Skip to content

v0.11.2

Choose a tag to compare

View all tags
@clintecker clintecker released this 27 Mar 15:21
v0.11.2
281e762

Security and reliability audit of all tool nodes across 3 pipelines. Fixes the class of bugs where shell scripts make rigid assumptions about LLM-generated file formats.

Critical fixes

  • Removed eval of LLM-generated verify commands — was arbitrary code execution from free-form text
  • PickNextMilestone no longer silently skips all milestones — flexible header matching + fails loudly on empty extraction

Other fixes

  • TestMilestone strips comments from known_failures, uses go test -skip
  • PickBest winner parsing works regardless of markdown formatting
  • Decompose prompt explicitly instructs expected file formats

Root cause

LLM agents write free-form text. Shell scripts expect rigid formats. Every mismatch is either a silent failure (build loop skipped) or a crash (invalid regex). The fix pattern: validate/sanitize before use, fail loudly on unexpected formats, never eval LLM content.

🤖 Generated with Claude Code

Assets 11
Loading