Python verification (partner contract)

This folder is the Python-first LangGraph / kernel demo surface. Commands for the Node reference emitter remain in the generated partner quickstart; this README stays prose-only per repo drift rules.

Canonical integrator guide: docs/integrate.md. LangGraph checkpoint trust statute (including capability matrix): docs/integrator-verification.md#langgraph-checkpoint-trust.

Run the kernel demo (requires repo root examples/partner-quickstart fixtures):

pip install -e "python/[dev]"
python examples/python-verification/run_partner_kernel_demo.py

Postgres (same certificates as the Node CLI): install pip install -e "python/[dev,postgres]" and set database_url to a connection string (e.g. postgresql://user:pass@localhost:5432/mydb) on VerificationSession. For direct kernel use, import verify_langgraph_checkpoint_trust from agentskeptic.kernel and pass that URL. Apply the same partner.seed.sql (or your schema) to the target database for parity with the partner quickstart.

CrewAI minimal example

Deterministic, in-process demo: shows where the pinned CrewAI hook surface (crewai.hooks.before_tool_call) is attached during AgentSkeptic.verify(framework="crewai") (via agentskeptic._integrations.crewai.attach_crewai), how buffered tool observations flow into the same contract-SQL verification kernel used elsewhere, and prints an Outcome Certificate JSON. Authority for pins and hook path: python/FRAMEWORK_LOCK.md.

pip install -e "python/[dev,crewai]"
python examples/python-verification/crewai_minimal.py

No LLM or API keys; uses the same examples/partner-quickstart SQLite seed and partner.tools.json as the LangGraph demo above.

If CrewAI is not installed, the script exits non-zero with a short message (no traceback).

Reading the certificate: the JSON includes trustDecision: safe, unsafe, or unknown, plus stateRelation, highStakesReliance, and relianceRationale. The embedded human-oriented report uses wording such as TRUSTED / NOT TRUSTED / ineligible paths—align machine fields with that narrative when integrating.