vinny/CMOnet
Watch
1
0
Fork
You've already forked CMOnet
 0
Code Issues 1 Pull requests Projects Releases Packages Wiki Activity Actions

A flow agent's schema repair names the field and its limit #147

Manually merged
vinny merged 0 commits from vinny/schema-repair into master 2026-10-17 10:16:15 -04:00 AGit
Conversation 0 Commits - Files changed -
vinny commented 2026-10-17 10:15:12 -04:00
Owner
Copy link

Summary

  • A flow agent whose final answer breaks its output schema is now told where (at $.findings) and, for a size or range keyword, the limit it broke ((at most 8 items)), instead of only the validator's message ([...] is too long).
  • Same text as the LLM workflow tester's runtime (its commit 5c22d60), which the agent runtime is ported from.
  • Found in lab run 10, trial 6 (tpl-mixed-takeover-heldout): an estimate node answered 9 items where its schema allows 8, and repeated the overlong list after its one repair round, failing the decision.

Test plan

  • tests/flows/test_agent_g3.py::test_a_schema_repair_names_the_field_and_its_limit (prompt JSON and native tools)
  • tests/flows/test_agent_g3.py, tests/flows/test_runner.py
  • Full tools/check.py --rev in the background

🤖 Generated with Claude Code

## Summary - A flow agent whose final answer breaks its output schema is now told where (`at $.findings`) and, for a size or range keyword, the limit it broke (`(at most 8 items)`), instead of only the validator's message (`[...] is too long`). - Same text as the LLM workflow tester's runtime (its commit 5c22d60), which the agent runtime is ported from. - Found in lab run 10, trial 6 (tpl-mixed-takeover-heldout): an estimate node answered 9 items where its schema allows 8, and repeated the overlong list after its one repair round, failing the decision. ## Test plan - [x] `tests/flows/test_agent_g3.py::test_a_schema_repair_names_the_field_and_its_limit` (prompt JSON and native tools) - [x] `tests/flows/test_agent_g3.py`, `tests/flows/test_runner.py` - [ ] Full `tools/check.py --rev` in the background 🤖 Generated with [Claude Code](https://claude.com/claude-code) 
A final answer that broke its output schema was sent back with only the
validator's message, so a list one item too long read "[...] is too long".
In lab run 10 (2026-10-17, tpl-mixed-takeover-heldout) an estimate node's
repaired answer was too long again and the decision failed. The repair now
says where the first mismatch is and, for a size or range keyword, the limit
it broke, as the tester's runtime does since its 5c22d60.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Version 0.7.90 (schema-repair)
Some checks failed
check / check (ubuntu-latest) (push) Has been cancelled
Details
check / check (windows-latest) (push) Has been cancelled
Details
check / check (windows-latest) (pull_request) Has been cancelled
Details
check / check (ubuntu-latest) (pull_request) Has been cancelled
Details
220ac8a6a7
vinny manually merged commit 220ac8a6a7 into master 2026-10-17 10:16:15 -04:00
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
vinny/CMOnet!147
No description provided.