@jinn-network/client 0.1.6 → 0.1.7-canary.17a8ecb8

package/CHANGELOG.md

@@ -2,6 +2,39 @@
 
 ## Unreleased
 
+### Embedded agent surface hidden by default (issue #326)
+
+- **The embedded Claude agent chat surface no longer renders in the operator
+  app by default.** The right-rail agent panel (running mode) and the
+  "Ask Claude" panel (onboarding) are hidden. The 2026-05-19 v0.1.6 dogfood
+  found the surface isn't robust enough for first-time operators — it is the
+  only daemon-side surface that strictly requires Claude auth, and its
+  action-authority / plugin-scope shape is still in design (#177 / #178).
+- **Feature flag: `JINN_ENABLE_EMBEDDED_AGENT`.** Set
+  `JINN_ENABLE_EMBEDDED_AGENT=1` (also accepts `true`) to re-enable the
+  surface for development. Default is off. When off, the daemon does not
+  mount the `/api/agent/ws` bridge and the SPA renders no agent panel; when
+  on, the dev-time path works end-to-end as before. (Issue #367: the SPA now
+  reads this flag through the injected `window.__JINN_FEATURES__`, the same
+  channel as the plug-in builder UI flag — no behaviour change for operators.)
+- **Claude-Code-as-a-solver-harness is unaffected.** Operators can still pick
+  Claude Code as their SolverNet harness — that path is independent of the
+  embedded chat surface and its WebSocket bridge.
+
+### Operator app
+
+- **Plug-in builder UI surfaces hidden by default (issue #327).** The `/build`
+  route and the Build top-tab no longer appear in the operator app. Hitting
+  `/build` by direct URL redirects to `/overview`. The plug-in substrate stays
+  fully live — `jinn solver-plugins publish` / `revoke`, the Ponder indexer,
+  the Discovery API endpoints, and the `client/docs/build/` tree are
+  unchanged, so direct-CLI builders following the docs are not blocked. The
+  surfaces are gated to keep the operator-app first-run UX focused; they will
+  be re-promoted once that UX is solid and plug-in indexing is end-to-end
+  populated. Set `JINN_ENABLE_PLUGIN_BUILDER_UI=1` on the daemon to re-enable
+  the full builder surface for development and design work — the daemon injects
+  the flag into the SPA via `window.__JINN_FEATURES__`.
+
 ### SWE-rebench v2 admission
 
 - **Admission semantics bumped from `'2'` → `'3'`.** Operators running the

package/README.md

@@ -312,7 +312,7 @@ JINN_PASSWORD=secret jinn run --config ./my-config.json
 | Config key | Env override | Default |
 |---|---|---|
 | network | JINN_NETWORK | `testnet` (Phase 1b default; flips to `mainnet` at Phase 2 launch) |
-| rpcUrl | BASE_RPC_URL / JINN_RPC_URL | network-appropriate public RPC |
+| rpcUrl | BASE_RPC_URL / JINN_RPC_URL | testnet: two-provider fallback chain `[publicnode, sepolia.base.org]`; mainnet: single `mainnet.base.org`. Accepts string OR array; env values split on commas. See "RPC fallback chain" below. |
 | claudeModel | JINN_CLAUDE_MODEL | claude-haiku-4-5-20251001 |
 | claudePath | JINN_CLAUDE_PATH | claude |
 | pollIntervalMs | JINN_POLL_INTERVAL_MS | 5000 |
@@ -325,6 +325,72 @@ JINN_PASSWORD=secret jinn run --config ./my-config.json
 
 `JINN_PASSWORD` is env-only (keystore encryption, never in config files). Alternatively, use `--password-fd <N>` to read from a file descriptor.
 
