1 Commits

Author SHA1 Message Date
453dab3271 fix(bridge): resolve real bridge key past OpenClaw redaction; sane provider timeout
Three install/bridge bugs that made every OpenClaw model call to the
bridge fail when driven by a non-bundled channel plugin (e.g. Fabric):

1. OpenClaw redacts secret-like keys before exposing pluginConfig to a
   plugin, so config.bridgeApiKey was the literal __OPENCLAW_REDACTED__
   sentinel. The bridge then validated Authorization against the
   sentinel while the model provider sent the real key -> permanent
   HTTP 401. Resolve the real shared secret from the raw on-disk config
   (same pattern resolveAgent already uses); if still missing/redacted,
   treat as no-auth on the loopback-only bridge instead of 401-locking.

2. install.mjs set the provider apiKey authoritatively but only
   setIfMissing the plugin bridgeApiKey, so a stale prior value desynced
   the pair. Make bridgeApiKey authoritative too (they must match).

3. The provider had no timeoutSeconds; a full bridged agent turn far
   exceeds OpenClaw's default model-fetch timeout, so OpenClaw aborted
   mid-turn and no reply was ever delivered. Default timeoutSeconds=600
   (preserves a user override).

Verified live: bridge now returns 200 for the real key and a valid
OpenAI SSE completion; the fetch-timeout abort is gone.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-16 09:57:32 +01:00
6 changed files with 62 additions and 135 deletions

View File

@@ -158,29 +158,10 @@ export async function* dispatchToClaude(
// detached:true puts claude in its own process group. Claude's Bash tool
// occasionally leaks shells/ssh that keep claude alive past end-of-turn; when
// that happens we SIGKILL the whole group rather than wait forever.
// Sanitize NODE_OPTIONS before spawning. Claude Code is a Node CLI; if
// the parent gateway runs with `NODE_OPTIONS=--inspect=...:9229`, every
// child Node process — including claude — tries to bind the same inspector
// port, fails (EADDRINUSE), and exits SILENTLY (no stdout, no stderr).
// Bridge then sees an empty stream and reports `claude did not return a
// session_id` with no useful diagnostic. Strip any --inspect* /
// --inspect-brk* / --debug* flag from NODE_OPTIONS; keep everything else
// (e.g. --max-old-space-size) in case operators depend on it.
const childEnv: NodeJS.ProcessEnv = { ...process.env };
if (childEnv.NODE_OPTIONS) {
const filtered = childEnv.NODE_OPTIONS
.split(/\s+/)
.filter((tok) => tok && !tok.startsWith("--inspect") && !tok.startsWith("--debug"))
.join(" ")
.trim();
if (filtered) childEnv.NODE_OPTIONS = filtered;
else delete childEnv.NODE_OPTIONS;
}
const child = spawn("claude", args, {
cwd: workspace,
stdio: ["ignore", "pipe", "pipe"],
env: childEnv,
env: { ...process.env },
detached: true,
});

View File

@@ -156,24 +156,10 @@ export async function* dispatchToGemini(
args.push("--resume", resumeSessionId);
}
// Sanitize NODE_OPTIONS before spawning — same reason as the claude
// adapter: gemini-cli is a Node binary; inheriting a parent
// `NODE_OPTIONS=--inspect=...:9229` makes every child silently EADDRINUSE.
const childEnv: NodeJS.ProcessEnv = { ...process.env };
if (childEnv.NODE_OPTIONS) {
const filtered = childEnv.NODE_OPTIONS
.split(/\s+/)
.filter((tok) => tok && !tok.startsWith("--inspect") && !tok.startsWith("--debug"))
.join(" ")
.trim();
if (filtered) childEnv.NODE_OPTIONS = filtered;
else delete childEnv.NODE_OPTIONS;
}
const child = spawn("gemini", args, {
cwd: workspace,
stdio: ["ignore", "pipe", "pipe"],
env: childEnv,
env: { ...process.env },
});
const stderrLines: string[] = [];

