Skip to content

v2.13.0 - Verdict trust, loud failures, model roles

Latest
Latest

Choose a tag to compare

View all tags
@admarble admarble released this 30 Aug 22:36
· 2 commits to main since this release
v2.13.0
633f19a
This commit was signed with the committer’s verified signature.
admarble
GPG key ID: E20FFDC10BFD9AF1
Verified
Learn about vigilant mode.

What's Changed

Features

  • awaiting_verification workflow state (#972, #977) — a NEEDS_VERIFICATION QA verdict no longer parks an issue in ready_for_merge: the follow-up qa re-run proceeds without --force, the state is surfaced in sequant status/dashboard with a re-run hint, and directly-merged PRs are swept to merged. MCP sequant_run now accepts force (previously silently ignored). New reference: docs/features/qa-verdict-workflow-states.md.
  • Model roles (#975, #979) — run.modelRoles maps semantic roles (fast/strong/frontier) to model strings; phase policy and --models accept role:<name> references, raw strings pass through verbatim, and a missing role fails loudly at config-resolution before any session spawns. Per-phase run metrics record the requested value and the resolved concrete model ID from modelUsage.
  • Constitution with teeth (#943, #978) — the constitution template is now the enforceable agent contract: a generated Definition of Done table drift-gated in CI (lint:constitution-dod), an AC authoring standard cross-linked from /spec, and Boundaries/Budgets sections where every rule names its real enforcing mechanism. Customized constitutions are preserved by update/sync.

Bug Fixes

  • API-errored turns no longer report as successful phases (#973, #976) — an SDK result with subtype:"success" + is_error:true (e.g. an unrecognized model name) now fails the phase loudly with structured terminal_reason/api_error_status context, instead of a zero-work "success".
  • sequant status hints named a nonexistent --phase flag; corrected to --phases (#977).

Maintenance

  • agents.model accepts any free string instead of the stale three-model enum (#975).
  • Dependency bumps: @anthropic-ai/claude-agent-sdk 0.3.242 (runtime-smoke-verified against the driver's is_error/modelUsage contract), hono 4.13.4, inquirer 14.1.0 (#969); dev-deps group (#970).

Security review: SECURE — no new process-execution, network, or privilege surface; supply-chain registry signatures and provenance attestations verified (npm audit signatures: 357 signed, 84 attested).

Full Changelog: v2.12.0...v2.13.0

Assets 2
Loading