+### RPC fallback chain
+
+Out of the box on testnet, `jinn run` uses a two-provider fallback chain —
+no key required, zero config:
+
+```text
+slot 0  https://base-sepolia.publicnode.com   (no-auth, 50k-block getLogs cap)
+slot 1  https://sepolia.base.org              (free public Coinbase, 2k cap)
+```
+
+If the primary returns a network error or HTTP 429 / 5xx, viem
+transparently falls through to the next slot. When every slot fails the
+daemon throws `AllRpcsFailedError` (visible in the dashboard's NetworkSection
+and in stderr).
+
+#### Bring your own paid primary
+
+Operators with a paid Alchemy / Tenderly / QuickNode endpoint should
+**prepend** their URL — the public free chain stays as automatic backup:
+
+```json
+{
+  "rpcUrl": [
+    "https://my-alchemy-key.example",
+    "https://base-sepolia.publicnode.com",
+    "https://sepolia.base.org"
+  ]
+}
+```
+
+Or via env (comma-separated, same chain capped at 4 providers):
+
+```bash
+JINN_RPC_URL='https://my-alchemy-key.example,https://base-sepolia.publicnode.com,https://sepolia.base.org' jinn run
+```
+
+The chain is built with viem's `fallback({ rank: false })` so operator slot
+order is preserved (no latency-based reshuffling).
+
+#### Automatic failure modes
+
+| Failure                                     | What used to happen           | What now happens                                |
+|---------------------------------------------|--------------------------------|-------------------------------------------------|
+| Primary RPC returns HTTP 429 (rate limit)   | Daemon hot-loops on every call | viem falls through to slot 1; daemon continues  |
+| Primary RPC unreachable                     | Daemon errors silently         | viem falls through to slot 1; daemon continues  |
+| Primary's API key quota exhausted           | Daemon silently degraded       | viem falls through to slot 1; daemon continues  |
+| Every slot fails (rare)                     | Daemon errors silently         | `AllRpcsFailedError` thrown with masked host list — surfaces as a state message in the operator app |
+
+On boot the daemon emits one log line per slot followed by the canonical
+summary line:
+
+```text
+[rpc] L2 base-sepolia.publicnode.com ok latency=87ms
+[rpc] L2 sepolia.base.org ok latency=141ms
+[rpc] L2 transport: fallback chain (2 providers) — primary=base-sepolia.publicnode.com
+```
+
+The probe is **log-only** — secondary-slot 429s warn but never gate
+startup. Only chain-id mismatch against the head URL is fail-loud
+(`checkRpcNetwork`).
+
+Note this is the JSON-RPC transport layer. It is distinct from
+`discovery.fallbackToOnchain`, which is a separate flag at the read-API
+layer (Ponder indexer → direct `eth_getLogs` floor). The RPC fallback chain
+operates beneath both.
+
 ## Tokens
 
 | Role | Phase 1b (testnet, Base Sepolia) | Phase 2 (mainnet, Base) |

package/deployments/deployment-jinn-mvi-l1-sepolia-fast.json

@@ -20,8 +20,13 @@
   },
   "messenger": {
     "mode": "mock",
-    "address": "0x30E2f1A5691a999053A54e8DAf9496CD7ccF2DFD",
-    "owner": "0x34516dB851f03f8E14d0c1cD56200175e914cCf0"
+    "address": "0xd229A2C20333B747675090Ce38B8a1Fb2dafe6AC",
+    "owner": "0x15e78734481bD31F6e183dad05225505a45ACd07",
+    "previousAddress": "0x30E2f1A5691a999053A54e8DAf9496CD7ccF2DFD",
+    "rotatedAt": "2026-05-20T17:14:48.591Z",
+    "deployTxHash": "0x44c0fc6f92e1c8247304dceb195b8cfa1ae114a79e7a6ce1305e4305de45b9af",
+    "setMessengerTxHash": "0x13313c9cc29ae9d7ce1e13ca78472359fdb6213b5d1d936d73f89ca051867c00",
+    "rotationBlockNumber": 10886858
   },
   "handover": {
     "renounceAdmin": false
@@ -31,6 +36,20 @@
     "TimelockController": "0x5CE9c7E24f0509490a91a1cD285DDFF70A9C058D",
     "JinnGovernor": "0xbbc43A1B5AC699de09577dAE4A1007a3bA704D9E",
     "JinnDistributor": "0xaC9CD847660d05e77D82A3684aFC4EbFd94fBfe6",
-    "Messenger": "0x30E2f1A5691a999053A54e8DAf9496CD7ccF2DFD"
-  }
+    "Messenger": "0xd229A2C20333B747675090Ce38B8a1Fb2dafe6AC",
+    "MockMessenger": "0xd229A2C20333B747675090Ce38B8a1Fb2dafe6AC"
+  },
+  "rotations": [
+    {
+      "type": "mock-messenger",
+      "previousMessenger": "0x30E2f1A5691a999053A54e8DAf9496CD7ccF2DFD",
+      "newMessenger": "0xd229A2C20333B747675090Ce38B8a1Fb2dafe6AC",
+      "owner": "0x15e78734481bD31F6e183dad05225505a45ACd07",
+      "distributor": "0xaC9CD847660d05e77D82A3684aFC4EbFd94fBfe6",
+      "deployTxHash": "0x44c0fc6f92e1c8247304dceb195b8cfa1ae114a79e7a6ce1305e4305de45b9af",
+      "setMessengerTxHash": "0x13313c9cc29ae9d7ce1e13ca78472359fdb6213b5d1d936d73f89ca051867c00",
+      "blockNumber": 10886858,
+      "rotatedAt": "2026-05-20T17:14:48.591Z"
+    }
+  ]
 }

