Skip to content

RustyCode v0.12.0

Choose a tag to compare

@github-actions github-actions released this 29 Aug 15:14
· 11 commits to main since this release

[0.12.0] - 2026-08-29

Features

  • Expose sandbox/permission mode in turn metadata via HarnessPolicy
  • Support GPT-5.6 and Claude Opus 5
  • Plugin skills/MCP registration with multi-scope install
  • Wire /checkpoint save/list/delete to CheckpointManager
  • Add RUSTYCODE_NO_ALT_SCREEN env var for terminal capture
  • Add benchmarks for bus, protocol, and compaction crates
  • Add benchmarks for tool-executor hot paths
  • Add exhaustive/ultracode mode to orchestration
  • Add SWE-bench bridge and cross-compile support
  • Eval_bridge accepts rustycode-bench native pred format
  • Submit best-scoring patch across swebench verify attempts
  • Record loop telemetry in swebench metrics sidecar
  • Docker-based SWE-bench verify in eval images
  • Add A/B comparison script for SWE-bench arms
  • Union F2P+P2P docker verify with failure-detail tracebacks (experiment-3 commit #16 + eval-only)
  • Feature-ADD prompt guidance + F2P-first best-patch scoring
  • Advisor auto resolves a strictly stronger advisor model
  • Bench prompt frames provided tests as the specification
  • Evidence-based agent defaults — on-demand advisor, opt-in continuation, de-escalated prompts
  • Persist and search architectural decisions
  • Route blocking review findings to team fixes
  • Add blur-aware TUI notifications
  • Add glm-5.3, glm-5.2, glm-5.3-flash to model catalog
  • Add glm-5.3 and glm-5.3-flash to zhipu provider

