merge:main into release

.claude/memory.md

@@ -216,7 +216,7 @@ Quick reference for anyone starting with Claude on this project. Updated by the
 
 ## Artifacts Domain (Issue #2776)
 
-- **Filesystem-backed persistence, no SQLite** — `src/openhuman/artifacts/` stores JSON metadata (`meta.json`) + binary blobs under `<workspace_dir>/artifacts/<uuid>/`. Pattern mirrors `memory/ops/files.rs` but simpler.
+- **Filesystem-backed persistence, no SQLite** — `src/openhuman/agent/artifacts/` stores JSON metadata (`meta.json`) + binary blobs under `<workspace_dir>/artifacts/<uuid>/`. Pattern mirrors `memory/ops/files.rs` but simpler.
 - **`"ai"` namespace in controller registry** — RPC methods are `openhuman.ai_list_artifacts`, `openhuman.ai_get_artifact`, `openhuman.ai_delete_artifact`. Future `ai_*` methods should use this same namespace.
 - **Two-layer path validation required** — (1) `validate_artifact_id` rejects empty strings, `/`, `\`, `..`, absolute Unix paths, Windows `C:` and UNC `\\` paths; (2) `assert_within_root` canonicalizes and checks containment. Replicate this pattern for any new filesystem-backed domain.
 - **`cargo test --lib` required for lib crate tests** — `cargo test -p openhuman -- "artifacts"` lists tests but filters to 0. Must use `cargo test -p openhuman --lib -- "artifacts"` because tests are in the lib crate, not integration test binaries.
@@ -310,7 +310,7 @@ Quick reference for anyone starting with Claude on this project. Updated by the
 
 ## Memory Sync Sources — Defaults ON + Per-Source UI (Issue #3293)
 
-- **Conservative caps registry** — `composio_defaults_for_toolkit(toolkit) -> (max_items, sync_depth_days)` in `src/openhuman/memory_sources/registry.rs` is the single source of truth. Values: gmail 100/30, slack 50/14, notion 30/30, linear 50/30, clickup 50/30, github 50/30, generic 30/14. Non-Composio defaults (GithubRepo 10PR/10issue/50commit, RSS 20, Twitter 7d) live in `apply_kind_defaults` in `rpc.rs`.
+- **Conservative caps registry** — `composio_defaults_for_toolkit(toolkit) -> (max_items, sync_depth_days)` in `src/openhuman/memory/sources/registry.rs` is the single source of truth. Values: gmail 100/30, slack 50/14, notion 30/30, linear 50/30, clickup 50/30, github 50/30, generic 30/14. Non-Composio defaults (GithubRepo 10PR/10issue/50commit, RSS 20, Twitter 7d) live in `apply_kind_defaults` in `rpc.rs`.
 - **`upsert_composio_source` now defaults ON** — Registers `enabled: true` with caps applied from the registry. Previously registered `enabled: false` with no caps.
 - **Cap enforcement: 3 construction sites, not 1** — `ProviderContext` (`memory_sync/composio/providers/types.rs`) carries `max_items`/`sync_depth_days`. All three sites must populate caps from the registry entry: `composio/mod.rs run_connection_sync`, `composio/periodic.rs`, `composio/bus.rs`. Each provider reads `ctx.max_items`/`ctx.sync_depth_days` for pagination + date clamping. Shared helpers: `pages_for_max_items` / `epoch_floor_from_depth` in `providers/helpers.rs`.
 - **First-sync gotcha on new connections** — `bus.rs on_connection_created` fires BEFORE `upsert_composio_source`, so caps are (None, None) on the brand-new connection's first sync — bounded only by internal `MAX_PAGES`, not registry caps. Documented in code; intentional.

.env.example

@@ -118,6 +118,12 @@ OPENHUMAN_TEMPERATURE=0.7
 # research turns). The window resets on every token, so valid long responses are
 # never cut. Leave unset to use the default.
 # OPENHUMAN_INFERENCE_STREAM_IDLE_TIMEOUT_SECS=
+# [optional] Wall-clock budget (milliseconds) for the store-time GET /auth/me
+# validation during sign-in (default 12000). Kept well under the desktop sign-in
+# RPC timeout so a reachable-but-slow backend fails fast into deferred
+# revalidation for a live-exp JWT instead of hanging until the RPC bounces the
+# user back to sign-in (#5166). Leave unset to use the default.
+# OPENHUMAN_AUTH_ME_STORE_TIMEOUT_MS=12000
 # [optional] Headless update restart contract: self_replace | supervisor
 # OPENHUMAN_AUTO_UPDATE_RESTART_STRATEGY=self_replace
 # [optional] Allow bearer-authenticated RPC callers to invoke update.apply/update.run
@@ -174,6 +180,18 @@ OPENHUMAN_WEB_SEARCH_TIMEOUT_SECS=10
 # QUERIT_API_KEY=
 # OPENHUMAN_QUERIT_API_KEY=
 
