Skip to content

v1.3.241

Choose a tag to compare

View all tags
@topcheer topcheer released this 18 Sep 06:43
· 1911 commits to main since this release
v1.3.241
9e904fe

ggcode v1.3.241

Highlights

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

Compare

Changelog

  • release: v1.3.241
  • fix(provider): wrap openai-responses stream goroutines in safego.Go
  • Anthropic PTC: verbatim caller echo-back + code_execution_tool_result round-trip (#2508)
  • feat(provider): OpenAI Responses background mode (sa-73) (#2518)
  • feat(provider): per-model/per-endpoint LLM call policy (request_timeout, max_retries) (#2528)
  • feat(provider): add service_tier processing-tier control for openai endpoints (#2530)
  • feat(anthropic): support server-side Tool Search Tool (regex/bm25, advanced-tool-use beta) (#2507)
  • sa-63: OpenAI Responses built-in server tools (code_interpreter, file_search) (#2505)
  • feat(provider): strict tool use (structured outputs) behind opt-in strict_tools knob [sa-60] (#2502)
  • feat(mcp): implement SEP-2243 HTTP header standardization (client side) (#2498)
  • feat(anthropic): server-side context editing (context-management beta) (#2488)
  • feat(memory): provenance & usage feedback loop for auto-memory (sa-85)
  • feat(responses): client-executed apply_patch tool (sa-67) (#2509)
  • feat(agent): resolve cross-file variable-receiver method refs via stdlib go/types (#2164) (#2470)
  • fix(tui): expose openai-responses in the new-vendor/new-endpoint protocol cycler
  • test(agent): cover experience recall matching and skip paths
  • feat(memory): Experience Case Bank — case-based recall across sessions
  • feat: bi-temporal validity for knowledge graph edges (Zep/Graphiti model)
  • test(mcp): guard task-envelope passthrough in the MRTR loop
  • feat(mcp): client-side MCP Tasks protocol (SEP-1686)
  • test(config): thinking_mode endpoint plumbing coverage
  • feat(anthropic): adaptive thinking carrier for Claude 4.6+/5.x
  • feat(tui): Alt+R reverse input history search (Claude Code v2.0 parity)
  • fix(memory): duplicate-check exact match must compare disk keys (#2520)
  • test(tui): guard i18n coverage for session pin/tag slash commands
  • feat(session): add session pinning and tags for long-lived session management
  • test(sa-74): coverage for gemini logprobs, agent confidence notice, config pass-through
  • feat(provider): token-logprob confidence telemetry (sa-74)
  • test(config): cover text_verbosity endpoint resolution (sa-72)
  • feat(provider): add GPT-5 text.verbosity control for Responses API (sa-72)
  • test(a2a): assert agent card serializes required protocolVersion field
  • fix(mcp): MRTR deferred sampling must pass ValidateSamplingParams (#2514)
  • feat(a2a): A2A v1.0 compatibility - ProtoJSON enums, well-known fallback, a2a+json media type
  • internal/mcp: implement 2026-07-28 error code allocation policy (sa-70)
  • feat(mcp): correlated subscription streams (MCP 2026-07-28 subscriptions/listen)
  • test(agentruntime): pin ApplyProviderToAgent memory tool gate wiring (#2511)
  • fix(agent): gate memory tool executor behind the provider declaration (#2511)
  • test(mcp): cover config stateless wiring in NewClientFromConfig
  • feat(mcp): 2026-07-28 stateless protocol core (SEP-2575) client support
  • test(provider,permission): cover memory tool declaration and whitelist
  • feat(agent): implement Anthropic Memory Tool (memory_20250818)
  • feat(provider): OpenAI Responses hosted server-side tools (web_search) (#62-sa)
  • feat(a2a): verify signed Agent Cards (spec §8.4) via JWS + JCS canonicalization
  • test(gemini): cover registry server_tools injection end-to-end
  • fix(agent): word-accurate anchor for commandFailed verification cross-check (#2500)
  • feat(gemini): built-in server tools (google_search/url_context) with grounding source echo
  • test(mcp): add end-to-end MRTR round-trip test over real HTTP transport
  • feat(mcp): client support for Multi-Request Tool Results (resultType: input_required)
  • mcp: client-side freshness cache for CacheableResult (SEP-2549)
  • feat(mcp): tool-enabled sampling per 2025-11-25 SEP-1577
  • responses: stateless encrypted reasoning round-trip
  • fix(agent): git_revert joins fileMutatingTools so the dedup epoch bumps (#2492)
  • mcp: implement Streamable HTTP SSE resumability via Last-Event-ID
