SABLE: one real CrewAI execution → auditable reliability trace in ~10 minutes

SABLE now has a reusable GitHub Actions path that converts one real CrewAI execution into an auditable sable.submission.v0.9 trace, normalized SABLE trace, and SHA-256 evidence manifest. The path has already been live-validated with LangGraph and CrewAI applications in the SABLE repository; validation, deterministic evaluation, replay integrity, task-success enforcement, and artifact upload all passed.

We'd like one independently maintained CrewAI example/app to run the same path so SABLE can capture an independent external trace.

Minimal integration:

jobs:
  sable:
    uses: socksninja/sable-agent-reliability/.github/workflows/external-verification-reusable.yml@main
    with:
      agent_command: python scripts/run_one_sable_task.py

The command only needs to run one real CrewAI task and write artifacts/submission-v09.jsonl. The reusable workflow handles validation and evidence packaging.

Docs: https://github.com/socksninja/sable-agent-reliability/blob/main/docs/10_MIN_EXTERNAL_VERIFICATION.md
Evidence implementation: https://github.com/socksninja/sable-agent-reliability/pull/18

No endorsement or runtime change is requested. The target is deliberately small: one real task, one run, inspectable evidence.

Start with docs/10_MIN_EXTERNAL_VERIFICATION.md and the referenced external-verification-reusable.yml workflow. Prepare a CrewAI example with an entry point such as scripts/run_one_sable_task.py, configure the reusable workflow, and run one real task. Done means artifacts/submission-v09.jsonl is produced and validation, evaluation, replay integrity, task-success enforcement, and artifact upload pass.