Dev to Main#97
Merged
Merged
Conversation
Reduce README from 1887 to ~275 lines. Move CLI / configuration / payment / security details into the existing docs/ tree. Add Documentation Map as the primary navigation surface. Fix three broken Quick Start commands (agent.provider / providers.openai.apiKey, serve not gateway, chat not run). See internal-docs/superpowers/specs/2026-05-23-readme-refactor-design.md for the design and internal-docs/superpowers/plans/2026-05-23-readme-refactor.md for the implementation plan. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Bare `./bin/lango` launches the Mission Workbench TUI; the multi-panel Cockpit dashboard is the explicit `./bin/lango cockpit` subcommand. README previously conflated the two. Update the Quick Start Run section to show both invocations and broaden the Architecture diagram's Surfaces row to reflect all three TUIs. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
7-track integration design under feature/adk-v1.3: - Track 0: deps bump v1.0.0 → v1.3.0 - Track A: wrapper slim-down (plugin.go delete + ThoughtSignature doc) - Track D: A2A v2 immediate migration - Track B: Skill Toolset adapter - Track C: Streaming function tools - Track E: Live API plumbing (Phase 1; Bubbletea cockpit Phase 1.5, audio Phase 2 deferred) - Track F: Channel voice infrastructure (Phase 1; per-channel integration Phase 2 deferred) - Track G (Discord realtime voice): deferred to future cycle Reviewed via reviewing-plan-artifacts-with-subagents (3 rounds on initial design, 2-3 rounds per track plan). Status Log captures completion dates for all tracks landed in this branch. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Upgrade ADK SDK three minor versions. v1.1 added FindAgent/FindSubAgent interface methods (transparent — Lango uses upstream agent factories, no custom implementations). v1.3 added model.LLMResponse transcription fields (additive, nil-safe). Transitive bumps include genai v1.57, a2a-go v0.3.15, new a2a-go/v2 v2.3.1, gRPC v1.81, OTel v1.43. Prerequisite for Tracks A-F (capability adoption). See internal-docs/superpowers/specs/2026-05-23-adk-v1.3-capability-first-design.md. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Compile-time tripwire that verifies the v1.3 API surface (Agent.FindAgent, LLMResponse transcription fields, tool/functiontool streaming, tool/skilltoolset, agent.LiveSession) is reachable. Guards against accidental downgrade or vendor drift. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…ndary Track A (wrapper slim-down) of ADK v1.3 capability-first design: A1 — ThoughtSignature handling: KEPT internal/adk/state.go round-trip. Initial plan called for removal because v1.3 added stream_aggregator ThoughtSignature propagation. Investigation showed upstream operates at the streaming layer (between Parts within one LLM response) while Lango's code operates at the persistence layer (ADK runtime ↔ Lango session storage). Orthogonal concerns. Removing Lango's code would break thought_signature continuity across session restarts. Added an explanatory block comment at functionCallToToolCall pointing to spec §4 Track A A1. A2 — plugin.go: DELETED. 290 LoC spike with zero production consumers (verified by grep). ADK v1.3 has a stable plugin API; if Lango adopts plugins later, a fresh design from v1.3 will be more accurate than the stale parity-gap doc the spike contained. Also removed the spike's test fragment from context_helpers_round_trip_agent_and_child_session_test.go (TestPluginConstructorsExposeCallbacksAndMapKeys + helper struct + interface assertion), preserving the file's 5 other valid tests. A3 — OTel reasoning attribute rename: no-op confirmed. grep returned zero hits for experimental.reasoning|reasoning_tokens|gen_ai.usage in Lango source. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Track D of the ADK v1.3 capability-first design. Switch internal/a2a/remote.go (+ remote_test.go) from google.golang.org/adk/agent/remoteagent (v1, now Deprecated) to .../v2: - A2AConfig.AgentCardSource string was removed in v2; replace with AgentCardProvider: remoteagent.NewAgentCardProvider(url), the v2 lazy URL/file resolver. Semantics equivalent (lazy resolution preserved). - All other A2AConfig fields, NewA2A() signature, and AgentCard handling are signature-compatible. internal/a2a/server.go is NOT migrated. It is Lango's custom AgentCard JSON server with P2P/pricing/ZK extensions, NOT an ADK adka2a executor. The previously-planned A2AClientProvider v1-fallback plumbing is unnecessary because Lango is an A2A *client*; outbound goes v2 immediately without affecting inbound peers. Verification: build clean; ./internal/a2a/... tests pass; whole-tree test suite matches Track 0 baseline; zero v1 agent/remoteagent imports remain in Lango source. a2a-go/v2 v2.3.1 entered the module graph as transitive // indirect during Track 0's go.mod bump; Track D does not promote it to direct. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Track B of the ADK v1.3 capability-first design. Wire ADK v1.2+ tool/skilltoolset into Lango so agents can discover markdown-defined skills from a configured directory via three tools: load_skill, list_skill_resources, load_skill_resource. Added: - internal/adk/skilltoolset.go: BuildSkillToolset(ctx, dir) helper using upstream skill.NewFileSystemSource + skilltoolset.New (no custom skill.Source — the upstream filesystem source already covers our requirements per agentskills.io frontmatter spec). - internal/adk/skilltoolset_test.go: 5 unit tests (empty / nonexistent / file-not-dir / valid / empty-dir). - internal/adk/agent.go: WithToolsets(...tool.Toolset) option plumbing into llmagent.Config.Toolsets. - internal/orchestration/orchestrator.go: Config.Toolsets field plumbed into the orchestrator's llmagent.Config.Toolsets. Sub-agents do NOT receive these toolsets — orchestrator only. - internal/config/types.go: AgentConfig.SkillsDir field (default empty = disabled). - internal/app/wiring.go: shared SkillToolset construction routed to BOTH single-agent (adk.WithToolsets) and multi-agent (orchCfg.Toolsets) paths. - assets/skills/example-skill/SKILL.md: demo skill with valid YAML frontmatter proving the wiring. .claude/skills/ migration is DEFERRED to a future change — existing files mix Claude-Code-specific frontmatter (with compatibility, metadata fields) and bare markdown. A clean migration requires deciding which Claude-Code fields to keep/strip per agentskills.io. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…reamingTool Track C of the ADK v1.3 capability-first design. Wire ADK v1.3's functiontool.NewStreaming[TArgs] into Lango so tool authors can yield partial results via iter.Seq2[string, error]. Added: - internal/agent/runtime.go: new StreamingToolHandler type and optional StreamingHandler field on Tool (nil-default — backward compatible). - internal/adk/tools.go: AdaptStreamingTool routes through NewStreaming when StreamingHandler is set, else falls back to functiontool.New. Per-call timeout and agent-name context injection work for both paths via the new injectCallContext helper. - internal/adk/tools_streaming_test.go: 3 unit tests covering streaming branch, fallback branch, and regression of non-streaming AdaptTool. Scope note: in non-live agent runs ADK aggregates yielded chunks into a single result string (base_flow.go:1106-1117). Real-time chunk surfacing in cockpit/turnrunner is deferred to Track E (Live API), where chunks flow to liveSess.Send. Track C only ships the adapter — non-live behavior is identical to a regular tool from the caller's perspective. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…r (Phase 1)
Track E Phase 1 of the ADK v1.3 capability-first design. Framework-
agnostic Live API plumbing; Bubbletea cockpit wiring is Phase 1.5,
audio I/O is Phase 2.
Added:
- internal/live/config.go: Config{Modalities, AudioSink, AudioSource,
MaxLLMCalls} + DefaultTextConfig() (text-only Live mode).
- internal/live/audio.go: AudioSink / AudioSource interfaces + NoOpSink /
NoOpSource defaults consumed by Phase 1 text mode. Phase 2 will swap
in malgo-backed implementations.