View File

@@ -18,6 +18,45 @@ const SERVER_KEY = "_contractorAgentBridgeServer";
/** Key for the live OpenClaw config accessor (getter fn) shared via globalThis. */
const OPENCLAW_CONFIG_KEY = "_contractorOpenClawConfig";
/** OpenClaw replaces secret-like config values with this before exposing
* pluginConfig to plugins. */
const OPENCLAW_REDACTION_SENTINEL = "__OPENCLAW_REDACTED__";
/**
* OpenClaw redacts any secret-like key (anything containing "apiKey") before
* handing `api.pluginConfig` to the plugin, so `pluginConfig.bridgeApiKey` is
* the redaction sentinel — never the real value. The bridge server validates
* inbound requests against this key, while the OpenClaw model provider sends
* the real `models.providers.contractor-agent.apiKey`; using the redacted
* value therefore guarantees a permanent HTTP 401.
*
* Resolve the real shared secret from the raw on-disk config instead (same
* pattern `resolveAgent` already uses). If it is still missing or redacted,
* return "" so the loopback-only bridge skips auth rather than hard-locking.
*/
function resolveBridgeApiKey(fallback: string): string {
try {
const configPath = path.join(
process.env.HOME ?? "/root",
".openclaw",
"openclaw.json",
);
const raw = JSON.parse(fs.readFileSync(configPath, "utf8")) as {
plugins?: {
entries?: Record<string, { config?: { bridgeApiKey?: string } }>;
};
};
const k = raw.plugins?.entries?.["contractor-agent"]?.config?.bridgeApiKey;
if (typeof k === "string" && k && k !== OPENCLAW_REDACTION_SENTINEL) {
return k;
}
} catch {
/* fall through to fallback handling */
}
if (fallback && fallback !== OPENCLAW_REDACTION_SENTINEL) return fallback;
return ""; // loopback-only bridge: skip auth instead of 401-locking
}
// ── Plugin entry ─────────────────────────────────────────────────────────────
export default definePluginEntry({
@@ -69,7 +108,7 @@ export default definePluginEntry({
api.on("gateway_start", () => {
const server = createBridgeServer({
port: config.bridgePort,
apiKey: config.bridgeApiKey,
apiKey: resolveBridgeApiKey(config.bridgeApiKey),
permissionMode: config.permissionMode,
resolveAgent,
logger: api.logger,

View File

@@ -61,51 +61,6 @@ function isRuntimeContextMessage(text: string): boolean {
return INBOUND_META_SENTINELS.includes(firstLine);
}
/**
* Strip *prefixed* metadata blocks from a single user message body.
*
* Background: OpenClaw's older / canonical convention is to emit metadata
* envelopes (chat_id, sender, reply target, …) as SEPARATE user-role
* messages folded into the openai-completions request right after the real
* one. `extractLatestUserMessage` skips those separate messages whole.
*
* The Fabric channel plugin (Fabric.OpenclawPlugin) does not split: it
* passes the metadata blocks and the actual user content as ONE merged
* user-message body, separated by blank lines. Result: a Fabric inbound
* turn has a single user message whose first line matches a sentinel, so
* `isRuntimeContextMessage` returned true and the whole turn was dropped
* with "no user message found" (HTTP 400). Symptom: every contractor agent
* subscribed to a Fabric channel silently fails to reply.
*
* Strip leading sentinel-prefixed blocks (sentinel header + optional code
* fence + blank-line separator) until we hit a block whose first line is
* NOT a sentinel — that's the real prompt. Returns "" if the entire body
* is metadata-only (still a no-op turn, same as before).
*/
function stripPrefixedMetadataBlocks(raw: string): string {
// Split on blank-line block boundaries. Within a metadata block the JSON
// code fence has only single newlines, so it stays in the same chunk as
// its sentinel header.
const blocks = raw.split(/\n\n+/);
const kept: string[] = [];
let stillStripping = true;
for (const block of blocks) {
if (stillStripping) {
const trimmed = block.trimStart();
const firstLine = trimmed.split("\n", 1)[0].trim();
if (
INBOUND_META_SENTINELS.includes(firstLine) ||
trimmed.startsWith(LEGACY_RUNTIME_CONTEXT_MARKER)
) {
continue; // drop this prefix block
}
stillStripping = false;
}
kept.push(block);
}
return kept.join("\n\n").trim();
}
/**
* Extract the latest user-authored message from the OpenClaw request.
*
@@ -130,18 +85,7 @@ export function extractLatestUserMessage(req: BridgeInboundRequest): string {
for (let i = userMessages.length - 1; i >= 0; i -= 1) {
const raw = messageText(userMessages[i]);
if (!raw) continue;
// First try the separate-message convention (Discord/Telegram path): a
// user message whose entire body is one metadata envelope — skip whole.
if (isRuntimeContextMessage(raw)) {
// ...but also try inline-prefixed-metadata recovery (Fabric path):
// some channels splice metadata + real content into one body. Walk
// past leading sentinel blocks; if any non-metadata block follows,
// that's the prompt. If nothing follows, this turn really is pure
// metadata and we keep skipping.
const stripped = stripPrefixedMetadataBlocks(raw);
if (stripped) return stripOpenClawTimestampPrefix(stripped);
continue;
}
if (isRuntimeContextMessage(raw)) continue;
return stripOpenClawTimestampPrefix(raw);
}
return "";