+# ---------------------------------------------------------------------------
+# Exa search (direct API, BYOK — https://exa.ai)
+# ---------------------------------------------------------------------------
+# [optional] API key from https://exa.ai. Select "Exa" in
+# Settings > Search engine, or set OPENHUMAN_SEARCH_ENGINE=exa. Calls go
+# straight to https://api.exa.ai with this key (never through the managed
+# backend) and power web_search_tool plus exa_search / exa_find_similar /
+# exa_get_contents. Engine silently falls back to "managed" if no key is
+# present, so the agent always has working search.
+# EXA_API_KEY=
+# OPENHUMAN_EXA_API_KEY=
+
 # ---------------------------------------------------------------------------
 # Brave search (direct API — https://brave.com/search/api/)
 # ---------------------------------------------------------------------------

.github/workflows/build-desktop.yml

@@ -183,7 +183,7 @@ jobs:
             libnss3 libnspr4 libatk1.0-0 libatk-bridge2.0-0 libcups2 libdrm2 \
             libxkbcommon0 libxcomposite1 libxdamage1 libxfixes3 libxrandr2 \
             libgbm1 libpango-1.0-0 libcairo2 libatspi2.0-0 libxshmfence1 libu2f-udev \
-            xvfb
+            xvfb dbus dbus-x11
           if [ "$MATRIX_TARGET" = "x86_64-unknown-linux-gnu" ] \
             && ! command -v apparmor_parser >/dev/null 2>&1; then
             sudo apt-get install -y apparmor

.github/workflows/ci-lite.yml

@@ -148,8 +148,11 @@ jobs:
               - 'src/**'
               - 'tests/**'
               - 'scripts/ci-cancel-aware.sh'
+              - 'scripts/check-linux-tls-dependencies.sh'
               - 'scripts/test-rust-e2e.sh'
               - 'scripts/test-rust-with-mock.sh'
+              - 'vendor/motosan-ai-oauth/**'
+              - 'vendor/tinychannels'
             # Changes here invalidate per-module test scoping → full suite.
             rust-core-full:
               - '.github/workflows/ci-lite.yml'
@@ -160,6 +163,9 @@ jobs:
               - '.cargo/config.toml'
               - 'rust-toolchain.toml'
               - 'scripts/ci-cancel-aware.sh'
+              - 'scripts/check-linux-tls-dependencies.sh'
+              - 'vendor/motosan-ai-oauth/**'
+              - 'vendor/tinychannels'
             rust-core-src:
               - 'src/**'
               - 'tests/**'
@@ -296,7 +302,12 @@ jobs:
         env:
           NODE_ENV: test
           FULL: ${{ needs.changes.outputs['frontend-full'] }}
-          CHANGED_FILES: ${{ needs.changes.outputs['frontend-src-files'] }}
+          # Same argument-limit guard as the rust-core lane below. The condition is
+          # inverted so the file list sits on the `&&` (truthy) side: GitHub's
+          # `a && b || c` ternary falls through to `c` whenever `b` is falsy, so
+          # `... == 'true' && '' || <files>` would hand back the full list in
+          # exactly the full-suite case this is meant to blank.
+          CHANGED_FILES: ${{ needs.changes.outputs['frontend-full'] != 'true' && needs.changes.outputs['frontend-src-files'] || '' }}
 
       - name: Normalize lcov source paths to repo root
         if: needs.changes.outputs.frontend == 'true'
@@ -349,6 +360,10 @@ jobs:
       - name: Check Rust formatting
         run: cargo fmt --all -- --check
 
