Skip to content

v3.2.20260415 - Vigilant Mantis

Latest
Latest

Choose a tag to compare

View all tags
@nullhack nullhack released this 15 Apr 08:38
v3.2.20260415
c823ffa
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

v3.2.20260415 - Vigilant Mantis

"Sharp eyes, patient method, adversarial by nature."

Changelog

Added

  • Adversarial verification mandate: Reviewer's default hypothesis is now "the code is broken despite green checks" — job is to find the failure mode, not confirm it works (#54)
  • Production-grade gate: New step 3 in verification — app must exit cleanly AND output must change when input changes; static output regardless of input = REJECTED (#54)
  • UUID Drift bash check: One-liner detects duplicate UUIDs across test functions; any duplicate = REJECTED with fix instructions (#54)
  • docs/academic_research.md: 15 cognitive and social science mechanisms with full citations grounding every workflow design decision (#54)
  • Design pattern decision table: Added to developer.md and implementation/SKILL.md; any detected anti-pattern = REJECTED (#54)
  • Architecture contradiction check: Developer must cross-check ADRs against ACs before writing production code (#54)
  • PO pre-mortem: Added at scope step and acceptance step (#54)
  • Semantic alignment rule: Tests must operate at same abstraction level as AC (#54)
  • Integration test requirement: Multi-component features require at least one integration test through the public entry point (#54)
  • Verification Philosophy section: Added to AGENTS.md (#54)

Changed

  • Verification order: Code review before automated commands; run app first as production-grade gate (#54)
  • All review sections converted to tables: Correctness, KISS, SOLID, ObjCal, Design Patterns, Tests, Versions/Build all have PASS/FAIL/Fix columns (#54)
  • UUID Uniqueness rule: If only Given varies it is a property — use Hypothesis @given + @example, not multiple test functions (#54)
  • Production-grade self-check in implementation: Developer must verify output changes with input before handoff (#54)
  • AGENTS.md: Research foundations reference removed — academic research doc is for stakeholders only, not agent context (#55)

Summary

This release hardens the verification workflow against the failure modes identified in the ping-pong incident: no adversarial review, rules as prose that LLMs can hedge, and tests that verify functions instead of user-observable behavior. The reviewer is now explicitly adversarial, all checklists are structured tables with PASS/FAIL/Fix columns, and a production-grade gate ensures the app produces real behavior before code review begins. The name reflects the release theme: a mantis is precise, patient, and adversarial by nature.

SHA: 8be1b47

Assets 2
Loading