Skip to content

Staging#18

Merged
suguanYang merged 14 commits intomainfrom
staging
May 11, 2026
Merged

Staging#18
suguanYang merged 14 commits intomainfrom
staging

Conversation

@suguanYang
Copy link
Copy Markdown
Contributor

Summary

  • describe the change
  • describe any API, worker, deployment, or migration impact
  • link the related issue or task

Verification

  • list the commands you ran
  • list any manual API, worker, or local-dev checks you performed
  • note anything intentionally not tested

Deployment Notes

  • note new or changed environment variables
  • note database migrations, queue changes, storage changes, or release-order requirements
  • note backwards compatibility or rollback concerns

Checklist

  • Tests were added or updated when behavior changed
  • Public docs, examples, or OpenAPI contracts were updated when needed
  • Database migrations are idempotent and safe to deploy
  • Logs, errors, and validation paths avoid leaking secrets or user data
  • The pull request description explains any breaking or user-visible change

suguanYang and others added 14 commits May 7, 2026 15:34
…tion

refactor: consolidate chunk connection logic and improve parsing in s…
* feat: enhance graph deletion, sync performance, and zip manifest

- Update `remove_document_graph` to use 3-way OR deletion (owner, source, target) and enforce user/namespace scopes.
- Optimize graph builder's disk sync performance by removing redundant chunk loads.
- Extend `ZipResultService` to include a nested title hierarchy in `manifest.json`.
- Simplify hierarchical prompt instructions.

* feat(agentic): implement LLMPolicy and unify agentic RAG ranking logic

* refactor: implement recursive bottom-up doc_nav summary with virtual document node

* fix: resolve ruff lint errors (unused imports and variables)

* test: update expected top_summary format in unit tests
…coring

- Fix doc accumulation in revision loop (selected_docs.clear)
- Fix missing ever_explored_doc_ids tracking for no_confident_match
- Implement KG-exhausted guard in terminal loop
- Replace max(agent, disc) scoring with dual-priority ranking to prevent 0-agent leak
- Promote bottom_discovery to mandatory pre-step
…scoring

fix: resolve E2E agentic retrieval bugs and implement dual-priority scoring
…-user-balance

fix: initialize missing user balance during tier lookup
# Remove unused source file since we reuse the cached image
try:
source_path.unlink()
except OSError:
'outline', 'chunks', 'assets_only', 'image_only', 'table_only',
})

_SCOPE_NAV_PROMPT = """\
Returning None means the agent loop should stop and proceed to
the fixed terminal step (hydrate_and_rank).
"""
_ACTIONS_BLOCK = '\n'.join(

class _JiebaModule(Protocol):
def lcut(self, sentence: str) -> list[str]: ...
def cut(self, sentence: str) -> list[str]: ...
@suguanYang suguanYang merged commit 5bd9136 into main May 11, 2026
13 checks passed
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.

3 participants