- internal/live/session.go: Session wrapper around runner.Runner.RunLive
with Send / Events / Close. Nil-safe Close; idempotent.
- internal/live/voicemode.go: VoiceMode state machine (Off / Active)
with Toggle / Send / Events / Close / IsActive / State / Indicator.
UI-framework-agnostic so Phase 1.5 can adapt to Bubbletea by wrapping
Toggle in tea.Cmd and bridging Events() → program.Send(liveEventMsg).
Local LiveExecutor interface lives here (in package live) to avoid the
import cycle that would arise if it lived in turnrunner.
- internal/live/session_test.go + voicemode_test.go: 11 unit tests
(5 session + 6 voicemode) with stubbed executor.
- internal/turnrunner/live.go: RunnerLiveExecutor default impl satisfies
live.LiveExecutor (compile-time check via
var _ live.LiveExecutor = (*RunnerLiveExecutor)(nil)).
Phase 1.5 (deferred — separate change):
- Ctrl+V keymap binding, tea.Cmd wrapping of VoiceMode.Toggle, goroutine
bridging Session.Events() → program.Send(liveEventMsg), header indicator.
Phase 2 (deferred — separate spec):
- malgo / portaudio CGO bindings for mic + speaker, PCM resampling,
genai.ModalityAudio, gemini-3.1-flash-live-preview.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…ni provider (Phase 1)
Track F Phase 1 of the ADK v1.3 capability-first design. Ships the
vendor-agnostic voice interfaces + Gemini-backed STT provider + config
plumbing. Per-channel integration is Phase 2.
Added:
- internal/voice/types.go: Audio{MimeType, Data}, STTOptions{Language},
TTSOptions{Language, Voice}, sentinel errors ErrNotConfigured /
ErrTTSNotConfigured.
- internal/voice/stt.go: SpeechToText interface.
- internal/voice/tts.go: TextToSpeech interface.
- internal/voice/noop.go: NoOpSTT / NoOpTTS defaults that return the
sentinel errors so callers can detect disabled-voice and fall back to
text-only behavior without crashing.
- internal/voice/gemini_provider.go: NewGeminiSTT (Gemini STT via
genai.Models.GenerateContent with audio Blob + transcribe prompt).
NewGeminiTTS returns ErrTTSNotConfigured wrapped — Phase 2 will
implement real Gemini TTS audio output.
- internal/voice/voice_test.go: 8 unit tests covering NoOp,
nil-client fallback, empty-audio validation, and sentinel-error
distinctness.
Modified:
- internal/config/types.go: VoiceConfig + VoiceChannelConfig +
VoiceCostGuard structs; Voice VoiceConfig field on top-level Config.
Defaults to disabled.
Phase 2 (deferred — separate change):
- Per-channel inbound: Telegram getFile → audio download →
voice.SpeechToText.Transcribe → inject as text message.
- Per-channel outbound: optional voice.TextToSpeech.Synthesize → sendVoice
/ file upload.
- Real Gemini TTS implementation.