package/deployments/deployment-jinn-mvi-l1-sepolia.json

@@ -20,8 +20,13 @@
   },
   "messenger": {
     "mode": "mock",
-    "address": "0x30E2f1A5691a999053A54e8DAf9496CD7ccF2DFD",
-    "owner": "0x34516dB851f03f8E14d0c1cD56200175e914cCf0"
+    "address": "0xd229A2C20333B747675090Ce38B8a1Fb2dafe6AC",
+    "owner": "0x15e78734481bD31F6e183dad05225505a45ACd07",
+    "previousAddress": "0x30E2f1A5691a999053A54e8DAf9496CD7ccF2DFD",
+    "rotatedAt": "2026-05-20T17:14:48.591Z",
+    "deployTxHash": "0x44c0fc6f92e1c8247304dceb195b8cfa1ae114a79e7a6ce1305e4305de45b9af",
+    "setMessengerTxHash": "0x13313c9cc29ae9d7ce1e13ca78472359fdb6213b5d1d936d73f89ca051867c00",
+    "rotationBlockNumber": 10886858
   },
   "handover": {
     "renounceAdmin": false
@@ -31,6 +36,20 @@
     "TimelockController": "0x5CE9c7E24f0509490a91a1cD285DDFF70A9C058D",
     "JinnGovernor": "0xbbc43A1B5AC699de09577dAE4A1007a3bA704D9E",
     "JinnDistributor": "0xaC9CD847660d05e77D82A3684aFC4EbFd94fBfe6",
-    "Messenger": "0x30E2f1A5691a999053A54e8DAf9496CD7ccF2DFD"
-  }
+    "Messenger": "0xd229A2C20333B747675090Ce38B8a1Fb2dafe6AC",
+    "MockMessenger": "0xd229A2C20333B747675090Ce38B8a1Fb2dafe6AC"
+  },
+  "rotations": [
+    {
+      "type": "mock-messenger",
+      "previousMessenger": "0x30E2f1A5691a999053A54e8DAf9496CD7ccF2DFD",
+      "newMessenger": "0xd229A2C20333B747675090Ce38B8a1Fb2dafe6AC",
+      "owner": "0x15e78734481bD31F6e183dad05225505a45ACd07",
+      "distributor": "0xaC9CD847660d05e77D82A3684aFC4EbFd94fBfe6",
+      "deployTxHash": "0x44c0fc6f92e1c8247304dceb195b8cfa1ae114a79e7a6ce1305e4305de45b9af",
+      "setMessengerTxHash": "0x13313c9cc29ae9d7ce1e13ca78472359fdb6213b5d1d936d73f89ca051867c00",
+      "blockNumber": 10886858,
+      "rotatedAt": "2026-05-20T17:14:48.591Z"
+    }
+  ]
 }

package/deployments/deployment-jinn-mvi-l2-baseSepolia.json

@@ -2,11 +2,12 @@
   "network": "baseSepolia",
   "chainId": 84532,
   "deployer": "0x15e78734481bD31F6e183dad05225505a45ACd07",
-  "deployedAt": "2026-04-29T11:37:34.801Z",
+  "deployedAt": "2026-05-20T15:23:10.000Z",
+  "deployBlock": 41761151,
+  "txHash": "0x97524384c024a2bcfe11b3699bad53e8533bc25f43e64101ce1e0e4170491f9b",
   "contracts": {
-    "JinnClaimEmitter": "0xDB2f776D78d9aF12792C2C0f9875465eDaa859E7",
-    "RestorationActivityCheckerV2": "0x0e1B5f264F4FAdcFAA950fb00c58d9A39C040f70",
-    "JinnRouterV2": "0x13dae57a54A7DF862113BcA8Ee2fd3dEf6a6A94A",
+    "TaskClaimEmitter": "0xF60055534E377F4020eEA80356C0643E02f4f307",
+    "TaskActivityCheckerV3": "0x0e1B5f264F4FAdcFAA950fb00c58d9A39C040f70",
     "ServiceRegistry": "0x31D3202d8744B16A120117A053459DDFAE93c855"
   }
 }

