| name | idea-discovery |
| description | Generate one repo-grounded oral-grade deep-learning idea together with a paper-level method, a full experiment program, and the first formal batch. |
| metadata | {"short-description":"Generate one oral-grade DL idea plus the first formal batch"} |
<codex_skill_adapter>
A. Skill Invocation
- This skill is invoked by mentioning
$idea-discovery.
- Treat all user text after
$idea-discovery as optional extra direction for the next idea cycle.
</codex_skill_adapter>
Generate exactly one repo-grounded oral-grade deep-learning idea with a paper-level method, a full experiment program, and the first formal batch.
<execution_context>
@/home/zhu/workflow/for_code/.codex/idea-workflow/workflows/idea-discovery.md
</execution_context>
Use actual registered child-agent delegation for the three discovery helpers.
Do not satisfy `paper-architect`, `theorem-architect`, or `experiment-designer` locally in the parent rollout.
If child-agent support is unavailable, stop and report that discovery cannot continue with the required helper stack.
After each child agent finishes, write its accepted output immediately into the matching section of `.codex/active_idea.md`.
Do not keep accepted child-agent output only in transient chat text.
Execute the discovery process from @/home/zhu/workflow/for_code/.codex/idea-workflow/workflows/idea-discovery.md end-to-end.
