/new-track

New Track

Create a track. A track owns a durable outcome, not a tiny implementation step.

Pre-flight (one check)

If conductor/workflow.md doesn't exist: tell user to run /conductor:setup first. Otherwise proceed.

1. Pick the mode

Tracks come in five modes. Pick one before anything else — verification weight, expected workflow, and required artifacts all key off it.

Mode	Use for
Implementation	Code, schema, API, UI, behavior, infrastructure, automation changes
Audit / Research	Investigation, data review, source-truth comparison, system mapping
Decision / Planning	Architecture choices, sequencing, scope/ownership, tradeoff-heavy work
Execution / Batch Ops	Migrations, data writes, imports, backfills, bulk edits
Publishing / PR	Preparing/pushing/opening work for review (often a sub-phase, not a track)

If user passed mode as first arg, use it. Otherwise ask which mode fits.

2. Capture the Track Shape

For every track, capture:

• Outcome — what durable result does this track own?
• Scope — what's included?
• Non-goals — what should NOT be pulled in?
• Dependencies — what must exist or be decided first?
• Blocks — what does this track unlock?
• Handoff artifacts — files / reports / PRs / scripts / datasets / decisions another agent will consume
• Verification — what proves the work is trustworthy? (Mode-appropriate per workflow.md §Verification)
• Risks — known failure modes, unresolved questions, operational hazards

Ask the user for what's missing. If the user provided a paragraph upfront, infer most fields and confirm — don't drill them with 6 sequential questions.

For substantive/risky/multi-week work, also ask whether to add an Outcome Contract:

• Success condition
• Non-negotiable invariants
• Unacceptable failure modes
• Quality bar and tradeoffs
• How a human/operator will know it worked

For small focused work (1-3 days), skip the Outcome Contract — basic Track Shape is enough.

3. Generate the track ID

Format: {shortname}_{YYYYMMDD}

• shortname: 2-4 word kebab-case, derived from the outcome
• YYYYMMDD: today's UTC date

Check conductor/tracks/ for collisions; append _2 if needed.

4. Create files

Directory: conductor/tracks/{trackId}/

spec.md — the track shape. Pre-populate from the user's answers. Project-specific frontmatter (per conductor/CLAUDE.md §Track declaration) lives at the top:

---
layers: [...]              # required if project enforces this gate; else omit
projections: [...]
decs: [...]
facts: [...]
---

Body:

# {Track Title}

**Mode:** {Implementation | Audit/Research | Decision/Planning | Execution/Batch | Publishing/PR}
**Created:** {YYYY-MM-DD}

## Outcome

{The durable result this track owns.}

## Scope

{What's included.}

## Non-goals

{What should NOT be pulled in.}

## Dependencies

{What must exist or be decided first. Other tracks, decisions, datasets, env access.}

## Blocks

{What this track unlocks. Other tracks, releases, decisions.}

## Handoff artifacts

{What another agent or human will consume. Files, reports, PRs, scripts.}

## Verification

{Mode-appropriate: focused tests / dry-run + sample inspection / decision artifact / audit evidence. See workflow.md.}

## Risks

{Known failure modes, unresolved questions, operational hazards.}

## Outcome Contract (optional — for important work)

- **Success:** {condition}
- **Invariants:** {non-negotiable}
- **Unacceptable failures:** {what must not happen}
- **Quality bar:** {tradeoffs}
- **Operator signal:** {how the user knows it worked}

plan.md — phased task list. For Implementation tracks, include test/verification tasks per phase. For non-Implementation modes, the plan can be lightweight (list of artifacts to produce):

# Implementation Plan: {Track Title}

**Track ID:** {trackId}
**Mode:** {mode}
**Status:** [ ] Not Started

## Phase 1: {Phase Name}

- [ ] Task 1.1: {description}
- [ ] Task 1.2: {description}

## Phase 2: {Phase Name}

- [ ] Task 2.1: {description}

## Done when

- [ ] {Mode-appropriate Done criteria — see workflow.md §Definition of Done}

Markers: [ ] pending · [~] in progress · [x] complete (with short SHA / PR ref for Implementation) · [!] blocked (with reason) · [-] skipped.

metadata.json — minimal:

{
  "id": "{trackId}",
  "title": "{Track Title}",
  "mode": "implementation|audit|decision|batch|publishing",
  "status": "scoping",
  "created": "{ISO}",
  "updated": "{ISO}",
  "current_phase": null,
  "current_task": null,
  "commits": [],
  "blockers": []
}

status values: scoping | ready | in_progress | shipped | superseded | absorbed.

5. Register

If project has conductor/status.md with a generator script: regenerate it.

If project tracks DEC reservations (see conductor/CLAUDE.md): remind user to reserve any DECs in docs/project_notes/reservations.md before authoring DEC bodies.

6. Confirm

Show:

Track created: {trackId}
Mode: {mode}
Location: conductor/tracks/{trackId}/

Next: edit spec.md/plan.md if needed, then /conductor:implement {trackId}

Don't

• Don't create tracks for tiny tasks. If the work fits in one focused PR with no coordination, skip the track and just do it.
• Don't add Outcome Contracts to small/low-risk work.
• Don't drill the user with sequential questions if they gave you the info upfront.
• Don't pre-populate fake content in spec.md fields the user didn't answer — leave them blank with a placeholder so it's obvious what's missing.