End-to-End: Claude + Conductor

This runbook is intentionally strict about unknowns. Anything not verified is marked with [RESEARCH REQUIRED].

Scope

• Environment: Conductor workspace + Claude-based coding agent
• Method: ACE (/auro.constitution -> /auro.charter -> /auro.plan -> /auro.tasks -> /auro.implement)
• Outcome: shipped change + traceable artifacts

Preflight

1. Confirm repository and branch strategy:

• Repo URL: <RESEARCH REQUIRED: canonical repo URL>
• Default branch: <RESEARCH REQUIRED: protected branch policy>

2. Confirm local prerequisites:

• git --version
• node --version
• npm --version

3. Confirm Auro CLI:

• auro --version

4. Confirm Conductor workspace setup details:

• <RESEARCH REQUIRED: exact Conductor UI flow/versioned labels>

5. Confirm Claude integration mode:

• <RESEARCH REQUIRED: Claude model + tool permissions policy>

Phase 1: Project Bootstrap

1. Create/open workspace in Conductor.
2. Ensure working directory is correct.
3. Initialize Auro (if missing):

auro init <PROJECT_NAME>

4. Validate .auro/ structure exists.

Expected artifacts:

• .auro/constitution.md
• .auro/templates/
• .auro/scripts/

Phase 2: Governance First

Run once per project:

/auro.constitution

Checklist:

• principles are testable
• constraints are explicit
• no conflicting rules

If uncertain:

• add [NEEDS CLARIFICATION: ...] instead of assumptions

Phase 3: Charter

Run:

/auro.charter "<feature request>"

Review for:

• independent user stories (P1/P2/P3)
• numbered requirements
• measurable success criteria
• edge cases

Stop condition:

• no unresolved ambiguity markers

Phase 4: Plan

Run:

/auro.plan

Validate phase gates:

1. Simplicity
2. Anti-Abstraction
3. Integration-First

Expected outputs:

• plan.md
• research.md
• data-model.md
• contracts/
• quickstart.md

Phase 5: Tasks

Run:

/auro.tasks

Validate:

• every task maps to a user story
• [P] markers only where safe
• phases are dependency-correct

Phase 6: Implementation

Run:

/auro.implement

Execution guardrails:

• small commits per task cluster
• run tests after each phase
• stop and revise charter/plan if behavior drifts

Phase 7: Verification

Run project checks:

<RESEARCH REQUIRED: exact lint command>
<RESEARCH REQUIRED: exact test command>
<RESEARCH REQUIRED: exact build command>

Manual verification checklist:

• happy path
• one expected failure path
• one boundary/edge case

Phase 8: Ship

1. Open PR against main.
2. Include artifact traceability:

• charter section -> plan section -> task IDs -> commits

3. Merge with CI green.
4. Verify deployed URL:

• <RESEARCH REQUIRED: environment URL>

Failure Recovery

If run gets stuck:

1. Return to latest valid charter.
2. Regenerate plan from charter.
3. Regenerate tasks from plan.
4. Continue implementation with smaller phase scope.

Research Placeholders To Resolve

1. Conductor version-specific UI flow
2. Claude configuration and permissions
3. Canonical CI commands for this repo
4. Deployment verification URL/process