Bug Fixes

  • Harden child processes with process groups, env scrubbing, and session ID injection
  • Preserve line endings in apply_patch and reject duplicate paths
  • Stabilize glob sort with path tie-breaker for equal modification times
  • Make event bus failures recoverable
  • Correlate in-process server responses
  • Harden session snapshot persistence
  • Make storage writes safe and deterministic
  • Make indexing and search deterministic
  • Isolate TUI wide overlays
  • Honor agent provider and profiler constraints
  • Enforce agent runtime session deadlines
  • Harden llm transport and tool selection
  • Harden runtime worktree lifecycle
  • Stabilize orchestration scheduling boundaries
  • Consume workflow fallback nodes once
  • Harden learning storage persistence
  • Enforce wiki schema invariants
  • Harden vector memory metadata handling
  • Bound and stabilize memory consolidation
  • Harden team learning persistence
  • Stabilize team ensemble result ordering
  • Harden bench job concurrency and task name parsing
  • Handle prefixes with internal underscores and add timestamp overflow guard
  • Harden output filter bridge I/O
  • Bound in-process server requests
  • Stabilize server protocol notifications
  • Make event bridge startup ready
  • Clip connector screenshots by character
  • Make retry configuration total
  • Stabilize ring buffer length snapshots
  • Keep visualization snapshots consistent
  • Stabilize executable discovery and chains
  • Align tauri frontend packaging
  • Stabilize in-memory task queries
  • Centralize environment mutation
  • Serve tool server frontend assets
  • Preserve frontend request pending state
  • Stabilize embedded agent registry ordering
  • Reap LiteRT fallback processes
  • Reset stale client state when steering
  • Validate dotted workflow output references
  • Clean up lazy WebSocket sessions
  • Reduce macOS unwind metadata
  • Surface team task failures
  • Parse multiline Bash tool arguments
  • Avoid spawning inert MCP child reaper
  • Preserve file read and diff semantics
  • Bound thinking sessions and unify pipeline turns
  • Bound MCP content-length headers
  • Harden synchronous MCP process cleanup
  • Bound synchronous MCP stderr capture
  • Enforce shared-memory limits on updates
  • Shared_memory validation, ask_user stuck detection, help text, MCP framing notes
  • Make Bash timeout units explicit
  • Retain fork identity when join tasks fail
  • Move watcher indexing off async workers
  • Flush rollout records on run completion
  • Surface runtime system messages in TUI
  • Complete TUI stream event forwarding
  • Preserve terminal TUI stream events
  • Enforce crash lock exclusion
  • Validate OAuth token endpoints
  • Increase shared runtime worker stack
  • Require plan completion before milestone success
  • Require milestone validation evidence
  • Bound milestone validation commands
  • Require harness validation evidence
  • Honor harness recovery validation timeout
  • Preserve harness commit failures
  • Fail closed on harness rollback errors
  • Preserve existing harness task queues
  • Discard multi-line text on Esc instead of joining with space
  • Surface agent completion results to TUI message list
  • Detect same-tool-success repetition in agent loop
  • Team orchestrator sends canonical tool format to LLM provider
  • Add turn-count fallback to StuckLoopPlugin path for loop convergence
  • Cancel stale running tool badges on Ctrl+C
  • Stop preserving old system messages during compaction
  • Scroll to confirmation after slash command effects
  • /compact reported ~0 tokens saved by comparing local estimate vs API max
  • /load displayed numbered indices but could not resolve them to session IDs
  • Suppress duplicate 'Press Ctrl+Q to quit' hint on repeated Ctrl+C
  • /learnings shows no feedback when loading asynchronously
  • /marketplace shows no feedback when loading asynchronously
  • Slash command feedback gaps found by dispatch audit
  • Use PluginManager::new(cwd) in TUI init, make entry_point optional
  • Plugin install --scope parsing when source precedes flag
  • Learnings --category flag parsing and palette query sync
  • Streaming bash path missing local executable bypass (#169)
  • Auto-scroll off-by-one in message render fast-skip
  • Load plugin MCP servers during TUI init and fix misleading install message
  • Register all protocol slash command aliases in TUI dispatch table
  • Command palette highlight breaks with args, cache case collision
  • Surface checkpoint errors as SystemMessage instead of propagating
  • Agent cancel race and invisible flat JSON sessions
  • Isolate crash lock per test to eliminate flaky parallel failures
  • Isolate test_config_save_and_load from parallel-test race
  • Remove dead _child param from ProcessManager::register — callers always pass None
  • Skip computer-use monitor tests gracefully on headless
  • Relax flaky test_config_validation assertion
  • Use org/repo for SWE-bench clone URL
  • Eval_bridge applies test_patch and uses testbed conda env
  • Eval_bridge Python-3.6 compat + batched pytest per file
  • Never-writer loop-guard stop ships empty patches
