v0.7.23

v0.7.23

Changes since v0.7.22:

• feat(sessionproxy): local short-circuit + skip self-tunnel in single-pod mode (ADR 2026-06-06 §5)
• feat(workflow): configurable multi-preset workflows + merge-gate default (ADR 2026-06-06 §4)
• refactor(runtime): split runtime Strategy interface + capability-driven label-write (ADR 2026-06-06 §1/§3)
• refactor(runtime): remove dead coordinator-managed AgentM publish path (ADR 2026-06-06 §1/§6)
• refactor(dispatch): ship per-repo agent config over the wire (ADR 2026-06-06 §2)
• docs: ADR runtime Strategy interface + two-era convergence (2026-06-06)
• fix(pipeline): surface review feedback to dev-agent; merge-agent handles conflicts
• fix(runtime): skip stale agent label transitions to prevent state revert
• fix(pm): stronger dedup — list all titles, grep service name before creating
• fix(pm): check all issues (open+closed) to prevent duplicates
• feat(pm): production-readiness PM agent, 15-min schedule
• feat: spec-driven engineering pipeline with submit_result tool
• fix(statemachine): add total transition cap to catch review↔merge loops
• fix(prompts): explicit --add-label + --remove-label in all agent prompts
• fix(verify): branch check uses remote only, not local worktree
• feat(control): control loop framework + agentm 0.2.3
• fix(control): wire resume ID to spec, handle review rejection, propagate observer errors
• feat(control): implement control loop framework (observer + objective + feedback + controller)
• refactor: goal-oriented prompts + remove auto-submit (control-loop harness)
• chore: agentm 0.2.2 + scenario rename agent_env_repo→sandbox
• chore(docker): install agentm from PyPI instead of git clone
• refactor: autonomous agent mode — remove coordinator-managed publish/labels
• fix(chart): use plain template vars in values.yaml prompts