Releases
v1.3.252
Compare
Sorry, something went wrong.
No results found
ggcode v1.3.252
Highlights
See the changelog below for the user-facing updates included in this release.
Compare
Changelog
release: v1.3.252
feat(r440): RETRACE bidirectional verification gate at task stop (#3204 ) (#3205 )
feat(r439): execution-based tie-breaking for best-of-N candidates (#3202 ) (#3203 )
feat(r438): memory repair loop - surface consolidation findings (#3200 ) (#3201 )
feat(r436): dynamic workflow orchestration - external task graph execution (#3198 ) (#3199 )
fix(tool): simplifyAndroidClass panics on degenerate class like '.' or '.View' (#3194 ) (#3197 )
feat(cost): pre-execution token/cost forecast via KNN over past runs (/forecast, r434) (#3196 )
fix(tool): searchFile scanner ErrTooLong silent miss (re-push #3193 , closes #3191 ) (#3195 )
fix(tool): grep silently missed matches in files with >64KB lines (#3192 )
feat(runeval): interaction-quality dimensions in run report (r432) (#3189 )
feat(mobile): scroll-to-bottom FAB - shows when far from list bottom (#3188 )
fix(mobile): LIVE session row showed previous session's title (global sessionInfo without provenance) (#3184 )
feat(context): parallel context compaction (PCC) for large-payload summaries (#3186 )
fix(tunnel,mobile): stop materializing blank chat bubbles from empty text frames (#3187 )
fix(im): /nick from IM inbound fell to Unknown command - never registered in slash registry (#3185 )
fix(mobile): drop /tmp/gradle-hosts jvmargs from repo gradle.properties (#3182 ) (#3183 )
fix(mobile): theme switch not instant - AppColors static getters carry no reactive dependency (#3181 )
refactor(agent): decompose checkCrossFileImpact orchestrator (complexity 47→11) (#3180 )
fix(tool,agent): output truncation consistency + session budget store release (#3176 ) (#3177 ) (#3178 ) (#3179 )
feat(agent): add compact_context tool - Context-as-Tool (CAT, ACL 2026) (#3175 )
fix(tui): share tool injection never wired - Register rejects duplicates + program read races Update copies (#3174 )
feat(agent/provider): tool-call repair — Python literals, unquoted keys, single-quoted strings + actionable fallback (#3172 )
feat(memory): preference capture full-input + all entry points (r414) (#3143 )
feat(agent): session time budget soft ladder + per-run budget flags (r415) (#3147 )
fix(tool): grep content default cap and cross-backend files order parity (#3170 ) (#3171 ) (#3173 )
fix(tool): diff_scan combined-hunk double-prefix stripping + dd( language gate (#3167 , #3168 )
fix(tool): create_skill O_EXCL placeholder closes Stat-to-write TOCTOU (#3163 )
fix(tool): commit_partition counts header-shaped diff content lines (#3164 )
fix(agent): attribution revert detection honors stash pathspecs without -- and path boundaries (#3161 )
fix(tool): apply_patch enforces opType-section kind agreement (#3158 )
fix(stream): broadcaster read-error exit must Stop its own targets (#3154 )
fix(agent): self-defense exemption keys on read-target fields only (#3155 )
fix(agent): pivot warning dead zone - success must clear warned flag (#3153 )
fix(attribution): revert intervention must target the suspect; CRS tie-break prefers most recent edit (#3150 )
docs(tool): note rg per-file cap in grep head_limit description (#3145 )
fix(tool): deterministic rg content ordering for grep pagination (#3145 )
fix(agent): TTL + render-side staleness for failure-attribution memory (#3146 )
feat(tool): start_share / stop_share agent tools for mobile share lifecycle
feat(agent): persisted failure attribution - task-level credit assignment across sessions (r413)
fix(tool): correct single-quote escape dance in linux argv builder (#3140 )
fix(tool): terminal tab/window honor working_dir; linux -e routes through sh (#3140 )
fix(security): read-time backstop scan closes r409 legacy-stock blind spot (#3137 )
fix(tool): candidate-side dotfile guard in file suggestions (#3138 )
fix(security): close memory-poisoning persistence channel (r409)
fix(tool): length-proportional cap on enum correction distance (#3135 )
fix(tool): cache Windows enum callback; diff scan accepts combined hunks (#3133 )
fix(security): cover retrieval-side peer-content tools in injection guard (#3132 )
test(agent): wiring probe - mode transition resets autonomy-dial window
feat(agent): advisory progressive-autonomy dial - reliability feeds back into mode guidance
fix(tool): AT-SPI probe guards death-prone accessors per node (#3130 )
feat(security): guard inter-agent message channels against prompt injection
test(agent): pin lastSuspect supply line for attribution experiment
feat(agent): attribution validation experiment - Dov-style intervention read-out
fix(tool): create_skill schema truth + atomic write + debug_log export error surfacing (#3128 )
fix(tool): partition plan covers rename/binary/mode files; failed jobs evicted; conflict start marker exact (#3126 )
fix(tool): degraded code-index instance now runs reload loop instead of serving stale snapshot forever (#3124 )
fix(tool): quote-wrapped flag variants no longer bypass Block/Ask/IsDestructive (#3125 )
fix(tool): grep Go-fallback empty result reuses suggestion formatter (r403)
fix(tool): report un-awaited rejected promises in code_execution (#3121 )
fix(memory): cross-process lock for .usage.json sidecar writes (#3120 )
feat(agent): detector guidance fire/suppress observability foundation (r402 P0)
fix(tool): compile-summary keyword gaps, cmd_snippet cross-instance safety, ci_status list branch-free (#3116 #3117 #3118 )
fix(memory): cross-process write lock for AutoMemory dir (r401 GAP-B)
fix(tool): browser profile GC protects live Windows Chromes (#3114 )
fix(agent): oversight digest emitted on all run exits - defer next to reset, idempotent via emitted gate (#3111 )
fix(tool): coerceNumber rejects inf/nan; apply_patch move half-migration disclosure (#3109 #3110 )
feat(tui): missed-skill detection - prompt-named skill never invoked (r398)
feat(agent): oversight triage - novel-vs-routine digest for human review attention (r397)
fix(tmux): separate stdout/stderr in output(), validate %N pane ids (#3107 )
fix(session): fold-aware snippet offsets - ToLower lengthening runes no longer shift search snippets (#3088 )
release: v1.3.251
fix(swarm): BroadcastToTeam mirrors #2788 shutting-down guard (#3104 )
fix(agent): perf baseline tokens dimension in worst-run selection + blind-spot doc (#3102 )
feat(agent): intervention point ledger - proactive defer hints (r395)
feat(mobile): file transfer V1 - agent to phone, any format
docs(design): mobile file transfer V1 design (agent→mobile, any format)
fix(subagent): template store atomic save, corrupt-file visibility, TOCTOU lock (#3100 )
fix(agent): constraint audit ordinal coherence - prose enumerations skipped (#3095 )
fix(stream): broadcaster exit paths stop their targets (#3094 )
feat(agent): perf baseline tracks token spend with cost-regression advisory (r394)
fix(agent): constraint audit recognizes Chinese ordinal lists (#3093 )
fix(im): daemonSessionTurnIndex reads UsageHistory/Metrics via snapshot accessors (#3091 )
fix(agent): pivot_decision family key strips shell prefixes; environmental failures excluded from streak (#3089 )
You can’t perform that action at this time.