  • Exclude .bak/.orig/.rej droppings from SWE-bench patches
  • Tolerate truncated Codex patch envelopes, convert for git fallback
  • Apply_patch git fallback was a silent no-op; add patch(1) chain
  • Retry shared-cache clone over HTTP/1.1
  • Re-apply test patch after retry reset
  • Handle bare/non-pytest F2P names in swebench eval runners
  • Drop crashing pyarrow from qemu-emulated swebench eval images
  • Eval_bridge enforces PASS_TO_PASS per official SWE-bench metric
  • Swebench runner glob test-file exclusion + boundary-safe verify match
  • Editable-reinstall /testbed in both SWE-bench evaluators
  • Py3.5-safe subprocess decode + ill-formed P2P excluded from tally
  • Reserve loop-guard force-stop for actual repetition
  • Eval space-truncated parametrize ids poison whole pytest batch
  • Eval container cleanup timeout crashes whole batch
  • Honest retry budget, py_compile gate, failures-last verify tail
  • Derive turn-count soft cap from max_turns
  • Scope advisor blueprint to named tests and disclose its token budget
  • Extend never-writer guard to the standalone loop-guard branch
  • Escalate syntax warnings in the py_compile gate
  • Recover from pytest-rejected node ids in SWE-bench evaluators
  • Best-patch tie-break prefers the later attempt
  • Eval_bridge keeps skipped names in the tally denominator
  • Runner verify keeps skipped names in tally denominators
  • Advisor consults only retry transient transport errors
  • Thinking-continuation cap nudges to act before stopping
  • Score verify fails from full output before truncation
  • Never-writer guard counts only successful workspace writes
  • Suppress pytest warnings summary in verify output
  • Empty-retry emphasis restates the missing-feature path
  • Strip sample-instance tokens from swebench prompt
  • Eval_bridge unique container name prevents stale-verdict reads
  • Resolve docstring test names via official parse_log_django semantics
  • Keep warnings plugin loaded in eval (filterwarnings mark)
  • Snapshot workspace at retry start so analysis-only retries count as empty
  • Stop unproductive agent loops honestly
  • Make orchestration policies role-aware
  • Make benchmark verification ambiguity-safe
  • Honor server session mode for continuation
  • Retry docker verify once on infra errors before shipping unverified
  • Sanitize reused work dirs at instance start
  • Exclude .venv/pycache droppings from captured diffs
  • Stream team tool-loop turns for providers that drop non-streaming args
  • Team streamed tool loop must ask the provider to stream
  • Preserve diagnostics and classify provider limits
  • Re-key team tool schemas to canonical input_schema
  • Validate tool args under schema names (file_path/relative_path)
  • Reject unsupported lsp path alias
  • Bound provider awaits with a stall watchdog in team LLM client
  • Retry transient provider failures in team LLM gate
  • Harden filesystem and command trust boundaries
  • Bound autonomous process and retry lifecycles
  • Tighten skill and workflow validation
  • Support paginated MCP catalogs and case-insensitive tools
  • Bind MCP OAuth tokens to issuer origin
  • Name target file in write approval descriptions
  • Key bash approvals by command class
  • Keep tail messages reachable in auto-scroll
  • Preserve non-utf8 swebench paths
  • Harden cross-platform shell execution and updates
  • Handle shell terminal title escapes
  • Sanitize shell noise in streamed output
  • Run drift-guard git scan off the async runtime
  • Raise bash rate limiter ceiling for lib tests
  • mcp-oauth: Single-flight token refresh to stop rotation wipe-out
  • mcp-oauth: 0600-at-create token file + runtime-independent store
  • ws-bridge: Evict token totals when tasks end or are steered away
  • Route ws loop-iteration events to owning session
  • auth: Callback server survives port probes; validate before success page
  • Fail approval risk classification closed for unknown tools
  • Route REST plan events to owning session
  • Correct misleading pairing-nonce prune log
  • List MCP server tools from live connection, not a stub
  • Confine /workflow validate file paths to the workspace
  • Stop viz adapter fabricating success for failed tool executions
  • Stop re-sending full state snapshot on every chat input
  • Reflect loop outcome in transcript note
  • Bound until-check command output
  • Scan every escalation level for destructive -c scripts
  • Serialize viz_snapshot seq in the wire payload
  • Align key-point snippet offsets with original text
  • Split string-form allowed-tools into separate tools
  • storage: Keep backfill SQLite work off the async runtime thread
  • Skip unreadable files in conversation history search
  • Deterministic order for same-second decisions
  • Cleanup_old_sessions panic on huge day counts
  • Sort skill overviews for deterministic wire order
  • Unify skill description fallback between installer and registry
  • wiki: Remove FK cascade on changelog_entries to preserve entries on page deletion
  • Sort active skills and tiebreak recommendations by id
  • Sort skill presentations by name across dir loads
