Skip to content

v2.12.0 - Parseable QA Gates & Human Merge Gates

Choose a tag to compare

View all tags
@admarble admarble released this 26 Aug 12:03
· 11 commits to main since this release
v2.12.0
a34add9
This commit was signed with the committer’s verified signature.
admarble
GPG key ID: E20FFDC10BFD9AF1
Verified
Learn about vigilant mode.

What's Changed

Features

  • Structured gap findings — /qa now emits a machine-readable six-category findings taxonomy that /loop and sequant ready consume; non-blocking findings (document/pause_for_human) are no longer burned as fix iterations (#937)
  • Mutation-verification gate — gate-test ACs must record a parseable SEQUANT_MUTATION marker in the PR body; missing markers cap the QA verdict, fabricated ones floor it (#939)
  • Tautology detector in CI — advisory job annotates PRs with likely-vacuous tests while a false-positive rate is measured (#940)
  • Evidence: clause on ACs — an AC can declare its own verification command, which /qa must execute before marking it MET (#938)
  • run.mcpAllowlist — explicit per-server opt-in to pass specific Claude Desktop MCPs to phase agents (#936)
  • /merger Named-Set Boundary — never merges a PR whose issue wasn't named in the invocation; out-of-set dependencies halt with a report (#961)

Behavior Changes

  • /fullsolve no longer merges the PR by default — it stops at PR creation; pass --auto-merge or set run.autoMerge: true for the old end-to-end behavior (#958)
  • Phase agents no longer inherit your Claude Desktop MCP config — they get the sequant server plus the project's .mcp.json; use run.mcpAllowlist for desktop-only servers (#936)

Bug Fixes

  • AC parser: fenced code-block examples no longer parse as real ACs (#947); keyword inference matches word boundaries, not substrings (#946)
  • /loop trusts the QA context embedded in its orchestrated invocation instead of re-fetching stale comments from GitHub (#960)
  • pre-tool.sh No-Changes Guard resolves cd targets instead of silently checking the wrong directory (#963); tautology detector no longer flags describe.each shell-script suites wholesale (#963)
  • Orchestrated sequant run posts the fresh QA verdict comment, not just a stale one (#964)

Dependencies

  • @anthropic-ai/claude-agent-sdk 0.3.218→0.3.234, chalk 5→6, hono/node-server/open bumps (#953); dev deps (#905); nanoid security patch (GHSA-2v37-7h3g-55p8)

Full Changelog: v2.11.0...v2.12.0

Assets 2
Loading