| name | research-execute |
| description | AnchoredAutoResearch execution entrypoint for Codex. Runs approved experiments in manual or bounded fast-loop mode and records evidence back into RSD. |
| version | 1.0.0 |
AnchoredAutoResearch — EXECUTE Entrypoint
Use this skill when the user invokes $research-execute in Codex.
MANDATORY: Read Protocol Before Any Action
- Read
AGENTS.md for project rules
- Read
RSD.md and confirm the current cycle is in EXECUTE with explicit human approval or revision
- Read the approved PLAN section and extract execution mode, verify command, and guard command
- Follow
../research/references/phase-protocol.md for manual execution
- Follow
../research/references/autonomous-loop.md when the plan specifies bounded fast-loop execution
Execution
- Verify the current PLAN has explicit human approval or revision in chat or
RSD.md
- Execute the approved experiments only
- Record artifact-linked evidence in
RSD.md
- Update Code Architecture if the structure changed
- Write the post-execute review summary into
RSD.md
- Run
bash scripts/compile_rsd.sh
- Commit
"research: cycle N execute complete"
- Advance to INTERPRET