mirror of
https://github.com/bytedance/deer-flow.git
synced 2026-08-01 10:56:02 +00:00
41658c5ff4
* feat(skills): add skill review quality gate * fix(skills): skip review eval fixtures in CI * fix(skills): ignore review eval fixtures in bundled scans * fix(skill-review): harden review gate boundaries * fix(skills): address skill review gate feedback
1.2 KiB
1.2 KiB
Effect Verification
Readiness and assurance are separate.
Assurance Levels
static_only: The package was inspected statically. No runtime evidence was retained.trigger_checked: Positive and negative routing cases were executed against a stated model/runtime and retained.behavior_verified: Behavior assertions passed for the reviewed package digest.regression_verified: A frozen baseline and reviewed package were compared with retained outputs and grading evidence.
Evidence That Raises Assurance
To move beyond static_only, evidence must include:
- subject digest;
- model ID;
- runtime or DeerFlow version;
- prompt inputs;
- tool trace;
- outputs;
- assertions or grading result;
- timestamp.
If any of those are missing, stale, contradictory, or attached to a different digest, name the limitation and keep the lower assurance level.
Risk-Based Evidence
Missing evals become material when a skill:
- performs destructive or externally visible actions;
- handles secrets or sensitive data;
- has known sibling-routing collision risk;
- claims measurable improvement over a baseline;
- produces high-stakes output;
- has a history of regressions.