  • Prefer ancestor .git over local project markers in workspace root detection
  • Order category-filtered wiki pages by updated_at
  • Quote oauth url for cmd start on windows
  • Tighten guard secret-prefix and rm -rf detection
  • Fall through when interpreter lacks pytest module
  • Default sandbox writes to workspace, enforce denied carve-outs
  • Read marketplace installed versions from metadata
  • Match bazel marker on file name in build detection
  • Parse the sources label that generate_layered_summary renders
  • Assert PrePublish hook error aborts bus publish
  • Sync base schema build_system enum with serde BuildSystem
  • Reject oversized skill before eviction drains active set
  • Stop warning on expected skill state at session end
  • Ignore out-of-range skill level instead of granting max effort
  • Clean up partial marketplace install on clone or validation failure
  • Uninstall before reinstalling in marketplace update all
  • Ignore same-command loop when an edit landed between runs
  • Replace stale FTS rows when saving session snapshots
  • Isolate compaction snapshot disk test to a tempdir
  • Wire copilot device flow into login and make expiry tests falsifiable
  • Skip empty tokens when parsing YES/NO verdict in assertion engine
  • Rewind preview no longer claims untracked files are deleted
  • Dedupe interleaved same-depth skill discovery dirs
  • Verify TDD RED phase when the new test fails
  • Reject '.' as a skill name in uninstall
  • Drop stale cross-map entry on skill re-registration
  • Scope health report quality coverage to the catalog
  • Drop user messages emptied by orphaned tool-result filtering in conversation fixer
  • Make seatbelt sandbox execute real commands, default fail-closed
  • Base_score counts chars not bytes in task classification
  • Use Google's OpenAI-compat path for gemini model discovery
  • Deterministic session-log cluster reporting in wiki lint
  • Disambiguate forked page slugs on wiki ingest
  • Parse memory.yaml documents with serde_yaml multi-doc
  • Correct viz gauge units and retried-task rows
  • output-filter: Resolve per-tool level override in bridge dispatch
  • output-filter: Route unhandled cargo subcommands to generic spec
  • git: Keep paths containing ' in ' intact in merge-tree conflict parsing
  • Send tmux literals and Enter correctly across connector backends
  • Enforce chat inbound limit in chars and stop google chat history replay
  • Classify worktree deletions as dirty and seed prune age from mtime
  • Apply worktree prune max-age flag
  • Guard log10(0) in extractor confidence scoring
  • Reject path-like ids in workflow resume
  • Stop viz bridge holding session lock across blocking send
  • Bound web.fetch timeout for untrusted MCP clients
  • Stop advertising unimplemented browser tool params
  • Match written entry id in auto-memory sync exists-check
  • Honor retry include/exclude patterns in native bench runner
  • Preserve multibyte UTF-8 in native script adapter
  • Quote CSV error fields containing newlines
  • Honor history list --dataset filter
  • Exclude directory contents in trial artifact collection
  • Honor --retry for docker and bollard benchmark trials
  • Keep huge-ttl leases and widen execute_batch direction enum
  • Correlate stdio JSON-RPC responses by request id
  • Guard reversed multiselect ranges
  • Mask multibyte secrets by chars
  • Quote yaml leading indicators in skills create
  • Remove no-op cli flags
  • Concatenate a2a stream chunks in collect_stream
  • Match sudo and runas at token boundaries in guard R01
  • Fold DOT attributes into the statement list in to_dot
  • Classify verify results by test summary line
  • Parse actual tool output formats in simplify gate
  • Emit terminal TaskCompleted on escalate/stop exits
  • Make local test verification channel-independent and fail-closed
  • Run hook stdin write inside the per-hook timeout
  • Reject nested batch calls
  • Merge overlapping redactions in sanitize_for_log instead of picking the wider span
  • Stop read-only bash classifier vouching for arbitrary payloads
  • Lexical_normalize under-resolves double-parent after a component
  • Block full CGNAT range in SSRF validation
  • Lowercase find trigger in search strategy routing
  • Make cron persistence writes crash-atomic
  • Cron health_stats recent_errors now newest-first, deterministic
  • Remove hollow team tools that wrote unread artifacts
  • Delete zero-consumer FilteredRegistry with broken tag filter
  • Clamp code_context line window against stale cached index
  • Delete recipe builder methods that silently dropped their input
  • Delete never-registered StructuredOutputTool and its dead schema plumbing
  • Stop docker run stuffing foreground output into container_id
  • Serialize remaining RUSTYCODE_PATH_ROOT tests on shared lock
  • Send_message reports non-delivery without a transport
  • Worktree_create always failed (--detach conflicts -b); worktree_exit description promised branch deletion it never did
  • Kill timed-out task_retry check commands instead of orphaning them
  • Block smuggled cmdlets in powershell validation

