fix(search): return actionable live read errors and drop unreadable Confluence matches

[waleedlatif1]

Summary

• read_document turned every provider-side read failure into the generic "Knowledge operation failed", because NativeSearchError escaped readLiveDocument without being classified. Reads now return errors the model can act on:
  • a 404 or 410 becomes not_found, with a hint to search again or read a different result;
  • a revoked grant becomes unauthorized, naming the provider to reconnect;
  • rate limits, 5xx responses, provider timeouts, MCP request timeouts and the 15 s read deadline become a retryable LiveReadError.
• The Assistant read_document tool now reports retryable (and retryAfterSeconds when set), the same way search_workspace does. The Search MCP read_document returns the classified message instead of the generic text.
• Confluence search labeled every non-blogpost CQL hit a page. Native CQL that matched attachments, comments, whiteboards, folders or databases therefore produced references whose v2 page read returns 404. Those kinds are now dropped, and the result message says so.

Test plan

• application.test.ts: covers 404, reconnect, rate limit, 503 and MCP timeout. All five fail with the fix reverted.
• atlassian.test.ts: covers native CQL matches that a page read can't open. It fails with the fix reverted.
• Focused suites pass locally: application, atlassian, policy, workspace-search, mcp server.
• CI

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated
docs	Skipped		Oct 10, 2026 10:19am UTC

[cubic-dev-ai]

2 issues found across 7 files

Confidence score: 3/5

• In application.ts, native requests can time out before the read deadline and arrive as raw transport errors, so they may miss the retry path. Classify recognized transport timeouts as retryable.
• In atlassian.ts, excluded CQL rows can leave partial false for a single site, so callers report ok despite incomplete results. Include excluded when computing partial.

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="apps/sim/lib/sim-search/live/application.ts">

<violation number="1" location="apps/sim/lib/sim-search/live/application.ts:795">
P2: Native provider request timeouts can arrive here as raw transport errors: `createNativeClient` uses a 10-second request timeout, before the 15-second read deadline. Classify recognized transport timeout errors as retryable `LiveReadError`s instead of letting them escape as generic failures.</violation>
</file>

<file name="apps/sim/lib/sim-search/live/atlassian.ts">

<violation number="1" location="apps/sim/lib/sim-search/live/atlassian.ts:212">
P2: This drops matching CQL rows but leaves `partial` false for a single site, so callers report status `ok` despite known omitted matches. Include `excluded` in `partial` so callers treat coverage as incomplete.</violation>
</file>

<sub>Reply with feedback, questions, or to request a fix.

Turn on auto-fix | Re-trigger cubic</sub>

[cubic-dev-ai]

P2: Native provider request timeouts can arrive here as raw transport errors: createNativeClient uses a 10-second request timeout, before the 15-second read deadline. Classify recognized transport timeout errors as retryable LiveReadErrors instead of letting them escape as generic failures.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At apps/sim/lib/sim-search/live/application.ts, line 795:

<comment>Native provider request timeouts can arrive here as raw transport errors: `createNativeClient` uses a 10-second request timeout, before the 15-second read deadline. Classify recognized transport timeout errors as retryable `LiveReadError`s instead of letting them escape as generic failures.</comment>

<file context>
@@ -742,6 +748,53 @@ export type LiveReadInput = ResourceOwner & {
+      `${name} took too long to return this document. Try again, or read a different result.`,
+      true
+    )
+  return error
+}
+
</file context>

[cubic-dev-ai]

P2: This drops matching CQL rows but leaves partial false for a single site, so callers report status ok despite known omitted matches. Include excluded in partial so callers treat coverage as incomplete.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At apps/sim/lib/sim-search/live/atlassian.ts, line 212:

<comment>This drops matching CQL rows but leaves `partial` false for a single site, so callers report status `ok` despite known omitted matches. Include `excluded` in `partial` so callers treat coverage as incomplete.</comment>

<file context>
@@ -183,22 +195,31 @@ export async function searchAtlassian(
     })
   )
   const documents = interleaveByRank(pages.map((result) => result.documents))
+  const excluded = pages.some((result) => result.excluded)
   return {
     documents,
</file context>

[greptile-apps]

Confidence Score: 4/5

[Medium impact] Changes how search handles read failures and filters Confluence results.

Fix native transport timeout classification before merging so temporary read failures receive the promised retry guidance.

Findings

1.  Native timeouts still look permanent ▶

Summary

This PR gives live document reads clearer errors and removes Confluence matches that the reader cannot open.

• Live document reads tell callers what happened and when to retry.
• Confluence search leaves out matches its read path cannot open.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Live document read] --> B[Resolve account and check access]
  B --> C[Read from provider]
  C --> D{Read outcome}
  D -->|Success| E[Check current access and return content]
  D -->|Known provider error| F[Classify missing, reconnect, or retry]
  D -->|Read deadline or MCP timeout| G[Retryable LiveReadError]
  D -->|Native transport timeout at 10 seconds| H[Unchanged error]
  F --> I[Assistant or MCP response]
  G --> I
  H --> J[Generic error; Assistant says not retryable]

Loading

_{Reviews (1) · Last reviewed commit: "fix(search): return actionable live read..." · Reviewed by Greptile}

[greptile-apps]

Native timeouts still look permanent

createNativeClient sets a 10-second request timeout, which can throw a plain Error with code: 'ETIMEDOUT'. The 15-second read deadline has not fired yet, so liveReadFailure returns this error unchanged. The Assistant then reports retryable: false, and MCP returns generic failure text instead of retry guidance.

Recognize the transport's timeout errors, including the dispatcher timeout path, and return a retryable LiveReadError.