+      - name: Enforce Linux TLS dependency policy
+        if: needs.changes.outputs['rust-core'] == 'true' || needs.changes.outputs['rust-tauri'] == 'true'
+        run: bash scripts/check-linux-tls-dependencies.sh
+
       - name: Run clippy (core crate)
         if: needs.changes.outputs['rust-core'] == 'true'
         run: bash scripts/ci-cancel-aware.sh cargo clippy -p openhuman -- -D warnings
@@ -409,7 +424,7 @@ jobs:
       - name: Check core builds with the default domain gates disabled
         # Lib-scoped on purpose. `--all-targets` would additionally compile the
         # `tests/` integration crates, which today name gated symbols
-        # (voice::reply_speech::test_seam, mcp_registry::store, x402::tools) without
+        # (voice::reply_speech::test_seam, mcp::registry::store, web3::x402::tools) without
         # a #[cfg] and so do not compile with the gates off. That cleanup is tracked
         # in #5021 alongside running the full gates-off suite; until it lands, this
         # lane checks the lib and RUNS the gate-contract lib tests below.
@@ -430,7 +445,36 @@ jobs:
         run: |
           bash scripts/ci-cancel-aware.sh cargo test --manifest-path Cargo.toml \
             --no-default-features --features tokenjuice-treesitter --lib -- \
-            core::all:: core::cli:: core::jsonrpc:: core::legacy_aliases:: core::runtime:: agent_registry::agents::loader::
+            core::all:: core::cli:: core::jsonrpc:: core::legacy_aliases:: core::runtime:: agent::registry::agents::loader:: memory::people::address_book::tests:: openhuman::platform::socket::event_handlers:: tools::schemas:: tools::ops::tests::
+          bash scripts/ci-cancel-aware.sh cargo test --manifest-path Cargo.toml \
+            --no-default-features --features tokenjuice-treesitter,mcp --lib -- \
+            mcp::server::resources::
+
+      - name: Guard — the kernel dependency floor may not grow
+        # The `--no-default-features --features flows` profile is the surface a
+        # second host would embed, and left unmeasured it grows: seven heavy
+        # dependencies (git2/vendored-libgit2, rusqlite/bundled, enigo, arboard,
+        # ethers-*, tokio-tungstenite) are unconditional today, and none of them
+        # would likely have landed that way had a number moved in CI when they
+        # did. This lane is that number.
+        #
+        # The ratchet is bidirectional on purpose: it fails on growth AND on a
+        # shed that was not written back into scripts/kernel-floor.limits, since
+        # an unratcheted improvement silently grows back. Lower the limit in the
+        # same PR that sheds the crates.
+        run: bash scripts/check-kernel-floor.sh --verbose
+
+      - name: Guard — the removal simulator still agrees with cargo
+        # scripts/dep-sim.py projects the effect of cutting a dependency cohort,
+        # and every gating decision in the kernelization program is sized with
+        # it. It parses `cargo tree` precisely so it cannot drift from cargo's
+        # own feature resolution — an earlier version walked `cargo metadata`'s
+        # resolve graph instead and over-reported by 36 crates, counting
+        # dev-dependencies and unenabled target-specific edges.
+        #
+        # This asserts the calibration still holds. If it fails, every
+        # projection built on the simulator is suspect until it is fixed.
+        run: python3 scripts/dep-sim.py --cut-nothing --expect-names 285
 
       - name: Guard — new feature-gated test modules must be acknowledged
         # Self-maintaining coverage: the set of source files that #[cfg]-gate a test on
@@ -444,26 +488,32 @@ jobs:
           EXPECTED=$(cat <<'EOF'
           core/all_tests.rs
           core/cli_tests.rs
+          core/dispatch.rs
           core/jsonrpc_tests.rs
           core/legacy_aliases.rs
           core/runtime/services.rs
           openhuman/agent/harness/builtin_definitions.rs
           openhuman/agent/harness/definition_tests.rs
           openhuman/agent/harness/session/tests.rs
           openhuman/agent/harness/subagent_runner/tool_prep.rs
-          openhuman/agent_registry/agents/loader.rs
-          openhuman/composio/action_tool.rs
+          openhuman/agent/registry/agents/loader.rs
           openhuman/inference/local/mod.rs
-          openhuman/tool_registry/ops_tests.rs
-          openhuman/tool_registry/schemas.rs
+          openhuman/memory/people/address_book.rs
+          openhuman/mcp/server/resources.rs
+          openhuman/platform/socket/event_handlers.rs
+          openhuman/platform/socket/ops.rs
+          openhuman/tinyplace/manifest.rs
+          openhuman/tools/registry/ops_tests.rs
+          openhuman/tools/registry/schemas.rs
           openhuman/tools/ops_tests.rs
+          openhuman/tools/schemas.rs
           openhuman/voice/compile_status.rs
-          openhuman/wallet/stub.rs
           openhuman/web3/stub.rs
-          openhuman/x402/stub.rs
+          openhuman/web3/wallet/stub.rs
+          openhuman/web3/x402/stub.rs
           EOF
           )
-          ACTUAL=$(grep -rlE '#\[cfg\((not\()?feature = "(voice|media|web3|meet|mcp|skills|flows|channels)"' src --include='*.rs' \
+          ACTUAL=$(grep -rlE '#\[cfg\((not\()?feature = "(voice|media|web3|meet|mcp|skills|flows|channels|prediction-markets|contacts)"|#\[cfg\((not\()?all\([^]]*feature = "contacts"' src --include='*.rs' \
             | xargs grep -lE '#\[test\]|#\[tokio::test\]|fn .*_test' 2>/dev/null | sed 's|^src/||' | sort -u)
           if ! diff <(echo "$EXPECTED" | sed 's/^ *//' | sort -u) <(echo "$ACTUAL"); then
             echo "::error::Gated-test file set changed. Update the EXPECTED allowlist in the rust-feature-gate-smoke lane, and extend the scoped 'cargo test' filter if the new module can carry an ungated-assert regression (see #5022)."
@@ -592,7 +642,14 @@ jobs:
         env:
           CARGO_BUILD_JOBS: "1"
           FULL: ${{ needs.changes.outputs['rust-core-full'] }}
-          CHANGED_FILES: ${{ needs.changes.outputs['rust-core-src-files'] }}
+          # Suppressed in full-suite mode. The list is ignored when FULL=true, but a
+          # large refactor can push it past the container job's argument/env limit,
+          # and `docker exec` then fails to START the step —
+          # "Argument list too long" — before the script can apply its own
+          # MAX_CHANGED_FILES fallback. Blanking it here keeps the step launchable.
+          # Condition inverted so the file list is on the `&&` side — `a && '' || b`
+          # always falls through to `b` because `''` is falsy.
+          CHANGED_FILES: ${{ needs.changes.outputs['rust-core-full'] != 'true' && needs.changes.outputs['rust-core-src-files'] || '' }}
           OUT: lcov-core.info
           # The tinyagents harness is the agent engine on every build now
           # (issue #4249); the suite exercises it by default. The legacy engine