Single-vendor by user decision: Gemini API only. Interface shape exists
so a future swap is not blocked.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Final comprehensive code review (post-merge) flagged two test gaps where the implementations were correct but the tests didn't exercise the spec-mandated behavior: Track C gate: spec §4 Track C said "sample streaming tool yielding N chunks". TestAdaptStreamingTool_WithStreamingHandlerReturnsStreamingTool only asserted construction. Added: - TestStreamingHandler_YieldsExpectedChunks: verifies the iterator yields the three expected chunks in order when consumed. - TestStreamingHandler_StopsOnConsumerCancel: verifies the iterator honors consumer early-termination (the contract AdaptStreamingTool's timeout-cancel closure relies on). Track B gate: spec §4 Track B said "an agent can call load_skill and receive its markdown body. Round-trip test asserts content equality." TestBuildSkillToolset_ValidSkillReturnsToolset only asserted the tool count was 3. Added: - TestSkillSource_RoundTripsSkillContent: writes a SKILL.md fixture, builds the upstream skill.NewFileSystemSource, then exercises ListFrontmatters + LoadFrontmatter + LoadInstructions to verify the metadata round-trips and the markdown body content is recoverable. This is the same data path that load_skill / list_skill_resources / load_skill_resource use; we test the layer Lango actually owns (skill.Source) without needing ADK's internal tool.Context. These tests close the spec gates without requiring access to ADK's internal/toolinternal package (which would be unreachable from outside the ADK module). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two follow-up artifacts captured by the final reviewing-code-changes-with-subagents pass on feature/adk-v1.3 (10 commits 6db812e..c4d7a37): 1. Cross-reference comments for the dual SkillsDir fields: - cfg.Agent.SkillsDir (Track B — ADK SkillToolset, agentskills.io frontmatter) - cfg.Skill.SkillsDir (pre-existing — knowledge module's internal registry) Both now point at each other so future readers don't conflate them. 2. Spec Status Log gains a "Post-merge follow-ups (non-blocking)" section listing 3 known scope expansions/future-phase items intentionally left out of this branch: - Agent.SkillsDir settings-form coverage (Track B Phase 1.5-equivalent) - SaveLiveBlob opt-in surface (Track E Phase 2) - End-to-end streaming tool test through ADK RunStream (Track C Phase 1.5) No production code touched; documentation only. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Per .claude/CLAUDE.md "Whenever you change core code, update all downstream
artifacts" — update user-facing docs for the new config + capabilities
landed in feature/adk-v1.3 (11 commits):
- docs/configuration.md:
- New `agent.skillsDir` row (Track B — ADK SkillToolset, agentskills.io)
- New `## Voice` section (Track F — voice.enabled, sttModel, ttsModel,
perChannel, costGuard) with Phase 1 / Phase 2 status note
- Skill section header reorganized to explain the dual subsystems
(knowledge skill registry vs ADK SkillToolset) and disambiguate
`skill.skillsDir` from `agent.skillsDir`
- docs/features/skills.md:
- Admonition at the top explaining the two independent skill systems
(knowledge registry, this page; ADK SkillToolset, separate)
- Cross-references to assets/skills/example-skill/ for the ADK side
- docs/features/channels.md:
- Phase 1 (voice infra shipped) / Phase 2 (per-channel integration
deferred) status note — honest about what works today vs Phase 2
- Tells users explicitly that voice.enabled is currently inert until
per-channel wiring lands
[[capability-honesty]]: docs accurately reflect what's implemented vs
what's deferred. No phantom-feature documentation.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
langowarny
added
bug
Resolve every warning so dev → main is gated on a clean lint run:
- unused (9): drop dead wrapper funcs/consts/stubs in
cli/alerts, cli/metrics, cli/chat/input, cli/p2p/sandbox_test,
cli/status, plus the now-unused strings import in cli/chat/input.
- ineffassign (5): retire ineffectual writes in tuicore form,
knowledge retriever, mcp server-connection, p2p firewall, and
testutil mock-graph-store tests; discard append result with _.
- govet testinggoroutine (1): replace t.Fatal inside a non-test
goroutine with a buffered channel that the test inspects after
watchServeSignals returns.
- govet structtag (1): drop the unused json tag on the private
field in TestRedactConfigGetReflectValue — the redaction logic
already gates on field.PkgPath, so tag presence was incidental.
- staticcheck (21):
QF1011 — remove redundant interface type in ADK v1.3 smoke test
QF1001 — clear double-negation in recovery prompt + testutil
doc/stream guards
QF1008 — drop embedded selectors in runledger writethrough
and docker_runtime test
S1011 — replace per-element loop with append(slice, batch...)
in graph admission
S1031 — drop unnecessary nil checks around map ranges in mcp
config_loader
SA4006 — stop reassigning page when the result is never read
in cockpit mission-control branch tests
ST1008 — reorder executeExtensionCmd return to (string, int,
error) and update all 9 call sites
SA1019 — annotate legacy P2P.KeyDir / Embedding.ProviderID
regression tests with //nolint:staticcheck + rationale (these
tests intentionally pin deprecated-field behavior)
SA1012 — annotate the nil-parent submitCmd test with
//nolint:staticcheck (code path is the test's subject)
Verified: go build, go vet, golangci-lint, and targeted go test on
all touched packages all clean. README docs-completeness guard
failures in internal/testutil pre-date this work and are tracked
separately.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…docs/cli
Two requested cleanups so dev → main goes in green on a stock host:
CI
- Remove the `docs-version-check` job from .github/workflows/ci.yml.