Performance

  • Cache tool specs as Arc-shared snapshot to avoid per-turn cloning
  • Cache MCP tool catalogs to avoid redundant per-turn RPCs
  • Share one full clone per repo across swebench instances
  • Bare-name eval runs file::name nodes before whole-file fallback
  • rollout: Batch-drain event store writes to close the lag window

Refactor

  • Guard steered session task ownership
  • Remove dead provider_health module (zero consumers)
  • Extract agents/ module from TUI to orchestration
  • Extract memory modules from TUI to core
  • Extract service and app modules from TUI
  • Extract state modules from TUI to core
  • Extract pipeline engine from TUI to rustycode-pipeline
  • Slim pipeline — move browser tools, delete DAG stub
  • Delete rustycode-pipeline crate, absorb into orchestration
  • Generalize advisor blueprint grounding/scope/length wording
  • Strip hand-tuned methodology from SWE-bench driver prompt
  • Extract team tool loop
  • Extract llm planner
  • Extract final review workflow
  • Extract role lane adapters
  • Extract plan step execution
  • Extract team orchestrator lanes
  • Extract local capabilities from team executor
  • Extract plan templates from plan manager
  • Extract role prompts from team executor
  • Isolate tool argument coercion
  • Extract prompt rendering and builder from skill manager
  • Extract shell guard from ws-server router
  • Extract rest handlers from ws-server router
  • Extract socket handling from ws-server router
  • Split tool argument coercion by type
  • Extract harness command bodies
  • Isolate vector memory persistence
  • Isolate swebench workspace lifecycle
  • Extract prompt assembly and project guidance from conversation service
  • Split memory command into handlers, format, and store
  • Extract remote session lifecycle from service manager
  • Extract conversation streaming from service manager
  • Extract channel plumbing from service manager
  • Extract session handlers from server router
  • Extract operation handlers from server router
  • Extract remaining handlers from server router
  • Move workflow definition tests to own module
  • Extract validation impls from workflow definition
  • Move lsp client tests to own module
  • Extract lsp request methods from client
  • Move rollout tests to own module
  • Split rollout module into replay and writer
  • Move playbook store tests to own module
  • Extract rule queries from playbook store
  • Extract chat turn methods from session workflow
  • Extract agent role prompts from multi_agent
  • Isolate vector memory index
  • Move multi_agent tests to own module
  • Extract agent communication hub to own module
  • Move delegation tests to own module
  • Extract task role taxonomy to own module
  • Move bedd funnel tests to own module
  • Move ast recovery tests to own module
  • Move ast ledger tests to own module
  • Move ast progress store tests to own module
  • Move ast tree tests to own module
  • Move skill manager tests to own module
  • Move agent lifecycle tests to own module
  • Extract agent lifecycle types to own module
  • Move task decomposer tests to own module
  • Move agent profiler tests to own module
  • Move supervisor tests to own module
  • Move agent learning tests to own module
  • Move skill workflows tests to own module
  • Move git worktree tests to own module
  • Move state runtime tests to own module
  • Move plan manager tests to own module
  • Move team executor tests to own module
  • Move briefing tests to own module
  • Move execution trace tests to own module
  • Move config schema tests to own module
  • Move shared memory tests to own module
  • Move agent registry tests to own module
  • Move state derivation tests to own module
  • Move musician tests to own module
  • Move ensemble strategy tests to own module
  • Move fork join tests to own module
  • Move agent executor tests to own module
  • Move autonomy tests to own module
  • Move orchestration error tests to own module
  • Move verification gate tests to own module
  • Move loop executor tests to own module
  • Move layered prompt tests to own module
  • Move agent prompts tests to own module
  • Move session core tests to own module
  • Move session compaction tests to own module
  • Move session manager tests to own module
  • Move session message tests to own module
  • Move snip tier tests to own module
  • Move persist tier tests to own module
  • Move summarize tier tests to own module
  • Move piggyback tests to own module
  • Move ws bridge tests to own module
  • Move ws protocol tests to own module
  • Move shell guard tests to own module
  • Move chat relay tests to own module
  • Move slack transport tests to own module
  • Move viz updater tests to own module
  • Move turn tests to own module
  • Move thought frame tests to own module
  • Move completion gate tests to own module
  • Move session post-turn tests to own module
  • Move stuck loop tests to own modules
  • Move agent protocol tests to own module
  • Move workflow type tests to own module
  • Move protocol message tests to own module
  • Move command spec tests to own module
  • Move mode tests to own module
  • Move protocol hook tests to own module
  • Move intent tests to own module
  • Move task routing tests to own module
  • Move task contract tests to own module
  • Move conversation fixer tests to own module
  • Move briefing role tests to own module
  • Move layered memory tests to own module
  • Move memory index tests to own module
  • Move memory lib tests to own module
  • Move storage tests to own modules
  • Move core validation and storage tests to own modules
  • Move provider registry tests to own modules
  • Move marketplace client tests to own module
  • Move id tests to own module
  • Move tool registry tests to own modules
  • Move bench environment tests to own module
  • Move skill install and activation tests to own modules
  • Move bus tests to own module
  • Move output filter tests to own module
  • Move prompt lib tests to own module
  • Move orchestration module tests to own files
  • Move routing and telegram tests to own modules
  • Split task_dispatcher into directory module
  • Split validation into directory module
  • Split cross_platform and permission test modules
  • Split large modules and test suites
  • Split MCP client tests
  • Split autonomous test module
  • Split pipeline module into submodules
  • Split tools-api registry and cron_registry into directory modules
  • Extract agents module test suite
  • Split runtime lib.rs into focused impl modules
  • Split bench binary into cli and commands modules
  • Split llm_client into mock, provider_lane, tests modules
  • Extract headless test suite to its own module
  • Split mcp-core manager, types, server, oauth modules
  • Split mcp-core stdio_client, http_transport, tools, transport
  • Split tmux_control server into tool-handler modules
  • Extract native environment script helpers to scripts.rs
  • Split bundled skill tests into bundled/tests.rs
  • Extract ContentBlock into message/content_block.rs
  • Decompose execute_milestone into phase modules
  • Move headless error hints into ordered registry
  • Share tool hint diagnostics across surfaces
  • Centralize headless tool failure hints
  • Make tool hints cross-platform
  • Route headless through shared agent runtime
  • Dedupe frontend message conversion in session manager
  • Share noninteractive agent session setup
  • Reuse noninteractive session wiring
  • Share coding task config in tui harness
  • tools: God-object decomposition phase A — thread live tool registry, harden headless logs & marketplace checks
  • State true intent tie-break order, drop dead Thai prefix re-check
  • Wire router defaults to shared constants
  • agent-runtime: Privatize tool_registry_arc behind set_tool_registry
  • Delete hollow process manager
  • Delete wire-dead plugin exports
  • Drop discarded OrchestratorConfig retry knobs
  • Drop dead local-capability probes
  • Delete wire-dead egress_detector module
  • Delete wire-dead large_response module and bench
  • Delete wire-dead threat-pattern database
  • Delete wire-dead permission and permission_store modules
  • Delete stale duplicate check_tool_permission from tools-api
  • Delete dead select_tools_by_budget API
  • Delete wire-dead validation module
  • Delete wire-dead privilege module
  • Delete wire-dead AuditLog subsystem

