| name | bmad-autonomous-planning |
| description | Portable autonomous BMALPH planning for Codex. Use when the user wants specs from a short task description and expects the current AI session to orchestrate BMALPH subagents without human interaction.
|
Use this skill from the current Codex session. Do not rely on repo-local bash
wrappers, make targets, or other launcher automation for the planning flow.
The canonical planning contract lives in
.claude/skills/bmad-autonomous-planning/SKILL.md. This wrapper is only the
Codex-specific handoff.
If _bmad/ is missing, stop and tell the user BMALPH assets must be initialized
for the repository before this skill can run. Otherwise:
- Read the local
bmalph skill wrapper, _bmad/COMMANDS.md, and the resolved
BMAD config file first.
- Read
.claude/skills/bmad-autonomous-planning/SKILL.md.
- Resolve or create a bundle directory under the configured planning artifacts
folder, typically
autonomous/<timestamp>-<task-slug>/.
- Run each BMALPH planning stage in a dedicated subagent, using the BMALPH
command surface as the handoff contract:
- research and repository context:
analyst
- product brief:
create-brief
- PRD:
create-prd
- architecture:
create-architecture
- epics and stories:
create-epics-stories
- implementation readiness:
implementation-readiness
- After each subagent returns, the main agent must decide the next step,
answer workflow questions on the user's behalf, and only then hand the next
stage to another subagent.
- Validate and improve each artifact for up to three rounds without blocking on
BMALPH approval menus. When another subagent pass is needed, prefer the
matching BMALPH validation command from
_bmad/COMMANDS.md when available.
- For every spawned planning or validation subagent, set
model: gpt-5.4 and
reasoning_effort: xhigh. Do not use gpt-5.4-mini for stage ownership in
this workflow.
For each subagent, name the BMALPH command first and provide only the backing
workflow or agent files required by that command. Do not hand off raw
workflow-file paths without the command context.
Minimal Codex trigger example:
Use the bmad-autonomous-planning skill to plan customer tagging for the core service. Follow the repository's BMALPH autonomous planning skill, work in the current session, keep the flow fully autonomous, and run each planning subagent on gpt-5.4 with xhigh reasoning.
The finished bundle should include:
research.mdproduct-brief.mdproduct-brief-distillate.md when usefulprd.mdarchitecture.mdepics.mdimplementation-readiness.mdrun-summary.md
Optional GitHub issue or specs-only PR work happens only after the planning
bundle is complete and only when the user explicitly asks for it.