It grep-enforced ADK version strings across docs/ and is no longer
the gate we want — version sync now lives in go.mod + release notes.
Docs guard tests
- The README was refactored into a landing page (docs(readme): refactor
into landing page with docs/ navigation); the README-as-CLI-quickref
layout it guarded no longer exists. Repoint every surviving
internal/testutil README guard to docs/cli/index.md (or the matching
docs/cli/<topic>.md) where the actual quick-reference rows now live.
- Drop README halves of `TestArchitectureAndREADMEUseCurrent*` inventory
guards (Memory, Contract) — the architecture half still pins the
inventory; the README half pinned a compact form that the new README
intentionally no longer carries.
- Delete the package/architecture inventory guards that hard-coded the
old plain-text README package map (`pkg/ # description`). The new
docs/architecture/project-structure.md uses markdown tables, so these
guards have no current target and rewriting them is its own task:
docs_quality_guard_test.go: 19 stale Architecture/README package-
inventory tests removed.
smartaccount_docs_inventory_guard_test.go: removed
(TestSmartAccountDocsUseCurrentCommandInventory only checked the
deleted README compact form; architecture-doc coverage lives in
docs_quality_guard_test.go).
- Rewrite `TestREADMEP2PGitSummariesStayTruthful` against the new
markdown table rows in docs/cli/index.md.
- For multi-target tests (p2p, metrics, provenance), drop the now-stale
README target rather than letting it shadow real coverage.
Locale-stable git invocations
- internal/p2p/gitbundle: introduce `gitCommand(ctx, args...)` helper
that forces `LC_ALL=C` + `LANG=C`, and route every git invocation
through it (runGit, CreateBundle, UnbundleAll, CreateIncrementalBundle,
VerifyBundle). The "empty bundle" / "missing prerequisite" detection
was string-matching English stderr from git, which broke on Korean-
locale hosts (저장소에 필수적인 다음 커밋이 없습니다 etc). With C
locale forced, the existing English string checks are stable.
Verified: go build, go vet, golangci-lint (still 0 issues), and
go test ./... all clean on a Korean-locale macOS host.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Failures only surfaced under CI's stricter environment (Linux + macOS runners, `-race`, fresh hosts without dev-machine state). Address each root cause rather than skip: 1) archtest depends on `rg` `internal/archtest/bootstrap_boundary_test.go` shells out to ripgrep for the package-boundary scans. CI runners did not have rg installed, so all 10 boundary tests failed at "executable file not found in $PATH". Add `ripgrep` to both the Linux (`apt-get install`) and macOS (`brew install`) dependency steps in .github/workflows/ci.yml, and also to the lint job. 2) handshake test pinned `time.Local` `time.Unix(1_700_000_000, 0)` returns a Time with Location set to `time.Local`. The handshake responder caches the bundle and the round-trip normalizes the Location to `time.UTC`. `assert.Equal` compares the *Location pointer*, so Local≠UTC even when both evaluate to the same instant on a UTC host. Pin the test's CreatedAt to `.UTC()` so both sides agree on Location. 3) handshake `Initiate`/`HandleIncoming` could hang past their timeout `TestInitiateCachesV2BundleAndRegistersAlias` hit a 10-minute test timeout in CI because `json.Decoder.Decode` does not observe ctx cancellation. When the peer (real or simulated) stalls, `cfg.Timeout` was effectively ignored and the test wedged on `net.Pipe.read`. Set `stream.SetDeadline(deadline)` from the `WithTimeout` deadline in both Initiate and HandleIncoming so the underlying network read terminates with the rest of the handshake. Best-effort: ignore errors from SetDeadline since not all stream implementations support it. 4) `internal/tools/exec` PTY race on timeout `RunWithPTY` launches a goroutine that `io.Copy`s into `output bytes.Buffer`, then on `ctx.Done()` reads `output.String()` while the goroutine may still be writing — race-detector flags `Buffer.grow()` vs `Buffer.String()` concurrent access. Wait on the copy goroutine's `done` channel after SIGTERM before reading the buffer. 