Documentation

  • Record crate review checkpoints
  • Record benchmark job lifecycle fixes in PROGRESS.md
  • Add slash command architecture and alias conventions
  • Clarify drain.rs routing intent — tool/workspace bypass TuiEvent deliberately
  • Update progress log
  • Record orchestrator extraction progress
  • Record round 3 extraction progress
  • Record skill manager extraction progress
  • Record ws-server router decomposition
  • Record harness_cmd decomposition
  • Record swebench workspace extraction
  • Record conversation_service decomposition
  • Record memory command decomposition
  • Record service_integration decomposition
  • Record server router decomposition
  • Record workflow definition decomposition
  • Record lsp client decomposition
  • Record rollout module decomposition
  • Record ws-server session workflow split
  • Record multi_agent decomposition step 1
  • Record vector memory index extraction
  • Record multi_agent decomposition steps 2-3
  • Record delegation module decomposition
  • Record tail-test split batch
  • Record second tail-test split batch
  • Record third tail-test split batch
  • Record fourth split batch
  • Record fifth split batch
  • Record sixth split batch
  • Log seventh decomposition batch (pipeline, agents, runtime, bench bin)
  • Log eighth decomposition batch in PROGRESS.md
  • Log ninth batch and campaign exhaustion in PROGRESS.md
  • Log zsh OSC session-marker fix in PROGRESS.md
  • Note workspace gate blocked by concurrent orchestration WIP
  • Log execute_milestone decomposition session in PROGRESS.md
  • Log async-blocking bug-hunt session in PROGRESS.md
  • Log gate-green session and limiter flake fix in PROGRESS.md
  • Log hints registry session in PROGRESS.md
  • Log clean compaction-crate audit in PROGRESS.md
  • Log mcp-core OAuth token-store review in PROGRESS.md
  • Log ws-server bridge review in PROGRESS.md
  • Log ws-server loop-mode cross-session bleed fix in PROGRESS.md
  • Log clean compaction tier-boundaries audit in PROGRESS.md
  • Log rustycode-auth callback-server review in PROGRESS.md
  • Log ws-server approval.rs audit in PROGRESS.md
  • Log ws-server session-layer sweep round 2 in PROGRESS.md
  • Log ws-server session-layer sweep round 3 in PROGRESS.md
  • Add socket.rs findings to ws-server sweep round 3 in PROGRESS.md
  • Round 3 fix 5 in PROGRESS.md
  • Round 3 fix 6 in PROGRESS.md
  • Round 3 auth.rs + types.rs clearance
  • Record shell_guard su -c bypass fix in round-3 sweep
  • Record viz seq fix and close round-3 ws-server sweep
  • Record round-4 rustycode-session sweep
  • Log rustycode-storage backfill review in PROGRESS.md
  • Record rustycode-skill audit round 5
  • Log clean rustycode-git audit in PROGRESS.md
  • Record round-6 rustycode-storage audit in PROGRESS.md
  • Record glm-5.3/5.2/flash support + auth audit round 7
  • Glm zhipu provider commit landed, unblock round 7
  • Record storage decisions-ordering fix (round 8)
  • Record session cleanup panic fix (round 9)
  • Record round-10 rustycode-session clean audit
  • Record round-11 rustycode-skill sweep results
  • Record round-12 rustycode-prompt sweep (1 fix)
  • Record round-13 rustycode-wiki sweep in PROGRESS.md
  • Record round-14 rustycode-auth sweep in PROGRESS.md
  • Record round-15 rustycode-guard sweep in PROGRESS.md
  • Record round-16 rustycode-verification sweep
  • Record round-17 rustycode-sandbox sweep
  • Record round-18 rustycode-marketplace sweep
  • Record rustycode-lsp sweep (round 19)
  • Record round 20 (rustycode-memory full-crate sweep)
  • Record round 21 (rustycode-bus full-crate sweep)
  • Record round 22 (rustycode-config full-crate sweep)
  • Record round 23 (rustycode-skill full-crate sweep)
  • Record round 24 (rustycode-marketplace full-crate sweep)
  • Record round 25 (rustycode-verification full-crate sweep)
  • Record round 26 (rustycode-storage full-crate sweep)
  • Record round 27 (rustycode-session full-crate sweep)
  • Record round 28 (rustycode-prompt full-crate sweep)
  • Record round 29 (rustycode-verification clean audit)
  • Record round 30 (rustycode-auth full-crate sweep)
  • Record rustycode-session sweep (round 31)
  • Record rustycode-id clean audit (round 32)
  • Record round-33 verification sweep in PROGRESS.md