  • feat(mcp): implement completion/complete argument autocompletion
  • fix(agent): dedup epoch must advance on shell-based workspace mutations (#2486)
  • mcp: propagate notifications/cancelled when the client abandons in-flight requests
  • fix(a2a): strip card-declared API key header on cross-host redirects
  • feat(anthropic): server-side tools (web_search/web_fetch) with verbatim echo-back
  • fix(mcp): send MCP-Protocol-Version header on HTTP requests after initialize
  • feat(a2a): protocol extension negotiation per A2A v1.0 spec
  • feat(mcp): surface server initialize instructions on tool descriptions
  • memory: AGENTS.md nested discovery (closest-wins) for monorepos
  • feat(mcp): implement resource subscriptions (MCP 2025-06-18)
  • test(anthropic): pin buildParams ctx image-routing fallback
  • feat(anthropic): wire Files API uploader into message conversion
  • security: sanitize terminal escape sequences at display boundaries (sa-41)
  • feat(provider): add OpenAI Responses API adapter (sa-40)
  • fix(tool): normalize new_text line endings to the file's convention in all edit paths
  • feat(memory): sleep-time memory consolidation at startup
  • feat(mcp): parse 2025-06-18 tool annotations and refine read-only blocking
  • anthropic: cache-aware top-level output_config.effort carrier (#1733)
  • feat(audit): tamper-evident hash-chained audit ledger for agent actions
  • test: cover the Linux sandbox launcher protocol
  • feat: Linux sandbox backend for agent shell commands (Landlock + seccomp)
  • fix(tool): home-index guard matches case-variant home spellings on Windows/macOS
  • fix(agent): eagerly initialize toolDedup ledger to satisfy state-field regression guard
  • feat(agent): duplicate-suppression ledger for non-idempotent mutating tool calls
  • fix(tool): disable code index when working dir is the user home (OOM)
  • fix(agent): stop Working-Tree Invalidation FPs on tree-preserving git commands (#2465)
  • fix(agent): stop claim-verify Build-failed false positives on grep+echo chains (sa-30)
  • fix(agent): silent-error detector no longer indicts read-only environmental git failures
  • fix(irrev-gate): tier read-only/verification shell commands as Tier 0 (sa-28)
  • fix(agent): exempt advisory-backed truncations from error-propagation chains
  • feat(mcp): support URL mode elicitation (MCP 2025-11-25)
  • test(tool): companion-test coverage for OS sandbox; add missing non-darwin stub
  • feat(tool): OS-level kernel sandbox for shell commands (sandbox.enabled)
  • fix(auth): JWKS stale-while-revalidate fallback keeps A2A auth alive during IdP outages (#H-05)
  • agent: wire toolreplay into safeExecute for deterministic record/replay (GGCODE_TOOL_TAPE)
  • agent: nil-receiver safety for mutatingLedger
  • agent: non-atomic outcome semantics for mutating tool calls
  • feat(mcp): per-server circuit breaker for MCP tool calls (sa-21)
  • feat(tools): add trial_fork tool for parallel multi-strategy exploration
  • test(mcp): cover SandboxSafe flag for read-only vs write adapters (sa-19)
  • feat(sandbox): bridge read-only MCP tools into code_execution (sa-19)
  • test: assert snapshotTasksInto captures workspace fingerprint
  • feat: resume reconciliation for session task board vs workspace drift
  • feat(subagent): close the model-tier cascade loop with escalation hints
  • feat: temporal context injection + current_time tool
  • refactor: extract buildPreExecBatch and preExecOne from preExecuteReadOnlyTools
  • test: align MixedBatchSkipped doc comment with conflict-aware semantics
  • agent: conflict-aware partial parallelism for mixed tool batches
  • fix(agent): use %q for rootKey in cascade guidance (CodeQL unsafe quoting)
  • test(agent): add Agent-level wiring tests for tool_search
  • feat(agent): deferred MCP tool schema disclosure via tool_search meta-tool
  • feat(agent): tool output offloading — persist truncated results to disk for recovery (#2446)
  • fix(tool): deterministic tool list ordering to preserve prompt/KV cache (#2445)
  • feat(agent): harness fingerprint for scaffolding-regression attribution
  • agent: add effect ledger — duplicate-side-effect awareness on shell retries
  • Consolidate dual prompt-injection scanners into one unified pipeline
  • feat(agent): add adversarial evaluator gate (generator-evaluator separation)
Assets 51
Loading