Work — structured AI coding work

Activation

Invoked via /work. Brings together Brain overview, Repo Passport, Context Governor, Change Receipts, and AI Mistake Firewall before any code work begins.

Supported forms

/work <task>
/work continue
/work status
/work context

Behavior by subcommand

/work <task>

Run this sequence before starting any work:

A. Brain overview

lstack brain overview

Show compact summary. Note any active Decisions, Failed Attempts, or warnings.

B. Claude context

lstack brain context --for claude

Show compact context. Note any active Task Contracts, protected files, or generated folders.

C. Check current receipt

lstack brain receipt status

D. Receipt handling

If no receipt is open:

lstack brain receipt start --title "<short title>" --goal "<full task>"

Derive a short title (≤60 chars) from the task. Preserve the user's full task as the goal.

If a receipt is already open, show it and use AskUserQuestion to ask:

• This task belongs to the open receipt — continue
• This is a new task — finalize the open receipt first
• This is a new task — abandon the open receipt first

Do not start another receipt silently.

E. AI Mistake Firewall precheck

lstack brain firewall status

If the task includes specific commands, files, or paths, check them:

lstack brain firewall check --command "<command>" --path "<path>" --changed-file "<file>"

Do not execute the checked command. Treat firewall warnings as guidance unless the user explicitly requested strict mode.

F. Work summary before starting

Before proceeding with actual work, show a compact constraint summary:

• Platform path rule (from passport/context)
• Open receipt id
• Active Task Contract if any
• Protected files or deny-listed paths
• Generated folders to avoid editing
• Active Decisions
• Relevant Failed Attempts
• AI Mistake Firewall status

Then proceed with the requested work, keeping the open receipt in scope.

/work continue

lstack brain overview
lstack brain receipt status
lstack brain context --for claude

If a receipt is open, continue working inside it. If none is open, use AskUserQuestion to ask whether to start one.

/work status

lstack brain overview
lstack brain receipt status
lstack brain doctor

Report status only. Do not start a receipt.

/work context

lstack brain context --for claude

Optionally follow with:

lstack brain overview

Report context only. Do not start a receipt.

Constraints

• Never bypass receipts: always check receipt status before work.
• Never hide firewall or doctor warnings.
• Never start a receipt when the user only asked for status or context.
• Never run tests until finalization or explicit user request.
• Never invoke Claude recursively or spawn subprocesses via AI CLI flags.
• Never run destructive or state-mutating git commands (reset, clean, restore, push, pull, commit).
• The CLI is the source of truth. Do not duplicate Brain storage logic.
• If a CLI command fails, show the error and tell the user what to run manually.