| name | compose-pipeline |
| description | Sequential one-command run of /to-prd → /prd-to-spec → /spec-to-tasks, then hands off to the user for manual /archon-code invocation. Use when you want the full PRD-driven development loop in one command. |
Compose Pipeline
Chain three skills sequentially via the Skill tool, then hand off to the user for the final coding step.
Process
Step 1 — Confirm with user
Before starting, confirm with the user:
"About to run /to-prd → /prd-to-spec → /spec-to-tasks on ''. After that finishes, you'll manually invoke /archon-code on the refined task tree. Proceed?"
Wait for explicit confirmation.
Step 2 — /to-prd
Invoke Skill(action=invoke, name=to-prd, args=[<feature>]) and wait for the result.
The PRD will be written to prds/<slug>/PRD.md. Note the path.
Step 3 — /prd-to-spec
Invoke Skill(action=invoke, name=prd-to-spec, args=[<prd path from step 2>]) and wait.
Step 4 — /spec-to-tasks
Invoke Skill(action=invoke, name=spec-to-tasks, args=[]) and wait.
Step 5 — Hand off
After the third skill returns, print:
Composition phase finished. Refined task tree at tasks/INDEX.md.
To run the 50-agent coding pipeline, invoke this manually:
/archon-code
(Skipping auto-invocation: the Skill tool cannot reach /archon-code,
which is a separate command-registry primary. Manual invocation also
ensures you confirm the expensive run step explicitly.)
Limitations
The Skill tool resolves names against register_builtins() only (the SkillRegistry). /archon-code lives in the broader command registry, so the Skill tool returns "skill 'archon-code' not found" if invoked. A future Phase 3.5 ticket can add an ArchonCodeSkill shim that registers in SkillRegistry and dispatches into the existing pipeline runner.