package/dist/adapters/mech/adapter.d.ts

@@ -33,6 +33,17 @@ export declare class MechAdapter implements ExecutionAdapter {
     private deliveryBlockCursor;
     private pendingEvaluations;
     private observedTasks;
+    /**
+     * Read-through cache for `restorationAnnouncementForTaskId` — the restoration
+     * task body looked up *while building an evaluation opportunity*. Kept
+     * SEPARATE from `observedTasks` (the `watchForTasks` discovery dedup set) on
+     * purpose: writing the restoration body into `observedTasks` made the
+     * TaskCreated scan skip that taskId as a *restoration* opportunity just
+     * because the daemon had built an *evaluation* opportunity for someone
+     * else's attempt on it. That blocked the creator's own daemon from claiming
+     * its own attempt on a multi-attempt (`maxClaims > 1`) task it posted.
+     */
+    private restorationBodyCache;
     private requestKinds;
     private claimedRestorationTaskIds;
     private evaluationOpportunities;
@@ -51,6 +62,9 @@ export declare class MechAdapter implements ExecutionAdapter {
     private persistPendingEvaluationSolutions;
     private rememberPendingEvaluationSolution;
     private forgetPendingEvaluationSolution;
+    private clearPendingDeliveryRecoveryState;
+    private recoveryDeliveryExpirySeconds;
+    private shouldSkipExpiredRecoveryDelivery;
     postTask(state: Task): Promise<PostedTask>;
     private signTaskDocument;
     private contractPolicyForTask;
@@ -58,6 +72,17 @@ export declare class MechAdapter implements ExecutionAdapter {
     private restorationAnnouncementForTaskId;
     private restorationAnnouncementFromDigest;
     private discoverSubgraphRestorationTasks;
+    /**
+     * Look up the Deliver-event envelope CID for a pending evaluation solution.
+     *
+     * Returns `null` when the Deliver event is not present in the configured
+     * lookback window. This is a terminal signal for the caller — re-running the
+     * same lookup later is deterministically futile when `solution.blockNumber`
+     * is set (toBlock is fixed at the SolutionDeliveryClaimed block), and is
+     * monotonically less likely to find the event when toBlock follows chain head
+     * (the window slides forward, away from any older Deliver event). Callers
+     * should prune the pending solution on `null` rather than retry — see #553.
+     */
     private deliveryEnvelopeCidForSolution;
     private evaluationAnnouncementForSolution;
     private retryPendingEvaluationSolutions;
@@ -74,8 +99,20 @@ export declare class MechAdapter implements ExecutionAdapter {
     }>;
     submitSolutionDelivery(requestId: RequestId, solutionDigest: Hex): Promise<void>;
     submitVerdictDelivery(requestId: RequestId, verdictDigest: Hex, verdictCode: VerdictCode): Promise<void>;
-    private evidenceHashForDelivery;
+    private deliveryClaimForDelivery;
     private ensureDeliveryClaimed;
+    /**
+     * Paginate `getLogs` over `[deliveryBlockCursor+1, currentBlock]` chunked by
+     * `DEFAULT_ROUTER_LOG_CHUNK_BLOCKS` to honor RPC provider block-range limits
+     * (Tenderly base-sepolia caps at 100k; sepolia.base.org ~1k). Advances +
+     * persists `deliveryBlockCursor` per chunk so a mid-scan RPC failure on a
+     * later chunk does not strand the cursor at the pre-poll value (#552).
+     *
+     * Yields each chunk's decoded Deliver entries so the consumer can process
+     * them with the live "current block" context (needed for the recovery-
+     * delivery timestamp cache).
+     */
+    private scanDeliveryLogChunks;
     watchForDeliveries(): AsyncIterable<DeliveredResult>;
     stop(): Promise<void>;
 }