Skip to content

fix(agent_explore): expose document_id and harden corpus tool I/O - #403

Merged
EricNGOntos merged 1 commit into
mainfrom
feat/wuchengke/2026-09-08-2
Sep 10, 2026
Merged

fix(agent_explore): expose document_id and harden corpus tool I/O#403
EricNGOntos merged 1 commit into
mainfrom
feat/wuchengke/2026-09-08-2

Conversation

@EricNGOntos

Copy link
Copy Markdown
Contributor

Summary

  • Show real document_id in grep/recall/outline/node_filter/read tool outputs so agent_explore stops passing filenames to corpus.read.
  • Serialize cursor_sdk step budget with a lock to prevent same-turn parallel tool calls from bypassing max_steps.
  • Normalize flat corpus.read argument shapes deterministically before validation.
  • Group sibling evidence blocks under parent section paths in retrieval evidence text.
  • Document CURSOR_API_KEY in ECS deploy README.

Test plan

  • make check (ruff + pyright)
  • test_agent_tools_read_normalize.py (10 tests)
  • test_agent_explore_harness.py (22 tests)
  • test_evidence_renderer_contract.py

Notes

Fix validated locally against 观心 647-doc corpus via debug_agent_explore_episode.py. Not on production until merged/deployed.

Made with Cursor

Show real document_id in grep/recall/outline/node_filter/read outputs so
models stop passing filenames to read. Serialize cursor_sdk step budget
with a lock and normalize flat read args deterministically. Group sibling
evidence under parent section paths and document CURSOR_API_KEY for ECS.

Co-authored-by: Cursor <cursoragent@cursor.com>
@EricNGOntos
EricNGOntos merged commit bf94dbe into main Sep 10, 2026
6 checks passed
@EricNGOntos
EricNGOntos deleted the feat/wuchengke/2026-09-08-2 branch September 10, 2026 10:37
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant