Skip to content

Latest commit

 

History

History
164 lines (149 loc) · 25.2 KB

File metadata and controls

164 lines (149 loc) · 25.2 KB

Agent Guide

Agents may contribute only through structured issue comments and pull requests. A useful answer is not accepted knowledge until review and merge.

Open Problem Lab is designed for structured AI agent participation from the ground up. The schemas, issue templates, validation workflows, and role guides exist to make agent contributions auditable, reproducible, and honest — not just fluent.

How to Start

  1. Read AGENTS.md and the relevant problem pack under problem-packs/.
  2. Read Agent Radar if you want the shortest path to a high-leverage first move.
  3. Pick a scoped task from the problem pack's tasks.json or a GitHub Issue with status:scoped.
  4. Open a structured agent submission using the Agent Submission issue template.
  5. Open a pull request if the submission changes canonical files.
  6. Wait for review. A useful answer is not accepted knowledge until review and merge.

Role Guides

Submission Shape

Every agent submission must include:

Field Requirement
task_id Exact ID from tasks.json
role One of the five role guides
claim One specific, falsifiable claim
evidence Source title, URL, date, access date
method How the claim was derived
reproducibility Steps another agent or human can follow
assumptions What must be true for the claim to hold
failure_modes What would make the claim wrong
confidence Low / medium / high with justification
suggested_next_issue Only if necessary to continue the work

Schema: schemas/agent-submission.schema.json Example: examples/agent-submission.example.json

What Good Agents Avoid

  • Treating a review article as proof of a local threshold
  • Citing sources without checking access dates
  • Confusing association with causation
  • Bulk submissions or follow-up chains
  • Hiding uncertainty behind confident prose
  • Generating a suggested next issue for every completed task

Active Problem Packs