  • Record round 34 rustycode-learning clean-audit sweep
  • Record round-35 rustycode-storage sweep in PROGRESS.md
  • Record rustycode-skill core sweep (round 36 part 1)
  • Record rustycode-skill manager+activation sweep (round 36 part 2)
  • Record rustycode-skill sweep parts 3-4 (workflows, install, bundled)
  • Record rustycode-skill sweep part 5 (mid modules, tests) — crate complete
  • Record rustycode-marketplace sweep (round 37) — clean audit, 0 defects
  • Record round-38 rustycode-prompt audit (clean)
  • Record round-39 rustycode-protocol audit part 1 (clean)
  • Correct stale doc comments in protocol session and text modules
  • Record round-39 rustycode-protocol audit part 2
  • Fix plan and query_guard doc comments in protocol
  • Record protocol sweep part 3/5 in PROGRESS
  • Record protocol sweep part 4/5 in PROGRESS
  • PROGRESS round 40 — rustycode-sandbox sweep complete
  • Document the #[tool] macro as an args-ignoring stub
  • PROGRESS round 41 — rustycode-macros sweep complete
  • PROGRESS round 42 — classification sweep complete
  • PROGRESS round 43 — bus-bridge clean audit
  • Record round 44 rustycode-ui-model clean audit
  • Record round 45 rustycode-observability sweep
  • Round 46 — rustycode-executable sweep complete
  • Round 47 providers audit — gemini discovery endpoint fix
  • Round 48 wiki audit — lint determinism + ingest slug disambiguation
  • Add round-49 rustycode-memory sweep to PROGRESS
  • Add round-50 rustycode-tasks clean audit to PROGRESS
  • Add round-51 rustycode-agents clean audit to PROGRESS
  • Round 52 viz sweep complete
  • Round 53 acp sweep complete
  • Round 54 a2a sweep complete
  • Round 55 env sweep complete
  • Add round 56 thread-guard sweep to PROGRESS
  • Add round 57 ringbuf sweep to PROGRESS
  • Add round 58 litert sweep to PROGRESS
  • Correct ContentModified retry deadline comment in lsp client
  • Round 59 progress entry (rustycode-lsp sweep)
  • Round 60 progress entry (rustycode-ui-core sweep)
  • Round 61 output-filter sweep record
  • Round 62 rustycode-git sweep record
  • Add round 63 connector audit to PROGRESS
  • Add round-64 rustycode-chat sweep to PROGRESS
  • Add round-65 rustycode-server-client clean audit to PROGRESS
  • Add round-66 rustycode-server-protocol clean audit to PROGRESS
  • Add round-67 rustycode-shared-runtime clean audit to PROGRESS
  • Add round-68 rustycode-id clean audit to PROGRESS
  • Add round-69 rustycode-verification clean audit to PROGRESS
  • Add round-70 rustycode-server clean audit to PROGRESS
  • Add round-71 rustycode-bus clean audit to PROGRESS
  • Add round-72 rustycode-agent-management clean audit to PROGRESS
  • Add round-73 rustycode-session clean audit to PROGRESS
  • Add round-74 rustycode-runtime sweep results to PROGRESS
  • Round-75 rustycode-learning sweep record
  • Round-76 rustycode-ws-server sweep record
  • Record round-77 MCP servers + tool-server sweep
  • Record round 78 rustycode-core sweep
  • agent-runtime: Document tool-registry aliasing invariant
  • Record round 79 rustycode-bench sweep
  • Record round 80 rustycode-tauri clean sweep
  • Record round-81 mcp-core sweep in PROGRESS.md
  • Record round 82 cli sweep
  • Record round 83 auth sweep
  • Record round 84 tool-server sweep
  • Record round 85 macros deletion
  • Record round 86 core deferred deletions
  • Record round 87 a2a collect_stream fix
  • Record round 88 ui-model dead-path deletion
  • Record guard R01 token-boundary fix (round 89)
  • Record config CLEAN audit (round 90)
  • Record vector-memory CLEAN audit (round 91)
  • PROGRESS round 92 — workflow audit (DOT fix + tautology)
  • Add round-93 agent-runtime audit to PROGRESS
  • State that architect validate() is a stub
  • Log round 94 (team crate audit)
  • Correct ToolInfo attribution in CLAUDE.md
  • Log round 95 (tool-integration audit)
  • Add round-96 tool-executor audit entry to PROGRESS.md
  • Record round 97 (tools-security) in PROGRESS.md
  • Correct CronEntry timestamp units to epoch milliseconds
  • Fix check_stale new-file claim to match caller-owned behavior
  • Add codex adoption review — prioritized adopt/consider/skip list
  • Round-98 tools-api sweep summary; fix tools-api README drift
  • Sync READMEs and references with tool dispatcher rename