.github/workflows/release-production.yml

@@ -401,7 +401,7 @@ jobs:
       # fork the core image doesn't need. The Dockerfile COPYs vendor/ because
       # [patch.crates-io] resolves Rust SDK crates from vendor/.
       - name: Init vendored Rust submodules
-        run: git submodule update --init --recursive vendor/tinyagents vendor/tinyflows vendor/tinycortex vendor/tinyjuice vendor/tinychannels vendor/tinyplace
+        run: git submodule update --init --recursive vendor/tinyagents vendor/tinyflows vendor/tinycortex vendor/tinyjuice vendor/tinychannels vendor/tinyplace vendor/tinyhumans-sdk
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v4
       - name: Log in to GHCR

.github/workflows/release-staging.yml

@@ -297,7 +297,7 @@ jobs:
       # fork the core image doesn't need. The Dockerfile COPYs vendor/ because
       # [patch.crates-io] resolves Rust SDK crates from vendor/.
       - name: Init vendored Rust submodules
-        run: git submodule update --init --recursive vendor/tinyagents vendor/tinyflows vendor/tinycortex vendor/tinyjuice vendor/tinychannels vendor/tinyplace
+        run: git submodule update --init --recursive vendor/tinyagents vendor/tinyflows vendor/tinycortex vendor/tinyjuice vendor/tinychannels vendor/tinyplace vendor/tinyhumans-sdk
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v4
       - name: Build image (no push)

.github/workflows/test-reusable.yml

@@ -175,7 +175,12 @@ jobs:
               sort
           }
 
-          bash scripts/ci-cancel-aware.sh cargo test -p openhuman --lib --bins
+          # `--features bin-tools` is load-bearing, not cosmetic: the six binaries
+          # under src/bin/ now declare `required-features = ["bin-tools"]`, so
+          # without it `--bins` silently matches NOTHING and this lane goes green
+          # having compiled none of them. fleet.rs and slack_backfill.rs each carry
+          # a `#[cfg(test)] mod`, so those tests would vanish too, with no error.
+          bash scripts/ci-cancel-aware.sh cargo test -p openhuman --lib --bins --features bin-tools
           bash scripts/ci-cancel-aware.sh cargo test -p openhuman --doc
 
           while IFS= read -r target; do

.gitignore

@@ -135,3 +135,6 @@ CHANGELOG.preview.md
 *.diff
 
 .claude/worktrees/
+
+# Python bytecode from scripts/dep-sim.py
+__pycache__/

.gitmodules

@@ -23,3 +23,6 @@
 [submodule "vendor/tinyplace"]
 	path = vendor/tinyplace
 	url = https://github.com/tinyhumansai/tiny.place.git
+[submodule "vendor/tinyhumans-sdk"]
+	path = vendor/tinyhumans-sdk
+	url = https://github.com/tinyhumansai/sdk.git