docs: add agent handoff note

docs/AGENT-HANDOFF-NOTE.md

@@ -0,0 +1,84 @@
+# Agent Handoff Note
+
+Open Problem Lab is designed for agents, but agent work becomes useful only when the next reviewer or contributor can continue without reconstructing hidden context.
+
+This document defines a small handoff note for agent-created or agent-assisted work.
+
+## Why this matters
+
+Agents often leave behind a clean artifact but not the reasoning boundaries around it:
+
+- which sources were checked but rejected;
+- what assumption is load-bearing;
+- where confidence is weak;
+- which validation passed;
+- what the next contributor should attack first.
+
+That missing context turns review into archaeology.
+
+## When to use
+
+Use a handoff note when an agent opens or substantially assists a PR, issue, evidence record, dataset note, or review packet.
+
+The note can live in the PR body, issue comment, or a nearby Markdown artifact.
+
+## Template
+
+```markdown
+## Agent handoff note
+
+**Task attempted:** One sentence.
+
+**Files changed:** Bullet list.
+
+**Primary claim or artifact:** What should reviewers inspect first?
+
+**Sources used:** Links or filenames, with dates when relevant.
+
+**Sources rejected:** Important sources inspected but not used, and why.
+
+**Load-bearing assumption:** The assumption most likely to break the result.
+
+**Known weak point:** The first thing a skeptical reviewer should attack.
+
+**Validation run:** Commands or checks run, or `not run` with reason.
+
+**Review route:** general | evidence | domain | quantitative | field | replication | maintainer.
+
+**Next useful action:** The smallest next step for a human or agent.
+```
+
+## Quality bar
+
+A good handoff note is not a progress diary. It is a compression of what the next reviewer needs in order to avoid wasting time.
+
+It should be short enough to read quickly and concrete enough to reduce review uncertainty.
+
+## Anti-patterns
+
+Avoid:
+
+- vague claims like "researched thoroughly";
+- hiding rejected sources;
+- saying validation passed without naming the check;
+- asking reviewers to infer the weak point;
+- including long chain-of-thought instead of auditable assumptions and evidence.
+
+## Reviewer guidance
+
+A reviewer may ask for a handoff note before reviewing a large or agent-generated contribution.
+
+The goal is not to punish agent work. The goal is to make agent work inspectable.
+
+## Acceptance test
+
+A handoff note is good when a reviewer can answer:
+
+1. What changed?
+2. What claim is being made?
+3. What evidence was used?
+4. What was rejected?
+5. What could be wrong?
+6. What should I check next?
+
+If those answers are missing, the artifact is not review-ready.