speckit-pro: v2.30.0

Highlights

• Skill guidance no longer instructs runner operations that were deferred or shell scripts that no longer ship, and the coach stops teaching an upgrade path that overwrites the constitution. Capability pointers now resolve from an installed plugin rather than only inside a checkout.
• The autopilot now validates phase gates by calling the validate-gate runner helper directly instead of dispatching a subagent, removing seven model calls per run and a duplicated contract that had drifted from the helper's actual output.
• The pre-implement confidence gate now computes its score from the five criterion values and applies severity deductions in code, instead of trusting the aggregate the synthesizer stated. Consensus routing, review-log aggregation and Phase 7 task partitioning move into runner helpers, and plural security keywords now correctly trigger the full three-analyst fan-out.
• The consensus synthesizer now runs at high reasoning effort rather than max, matching its bounded rule-applying role; every agent that authors specs, plans or code keeps max. A new test keeps the three consensus analysts from drifting apart.

Commit appendix

2.30.0 (2026-09-03)

Features

• speckit-pro: move consensus arithmetic and routing into runner helpers (#525) (897041b)
• speckit-pro: replace the gate-validator agent with a direct validate-gate helper call (#524) (588f4d6)
• speckit-pro: right-size the synthesizer effort and guard the analyst roster (#526) (92ea316)

Bug Fixes

• speckit-pro: repair stale paths, retired helpers and version pins in skills (#523) (f14e084)