claude/skills/. Run any command inside your AI tool with /command-name. This page covers the most used ones; HQ ships more than 100.

Session Commands

Command	Description
/startwork	Resolve current HQ context and surface useful next work options
/checkpoint	Save a lightweight session checkpoint to workspace/threads/
/handoff	Save thread and prepare for session continuation
/learn	Turn reusable findings into scoped policies or insight files. /learn --hard records a hard-enforcement rule
/conduct	Orchestrator mode: coordinate only, sending small tasks to subagents and worker areas to standing HQ lanes

/checkpoint

Saves git state, active files, and a summary. Use when you want to bookmark progress. When Claude Code autocompacts the conversation, a hook tells the agent to run /checkpoint on its next turn.

/handoff

Writes a thread at workspace/threads/ and a handoff.json with the context needed to resume in a fresh session. Run it before you end a session that has work in progress.

Deficiency report

Each handoff also lists problems with the HQ setup that slowed the session down. The agent answers a fixed checklist of seven areas from memory: finding files, automated checks, coding conventions, injected context, tool cost, steps that did nothing, and access to information. Each problem it records needs concrete evidence from the session. Items without evidence are dropped.

Every problem gets one of three kinds, and each kind goes to a different place:

• local-setup: problems with your own machine or account. The handoff summary lists them under Your setup, with the command to fix each one when there is one.
• repo: problems in the code repo you were working in. These are saved as repo-scoped learnings.
• hq-core: problems in HQ itself. On a standard install, HQ files these with the HQ team as one feedback report per handoff.

The feedback report contains only the hq-core items (area, severity, summary, evidence and proposed fix) and a short system_info block. It never includes log files. Before it is sent, the report is checked for secrets, and a match stops the submit. An item that was already reported in the last 7 days is not sent again.

Turning off automatic reports

Set HQ_HANDOFF_AUTOFILE=0 to stop handoffs from filing reports:

export HQ_HANDOFF_AUTOFILE=0

The problems are still saved in the handoff thread and shown in the summary. Nothing is sent to the HQ team.

What system_info contains

Field	Value
hq_version	The HQ version from core/core.yaml
hq_cli_version	The installed hq CLI version, if any
host	cli, desktop or sdk
os	The operating system name, for example darwin or linux
hooks_enforced	true when HQ hooks are confirmed to run in this session

system_info contains no file paths, email addresses or tokens.

Worker Commands

Command	Description
/run {worker} {skill}	Execute a worker’s skill
/newworker	Scaffold a new worker with skills, tools, and knowledge

/run

The primary way to invoke workers. Examples:

/run task-executor execute
/run architect system-design
/run content-brand voice-analysis
/run code-reviewer review-pr

/newworker

Interactive wizard to create a custom worker:

1. Name, type, and description
2. Skill definitions
3. Knowledge context
4. Execution rules
5. Generates worker.yaml + skill files

Project Commands

Command	Description
/brainstorm "idea"	Explore approaches and decide whether to plan
/prd "description"	Create a Project Requirements Document
/deep-plan "description"	Plan a large project with research and a longer interview
/grill-me "topic"	Run a decision interview on any topic
/run-project	Execute a project’s PRD stories through HQ worker lanes
/execute-task	Execute one PRD story through coordinated worker phases with test, lint, and typecheck gates

/brainstorm

Compares approaches and surfaces unknowns through a short interview. The interview runs on the shared HQ interview engine, the same one /prd and /grill-me use. At the end, /brainstorm asks what to do next:

• Plan it now: continues into /prd in the same session and writes prd.json. Your brainstorm answers carry over, so you are not asked them again.
• Upgrade to deep-plan: hands off to /deep-plan for research and a longer interview.
• Stop here: keeps the brainstorm and ends.

/prd

Runs its interview on the shared interview engine. If the project has a brainstorm, /prd loads those answers as known facts and skips questions they already settle. It records how many questions were asked, answered, and assumed in metadata.interview in prd.json. It generates a structured prd.json with:

• Project metadata (name, repo, branch)
• User stories with acceptance criteria
• Task breakdown
• Architecture decisions

/grill-me

Interviews you about any topic, one question at a time, and saves a summary of your decisions. Pass --rounds to batch questions.

/run-project

Orchestrates multi-task execution using the Ralph methodology:

1. Load PRD
2. Pick next story
3. Route to workers (architect → dev → QA → review)
4. Validate completion
5. Capture learnings
6. Repeat

System Commands

Command	Description
/search "query"	Search HQ content and indexed repos with qmd, falling back to grep
/cleanup	Audit and clean HQ structures and indexes. /cleanup --reindex rebuilds all INDEX.md files
/setup	Interactive setup wizard for new installations
/update-hq	Upgrade HQ from the latest hq-core release

Cloud and Team Commands

Command	Description
/hq-sync	Run a full bidirectional sync for cloud-backed HQ companies
/hq-files	Inspect or change HQ vault ACLs, grants, and shared prefixes
/hq-share	Share HQ vault paths through single-use links or direct ACL grants
/hq-secrets	Use HQ CLI secrets safely through env injection or one-off exec
/deploy	Deploy or share generated HQ artifacts through hq-deploy
/new-hire	Onboard a teammate end to end: invite, role, groups, secrets, file access, and onboarding packet
/delegate	Hand a project to a person or fleet agent: grants, dossier, ownership transfer, and a pickup DM