claude-sdk-rs: Command for Claude Code

Backlog Ticket — Capture a queued idea into the HQ backlog

Variables

$ARGUMENTS — free-form description of the idea, improvement, or research thread. Can include a target repo, type hint, or just a raw description.

Execution Model

Run entirely inline. Spawn no subagent. This is a single-file append plus a state.json node write — small enough that a subagent round trip adds latency without adding value.

Instructions

1. If $ARGUMENTS is not provided, stop and ask the user to describe the idea.

2. Find the HQ brain root. From the current working directory, walk up parent by parent looking for a brain.toml file (its first line begins # brain.toml). The directory containing it is BRAIN_ROOT. If not found, abort and explain this command requires the HQ brain context.

3. Read $BRAIN_ROOT/core/planning/backlog.md to understand the current format and check for near-duplicates. This command ALWAYS writes to the HQ backlog, never a sub-brain or project repo backlog.

4. From $ARGUMENTS infer the following fields:

   title — a concise, action-oriented title (5–10 words)

   repo — the primary repo/area this belongs to. Pick one: base-template · orchestrator · learn-ai · bastion · mev · brain · business · cross-repo If the description clearly targets multiple repos and neither dominates, use cross-repo.

   type — what kind of work this is:

   • feature — new capability or user-facing addition
   • improvement — refine, fix, or simplify something existing
   • research — investigation or scoping pass before building
   • content — writing, course material, or learning content
   • business — CV, outreach, rates, career, client ops
   • planning-session — needs a design/planning pass before implementation work can begin

   status — default to idea unless the description implies it's already scoped and ready (ready)

   related — optional; any decisions (D##), master-plan blocks, or doc paths mentioned or clearly implied by the description. Omit if nothing obvious.