Skip to content

v0.9.4

Choose a tag to compare

@zw008 zw008 released this 16 Sep 12:32
· 8 commits to main since this release

Fixed

  • redis_memory_pressure_rca's top-level pressure flag no longer decides by comparing another finding's cause against the literal sentence "Healthy — within thresholds". Behaviour is unchanged, but rewording that sentence — or normalising its em dash in a style sweep — would have reported pressure on a healthy broker. It now reads the measured condition, and an AST-based structural test pins the whole package against a decision being taken on a finding's prose.
  • agent-guardrails.md and SKILL.md claimed the four RCAs "rank findings worst-first" and that "priority is in the payload, not implied by list position". Neither was ever true here: there is no rank key, and findings are appended in the order the checks run — and the claim sat in the table telling operators not to spend prompt budget on it. The docs now say order is not severity, and point at the evidence number each finding does carry.

Full notes: CHANGELOG.md