Budibase/budibase · error

Invalid agent request link response

Error message

Invalid agent request link response

What it means

analyzeAgentRequestLink asks an LLM to decide whether a prompt belongs to an existing agent request thread, expecting a strict JSON reply. extractJson parses and validates that reply; if it cannot produce a valid decision object (missing/unknown decision value, or existing_thread without requestId/entryAction), the function throws Error("Invalid agent request link response").

Source

Thrown at packages/server/src/sdk/workspace/ai/agentRequests/helpers.ts:133

      prompt: JSON.stringify({
        currentSessionId: sessionId,
        latestPrompt: normalizedPrompt,
        recentChatContext,
        candidateRequests: candidateRequests.map(buildCandidateSummary),
      }),
    })
  } catch (error) {
    console.error("Failed to analyze agent request link", {
      agentId,
      sessionId,
      error,
    })
    throw error
  }

  const decision = extractJson(result.text || "")
  if (!decision) {
    throw new Error("Invalid agent request link response")
  }

  if (
    decision.decision === "existing_thread" &&
    !candidateRequests.some(request => request._id === decision.requestId)
  ) {
    throw new Error(
      `Invalid agent request link response: unknown requestId "${decision.requestId}"`
    )
  }

  return {
    decision: decision.decision,
    requestId: decision.requestId,
    entryAction: decision.entryAction,
  }
}

View on GitHub (pinned to a81a902e9a)

Solutions

1. Use a model/aiconfig that reliably follows JSON-only instructions; raise temperature constraints if configurable
2. Catch the error and fall back to decision "new_thread" so the agent flow continues
3. Retry the generateText call once on failure
4. Strengthen output handling: request structured output / response_format JSON if the provider supports it

Example fix

// before
const analysis = await analyzeAgentRequestLink({...}) // throws on bad LLM reply
// after
let analysis
try {
  analysis = await analyzeAgentRequestLink({...})
} catch {
  analysis = { decision: "new_thread" }
}

let analysis
try {
  analysis = await analyzeAgentRequestLink({ latestPrompt, candidateRequests, ... })
} catch (err) {
  if (err instanceof Error && err.message === "Invalid agent request link response") {
    analysis = { decision: "new_thread" }
  } else {
    throw err
  }
}

When it happens

Trigger: The LLM returns prose instead of JSON, malformed JSON, a decision value other than existing_thread/new_thread, or existing_thread without a valid requestId and entryAction.

Common situations: A weak or non-instruct-tuning-capable model ignoring the JSON-only instruction; the model wraps JSON in markdown fences or adds commentary (only the first {...} block is extracted); truncated responses cutting off the JSON.

Related errors

• Invalid agent request link response: unknown requestId "${de
• Invalid agent request title response
• Invalid tool call summary response
• Error generating tables
• LLM not available

AI-assisted analysis of Budibase/budibase@a81a902e9a (2026-08-29). Data as JSON: /api/errors/4c8d3bfe91ffba9a. Report an issue: GitHub.