View File

@@ -596,47 +596,14 @@ export function createBridgeServer(config: BridgeServerConfig): http.Server {
return;
}
const execFn = (toolInstance as { execute: (id: string, args: unknown) => Promise<unknown> }).execute;
const toolResultRaw = await execFn(randomUUID(), toolArgs);
const execFn = (toolInstance as { execute: (id: string, args: unknown) => Promise<{ content?: Array<{ type: string; text?: string }> }> }).execute;
const toolResult = await execFn(randomUUID(), toolArgs);
// Normalize the result shape. OpenClaw plugins return one of:
// (a) AgentToolResult — { content: [{type:'text', text:'...'}, ...] }
// used by tools that wrap via asContent() (e.g. all
// Dialectic.OpenclawPlugin tools).
// (b) raw JSON-able object — { ok:true, ...domain fields }
// used by tools that return data directly (e.g. every
// Fabric.OpenclawPlugin tool: fabric-channel-list,
// fabric-guild-list, fabric-send-message, etc).
// (c) null / undefined — fire-and-forget (rare).
//
// Pre-2026-05-24 the bridge ONLY handled shape (a); shape (b)
// silently returned "(no result)" to Claude Code (bug:
// fabric-channel-list and friends were unusable from contractor
// agents even though the tool actually ran successfully).
let text = "";
if (toolResultRaw == null) {
text = "";
} else if (
typeof toolResultRaw === "object" &&
Array.isArray((toolResultRaw as { content?: unknown }).content)
) {
// Shape (a)
const content = (toolResultRaw as { content: Array<{ type: string; text?: string }> }).content;
text = content
// Extract text content from AgentToolResult
const text = (toolResult.content ?? [])
.filter((c) => c.type === "text" && c.text)
.map((c) => c.text as string)
.join("\n");
} else if (typeof toolResultRaw === "string") {
text = toolResultRaw;
} else {
// Shape (b) — serialize the whole object as JSON. Claude Code
// is happy to parse JSON tool results.
try {
text = JSON.stringify(toolResultRaw);
} catch {
text = String(toolResultRaw);
}
}
sendJson(res, 200, { result: text || "(no result)" });
} catch (err) {

View File

@@ -79,6 +79,11 @@ function install() {
baseUrl: `http://127.0.0.1:${BRIDGE_PORT}/v1`,
apiKey: BRIDGE_API_KEY,
api: "openai-completions",
// The bridge wraps a full Claude/Gemini agent turn (tool use, multi-step),
// which routinely takes far longer than OpenClaw's default model-fetch
// timeout. Without a generous timeout OpenClaw aborts the request mid-turn
// and no reply is ever delivered. Preserve a user override if set.
timeoutSeconds: existingProvider.timeoutSeconds ?? 600,
models: [
{
id: "contractor-claude-bridge",
@@ -120,11 +125,16 @@ function install() {
cfg.plugins.entries[PLUGIN_ID] = cfg.plugins.entries[PLUGIN_ID] ?? {};
cfg.plugins.entries[PLUGIN_ID].enabled = true;
// Set default config — setIfMissing so user values are preserved
// Set default config — setIfMissing so user values are preserved.
const pluginCfg = cfg.plugins.entries[PLUGIN_ID].config ?? {};
setIfMissing(pluginCfg, "bridgePort", BRIDGE_PORT);
setIfMissing(pluginCfg, "bridgeApiKey", BRIDGE_API_KEY);
setIfMissing(pluginCfg, "permissionMode", "bypassPermissions");
// bridgeApiKey is the shared secret between the bridge server (this plugin)
// and the model provider written above. The provider apiKey is set
// authoritatively (= BRIDGE_API_KEY); the bridge side MUST stay in lockstep
// or every request 401s. Set it authoritatively too — never setIfMissing
// (a stale prior value would desync the pair).
pluginCfg.bridgeApiKey = BRIDGE_API_KEY;
cfg.plugins.entries[PLUGIN_ID].config = pluginCfg;
writeConfig(cfg);