v1.3.247

ggcode v1.3.247

Highlights

• See the changelog below for the user-facing updates included in this release.

Compare

• Full diff: v1.3.246...v1.3.247

Changelog

• release: v1.3.247
• feat(memory): /init generates standard AGENTS.md with exploration-first prompt
• feat(tool): govern background command job lifecycle (cap, memory floor, shutdown reaping)
• fix(knight): take k.mu when reading status fields in Status/CanPerformTask (#2757)
• fix(agent): subgoal_track no longer matches keywords against tool names (#2758)
• fix(agent): reproducer rerun detection validates command content (#2752)
• fix(tui): statusline failed/timed-out refresh keeps last good output (#2748)
• fix(agent): bare shell 'test' builtin no longer counts as test verification (#2754)
• fix(im): twitch sendRaw write deadline via net.Conn interface, not *net.TCPConn assertion (#2747)
• fix(agent): detect double-lock re-acquire in lock_without_unlock check (#2740)
• fix(agent): scope declarations supersede instead of accumulating (#2733)
• fix(agent): match JS/TS extends/implements class forms in duplicate decl check (#2734)
• fix(knight): strip show-prefix from rename dst side in gitStatusSnapshot (#2755)
• fix(im): lowercase matrix mention needle for mixed-case localparts (#2749 follow-up)
• feat(agent)!: remove FinOps heterogeneous-model guidance injection
• fix(im): matrix mention boundary handles non-word localpart endings (#2749)
• fix(agent): match playbook task-type keywords on word boundaries (#2745)
• docs(context): correct stale claims about retired ClearOldToolResults pipeline
• fix(agent): commit->revert annihilation requires revert target match (#2684)
• refactor(tui): decompose handleKnightCommand monolith into per-subcommand helpers (sa-162)