Testing

  • Make A2A redaction assertion lint-clean
  • Cover TUI double Escape input clearing
  • Add exhaustive mode delegation tests
  • Add compaction integration tests (12 tests)
  • Add verification integration tests (23 tests)
  • Add django docstring verifier sync tests
  • Assert wire tool schemas align with validator keys
  • Remove leftover delivery probe from bridge tests
  • Replace tautological rewind assertion with real coverage
  • Replace tautological from_dir assertion with falsifiable check
  • Make atomic-write leftover check falsifiable
  • Make session capture cap tests actually pin the caps
  • Exercise briefing truncated_to_budget with real builder-budget cases
  • Make analytics transport tests hermetic with local TCP sink
  • Isolate workspace_memory tests into per-test temp dirs
  • Make runtime creation assert success instead of a tautology
  • Exercise real 429 handling; drop dead pending_requests state
  • Delete literal-loop tautology in bridge tests
  • Assert whitespace-only expression parse fails
  • Delete tautological snip token test
  • Drop tautological profiler assertion
  • Rewire security_integration to the real validation API
  • Remove stale stub-facade comments, strengthen sandbox builder tests

Chores

  • Update lockfile for nix signal feature addition
  • Align Tauri application version
  • Delete dead code from TUI crate
  • Wire orphaned benchmarks, delete dead root benches/
  • Fix clippy 1.98 lint fallout
  • Replace remaining infallible unwraps
  • Fix leading-blank-line fmt drift in split test modules
  • git: Remove stale copies of tools git provider
  • Delete dead deadlock detector module
  • Delete dead runtime monitor module
  • Delete dead secure_env and tautology tests
  • Sync Cargo.lock with committed manifests
  • Delete dead macros crate
  • Delete dead ToolResultStorage spill-to-disk code
  • Delete dead RecoveryEngine and orphaned recovery modules
  • Delete dead ui-model accumulator and RunController
  • Delete security_patterns re-export shims

Security

  • Harden A2A server boundaries
  • Isolate ACP protocol transport
  • Harden provider authentication
  • Enforce permission and ID parsing invariants
  • Isolate server shell process trees
  • Fail closed for websocket approvals
  • Redact and atomically persist config
  • Bound chat webhook payloads
  • Bound verification output and verdicts
  • Fail closed when Landlock is unavailable
  • Tighten OAuth callback boundaries
  • Bound guard hook input