Autoresearch

Run bounded research until an explicit validator says the deliverable is good enough.

This is a native Codex adaptation of OMX $autoresearch: keep the measured research loop, validator-gated completion, and artifact discipline; replace OMX CLI, .omx/state, hook persistence, and tmux assumptions with normal Codex tools, repo-local files, optional goal tools, and D:\todo / project work traces when durable tracking is needed.

Use When

• The research output itself is the deliverable.
• The user wants more than quick best-practice lookup.
• Completion must be judged by an explicit script, rubric, reviewer, benchmark, or acceptance artifact.
• The work may require iterative source gathering, synthesis, critique, and revision.

Do not use this for ordinary implementation, simple repo-local analysis, or casual docs lookup. Use $codex-best-practice-research for read-only external guidance and $codex-consensus-plan for implementation planning.

Intake Contract

Before research execution, define:

• Mission: the research question and deliverable.
• Scope: sources, domains, timeframe, language, and exclusions.
• Validator mode:
  • script-validator: a command or deterministic check decides pass/fail.
  • rubric-review: a written rubric and reviewer/critic judgment decide pass/fail.
  • acceptance-artifact: a required output file contains status, evidence, and approval.
• Completion artifact path: where the final validator result or review record will live.
• Evidence ledger path: where sources, attempts, failures, and decisions are recorded.

If these are unclear, run a short $codex-deep-interview style pass first.

Artifact Defaults

Use the repo's existing artifact convention. If none exists, prefer:

reports/autoresearch/<slug>/mission.md
reports/autoresearch/<slug>/evidence-ledger.md
reports/autoresearch/<slug>/draft.md
reports/autoresearch/<slug>/validation.md

In D:\todo-tracked work, also write a task card or work trace when the research is nontrivial or cross-session.

Loop

1. Write or confirm mission, scope, validator, and artifact paths.
2. Gather sources from primary/official/public evidence where possible.
3. Record source quality, dates, and unresolved gaps in the evidence ledger.
4. Produce or revise the deliverable.
5. Run the validator:
   • execute the script or command,
   • apply the rubric,
   • request an independent review lane when available and useful,
   • or update the acceptance artifact.
6. If validation fails, record the failure and revise.
7. Stop only when the completion artifact records pass/approval or when a real blocker remains.

Completion Gate

Do not declare success because the prose sounds complete. Completion requires:

• deliverable exists,
• evidence ledger exists,
• validator result is recorded,
• status is passed / approved,
• residual risks are named,
• any active Codex goal is completed only after the evidence audit passes.

Output Shape

• Mission
• Validator mode and artifact paths
• Sources / evidence summary
• Deliverable path
• Validation result
• Iterations and failures
• Residual risk
• Next handoff, if any

References

Read references/native-autoresearch-checklist.html for a compact readiness and completion checklist.