5) glamour `BlockStack` race in chat markdown renderer `internal/cli/chat/markdown.go` cached a single `glamour.TermRenderer` in a package-level variable on the assumption that bubbletea serializes message dispatch. Tests (notably parallel `cli/cockpit/pages` tests) invoke `renderMarkdown` from multiple goroutines, and glamour's internal `BlockStack` is not safe for concurrent use. Guard the cached renderer + the render call with a `sync.Mutex`. 6) gitbundle `commit-tree` lacks author identity on fresh CI hosts `branch.go` merge runs `git commit-tree` against a bare repo. Bare repos do not pick up a user.name/user.email config, and CI runners have no global git identity, so the command fails with "Author identity unknown". Pass `GIT_AUTHOR_NAME`/`GIT_AUTHOR_EMAIL` and the matching `GIT_COMMITTER_*` env vars (using a stable `Lango Bundle <bundle@lango.local>` identity) so commit-tree always has the metadata it needs. 7) streamx `TestParallelReadOnlyExecutor_ContextCancellation` deadlock The test wired two parallel handlers to send a "started" signal on an *unbuffered* channel before waiting on `<-ctx.Done()`. The test reads `started` exactly once, so the second handler's send blocks forever and never reaches the ctx.Done() arm — even after cancel(). Under high race-detector overhead the test sometimes hung the full suite for >10 minutes. Buffer `started` to the handler count. Verified: targeted `go test -race ./internal/streamx/... ./internal/ tools/exec/... ./internal/p2p/handshake/... ./internal/p2p/gitbundle/ ... ./internal/cli/cockpit/pages/... ./internal/cli/chat/... ./internal /archtest/...` clean; golangci-lint still 0 issues. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Large integration release rolling up ~3 months of work on dev. Highlights span
agent runtime, voice/live channels, teammate platform, cockpit operator UX,
graph admission telemetry, and a sweeping coverage push.
Agent Runtime
ThoughtSignature boundary documented, plugin.go spike removed.
Tool.StreamingHandler+AdaptStreamingTool.internal/voiceSTT/TTSinterfaces with a Gemini provider and noop fallback.
internal/liveSession, VoiceMode,LiveExecutor.
remoteagent/v2.Teammate Platform
cutover), with approval identity policy and runledger mirroring of
approval-blocked state.
from instructions.
Cockpit & Operator UX
durable-first reads, write paths, collaboration context, proposal
registry (prepare/accept/dismiss), operating loop agenda.
langoroutes to workbench.(subtype/reason/dispatch/actor/time/family), retry, reset, summary
strips (actor, dispatch, family), grouped summaries, refresh recovery.
Graph / Telemetry
write-failure baseline event, unmapped admission source telemetry,
extractor drop telemetry.
Reliability / Runtime
bridge API.
appboundary (refactor: decouple cli from app boundary).Security
shutdown lifecycle hardened, sandbox worker exit codes, doctor bind
failure classification.
P2P / Knowledge
retriever residual coverage; FTS/vector path consolidation.
Test Coverage
"slices") targeting ADK, app wiring, runledger, p2p, security,
ontology, cockpit, smartaccount, etc. Spec gate tests added for ADK
v1.3 Tracks B/C.
Docs
docs/navigation; Workbenchvs Cockpit distinction clarified.
docs/superpowers/.adjudication, knowledge exchange, identity/trust).