@inetafrica/open-claudia 3.0.8 → 3.0.9

package/CHANGELOG.md

@@ -1,5 +1,11 @@
 # Changelog
 
+## v3.0.9 — one fat record no longer kills the turn
+
+- **Decoder failures are warnings, not run failures.** The v3 JSONL stream decoder capped provider records at 1MB and surfaced any oversized or malformed line as a terminal `error` event — so a single base64-image tool result (routine in image workflows) killed an otherwise healthy live turn with `Claude Code run failed: Provider JSONL record exceeded 1048576 bytes` (rahil, v3.0.8, 2026-07-12). `LINE_TOO_LARGE` / `MALFORMED_JSON` now normalize to a new non-terminal `warning` event: the record is skipped, the stream continues, and the warning is logged. If the dropped line happened to be the terminal record, the existing `PROVIDER_MISSING_TERMINAL` net still fails the run cleanly.
+- **Line cap raised 1MB → 16MB.** Image-bearing records are legitimate; the cap now exists to bound memory, not to police normal payloads.
+- **Agent-space / OpenClaw:** no new deps, no new env, no schema change; Docker image builds identically (FROM the pre-baked `:base`). Coverage: warning-shape assertions in `test-provider-events.js`, default-cap pin in `test-provider-stream-decoder.js`.
+
 ## v3.0.6 — cache-bust burn is now a one-line query
 
 - **Usage records carry a `coldStart` flag.** The restart→cache-bust burn (each bot restart re-bills the whole accumulated window as cache *writes* instead of ~13×-cheaper cache *reads*) was invisible to every token-count analysis — same token counts, different price class — and took a forensic reconstruction of the 22-day usage history to quantify (~8% of all spend, ~$12/day at its worst). Now the first usage record a process writes for each session is tagged `coldStart: true`: a cold start mid-conversation is the restart fingerprint, so the burn is one jq filter away instead of a modelling exercise. Records already carried `cacheReadTokens`/`cacheCreationTokens`; this adds the missing "was the cache necessarily cold?" dimension.

package/core/providers/events.js

@@ -10,9 +10,12 @@ const NORMALIZED_EVENT_TYPES = Object.freeze([
   "usage",
   "result",
   "error",
+  "warning",
 ]);
 
-const DEFAULT_MAX_LINE_BYTES = 1024 * 1024;
+// 16MB: tool results carrying base64 images legitimately exceed 1MB — the old
+// cap killed rahil's image-workflow turn (v3.0.8) on a single oversized record.
+const DEFAULT_MAX_LINE_BYTES = 16 * 1024 * 1024;
 
 function decoderError(code, message, details = {}) {
   return { type: "decoder.error", code, message, ...details };
@@ -133,11 +136,13 @@ function providerErrorFlags(message) {
 
 function normalizeDecoderError(rawEvent) {
   if (!rawEvent || rawEvent.type !== "decoder.error") return null;
+  // Warning, not error: a single oversized or malformed record is skippable.
+  // If the dropped line was the terminal record, PROVIDER_MISSING_TERMINAL
+  // still fails the run cleanly.
   return {
-    type: "error",
+    type: "warning",
+    code: rawEvent.code || "DECODER_ERROR",
     message: rawEvent.message || "Provider stream decoding failed",
-    authError: false,
-    usageLimit: false,
   };
 }
 

package/core/runner.js

@@ -361,6 +361,8 @@ async function executeProviderInvocation(options = {}) {
     } else if (event.type === "error") {
       terminalSeen = true;
       rawErrors.push(event);
+    } else if (event.type === "warning") {
+      console.error("provider stream warning:", redactSensitive(String(event.message || event.code || "unknown")));
     }
     if (onEvent) {
       eventChain = eventChain.then(() => onEvent(event)).catch((error) => {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@inetafrica/open-claudia",
-  "version": "3.0.8",
+  "version": "3.0.9",
   "description": "An always-on, provider-agnostic coding-agent harness for Claude Code and OpenAI Codex via chat",
   "main": "bot.js",
   "bin": {

package/test-provider-events.js

@@ -24,6 +24,7 @@ assert.deepStrictEqual([...NORMALIZED_EVENT_TYPES].sort(), [
   "tool_end",
   "tool_start",
   "usage",
+  "warning",
 ].sort());
 
 const claudeState = createClaudeParserState();
@@ -154,10 +155,10 @@ assert.deepStrictEqual(codexAuth, [{
 
