Repository navigation
Fix OWASP agent probe gaps (chapters, people, membership, CRE facts) - #1127
northdpole wants to merge 18 commits into
Conversation
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. 🗂️ Base branches to auto review: 1
Please check the settings in the CodeRabbit UI or the ⚙️ Run configuration
You can disable this status message by setting the Use the checkbox below for a quick retry:
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
fee35d3 to
b6319c7
Compare
86fd285 to
a0cf4e4
Compare
b6319c7 to
2a731f4
Compare
a0cf4e4 to
90b84e5
Compare
2a731f4 to
629eb20
Compare
90b84e5 to
91e5190
Compare
629eb20 to
5831fe9
Compare
There was a problem hiding this comment.
There is an SDK available -- https://github.com/owasp/nest-sdk-python
|
Folded in chat memory + OWASP entity fixes (
Stack restacked on top: #1132 → #1137 → #1142 (and sibling #1136). |
70c851e to
636089a
Compare
|
Correction: replaced the bundled chat-memory+entity commit with entity-only |
eb0cd49 to
69641cd
Compare
5831fe9 to
dfec9a8
Compare
69641cd to
6fd8b0b
Compare
…facts. Route hard probes correctly (inactive chapters, geo neighbors, leaders, membership pricing, fail-closed talks, CRE-844-486) and extend the index/crawler so gold+synth evals pass.
Vendors the 400 unique offline probes plus router/slot fixes that lift agent-path scoring to pass_rate>=0.90 with zero hallucinated talks.
Adds a package README (keys, sync, chat/MCP, suite scoring), a public fixture scorer script, and a Makefile target so contributors can reproduce agent-path eval without the private homeserver harness.
Ingest project leaders from leaders.json, add person/project/compare routing, and ground AI Exchange lethal-trifecta answers so Rob/AIX/GenAI questions stop falling through to CRE RAG hallucinations.
Pass a capped messages[] history through /completion into LiteLLM and the OWASP agent so clarify follow-ups keep prior context; UI gets a New chat control that clears the local thread.
Prefer main DATABASE_URL for the agent index, fail-closed talk+CRE compounds, board-person without year prompts, and geocode fallbacks used on staging.
bb41ecc to
32aefc1
Compare
Replace brand/AIX keyword classifiers with cre_graph / corpus / owasp_agent_meta scoring; embed agent meta (+ evidence notes) into the shared embeddings table behind source_channel.
Use Gemini REST by default with batched commits so litellm/Flask no longer OOM-kills the backfill.
Store gemini/gemini-embedding-001 so meta rows match cre_graph/corpus and the completion embedding contract check passes.
Sim-lane clarify no longer overrides clear intents; project who-leads extracts the project name instead of stealing chapter lookup.
Adds adversarial failures for Bangalore/São Paulo chapter contact questions into the synth suite.
Avoids per-turn create_app OOMs that produced noisy HTTP 500 failures on staging.
… ontology Prefer embedding top-k plus constrained LLM pick over brittle match_* paths. Embed chapter topics and board roster as corpus; keep AIX on evidence without CRE fallthrough. Extend adversarial asker for staging_ssh LLM_PICK/TOP_K runs.
Match meta hints on word boundaries so "prevent" is not "event", drop bare "chapter" as a meta keyword, route what-is OWASP chapter phrases, and remove Athens/Thessaloniki/LA (and suburb city) hardcodes from intent.
Remove meta_hints/normative_hints and the has_meta/has_norm cascade that misrouted asks like prevent→event. Keep phrase-shaped intents and malicious refuse; default to cre_normative and let sim-first answer community meta. Co-authored-by: Cursor <cursoragent@cursor.com>
Remove CRE/AIX keyword factoid gates, metro chapter maps, city allowlists, and CapWords person forests. Match projects on ingestion-derived aliases, resolve people from the index, geocode nearest chapters, and casefold/ strip adversarial wrappers so terse AIX prompts retrieve evidence.
Summary
cursor/owasp-agent-oie-a097): hardens the experimental OWASP metadata agent against real probe failures from homeserver eval.cre_facts.py/membership.pyhelpers and expandsowasp_agenttests; gold (40) + synth (400) eval suites pass on the homeserver image.Docs
application/utils/owasp_agent/README.mdowasp_agent_synth_v2_questions.txtTest plan
owasp_agentunit tests green