 const decoderError = { type: "decoder.error", code: "MALFORMED_JSON", message: "Malformed provider JSONL record" };
 assert.deepStrictEqual(normalizeClaudeEvent(decoderError, createClaudeParserState()), [{
-  type: "error", message: "Malformed provider JSONL record", authError: false, usageLimit: false,
-}]);
+  type: "warning", code: "MALFORMED_JSON", message: "Malformed provider JSONL record",
+}], "decoder failures must be non-fatal warnings, not run-killing errors");
 assert.deepStrictEqual(normalizeCodexEvent(decoderError, createCodexParserState()), [{
-  type: "error", message: "Malformed provider JSONL record", authError: false, usageLimit: false,
+  type: "warning", code: "MALFORMED_JSON", message: "Malformed provider JSONL record",
 }]);
 
 assert.deepStrictEqual(normalizeClaudeEvent({ type: "future.claude.event" }, createClaudeParserState()), []);

package/test-provider-language.js

@@ -32,11 +32,11 @@ assert.doesNotMatch(setup, /Your Claude Code bot is connected|Description=Claude
 
 const pkg = JSON.parse(read("package.json"));
 assert.match(pkg.description, /provider-agnostic coding-agent harness/i);
-// 3.0.8 approved by Sumeet 2026-07-12 (fast-path release after the v3.0.7
-// docker image was cancelled mid-build: same code as 3.0.7 — zombie-poll-loop
-// fix + turn-observer restore — plus the pre-baked :base CI image and codex
-// gpt-5.6-sol-pro / max / ultra support).
-assert.strictEqual(pkg.version, "3.0.8", "release version must remain unchanged without explicit approval");
+// 3.0.9 approved by Sumeet 2026-07-12 ("ok then please fix that"): decoder
+// LINE_TOO_LARGE/MALFORMED_JSON downgraded to non-fatal warnings and the JSONL
+// line cap raised 1MB→16MB after a base64-image tool result killed rahil's
+// live turn on v3.0.8.
+assert.strictEqual(pkg.version, "3.0.9", "release version must remain unchanged without explicit approval");
 for (const keyword of ["claude", "codex", "coding-agent", "provider-agnostic"]) {
   assert.ok(pkg.keywords.includes(keyword), `package keyword missing: ${keyword}`);
 }

package/test-provider-run-lifecycle.js

@@ -232,8 +232,29 @@ async function main() {
     proc.stdout.end();
     proc.emit("close", 0, null);
     const malformed = await malformedRun;
-    assert.strictEqual(malformed.ok, false);
-    assert.match(malformed.error.message, /Malformed provider JSONL record/);
+    // A single bad record is skipped (warning), not fatal: a >1MB base64-image
+    // tool result killed a healthy live turn on v3.0.8.
+    assert.strictEqual(malformed.ok, true, "one undecodable record must not fail a run that reached its terminal event");
+    assert.strictEqual(malformed.text, "fixture");
+  }
+
+  {
+    // But if the stream NEVER recovers a terminal event, the run still fails.
+    const proc = fakeChild();
+    const lostTerminalRun = executeProviderInvocation({
+      runContext: context(),
+      provider: provider(),
+      invocation: invocation(),
+      cwd: "/fixture/project",
+      spawnProcess: () => proc,
+      cleanup: async () => {},
+    });
+    proc.stdout.write("{not-json}\n");
+    proc.stdout.end();
+    proc.emit("close", 0, null);
+    const lostTerminal = await lostTerminalRun;
+    assert.strictEqual(lostTerminal.ok, false);
+    assert.match(lostTerminal.error.message, /terminal/i);
   }
 
   {

package/test-provider-stream-decoder.js

@@ -3,7 +3,13 @@
 "use strict";
 
 const assert = require("assert");
-const { createJsonlDecoder } = require("./core/providers/events");
+const { createJsonlDecoder, DEFAULT_MAX_LINE_BYTES } = require("./core/providers/events");
+
+assert.strictEqual(
+  DEFAULT_MAX_LINE_BYTES,
+  16 * 1024 * 1024,
+  "default line cap must fit base64-image tool results (1MB cap killed live image turns in v3.0.8)",
+);
 
 const unicodeEvent = { type: "fixture", text: "café 👋" };
 const unicodeLine = Buffer.from(`${JSON.stringify(unicodeEvent)}\n`, "utf8");