fix(voice): create ElevenLabs tools before agent setup

[Shujakuinkuraudo]

Summary

• replace deprecated ElevenLabs prompt.tools usage with prompt.tool_ids
• auto-create or reuse the required client tools before creating/updating the Hapi voice agent
• keep the shared web helper aligned with the new ElevenLabs tool-based agent setup

Testing

• verified createOrUpdateHapiAgent() succeeds with an ElevenLabs key that has ConvAI write access
• verified the resulting ElevenLabs agent includes messageCodingAgent and processPermissionRequest
• built the Linux x64 single-executable binary successfully via bun run build:single-exe

[github-actions]

Findings

• [Major] Auto-discovery now hard-fails without ConvAI write scope — getOrCreateAgentId() now treats tool creation and agent patching as mandatory before it will return an existing agent ID, so installations that rely on ELEVENLABS_API_KEY only can start returning Failed to create ElevenLabs agent automatically even when the agent already exists and token minting would otherwise be possible. Evidence hub/src/web/routes/voice.ts:203, hub/src/web/routes/voice.ts:217.

Summary

• Review mode: initial
• 1 finding. The new auto-provisioning path turns token issuance into a write operation for existing agents, which is a regression for accounts without ConvAI write permission.
• Residual risk: Not found in repo/docs: automated coverage for the write-denied / existing-agent path.

Testing

• Not run (automation; untrusted PR code)

HAPI Bot

[github-actions]

Findings

• [Major] Existing auto-discovered agents now require tool write access — getOrCreateAgentId() calls ensureHapiToolIds() before it even looks for an existing Hapi Voice Assistant, and then returns null if patching that existing agent fails. That makes /api/voice/token fail for deployments with a usable existing agent but an ElevenLabs key that cannot create/list tools or patch agents, even though token minting only needs the agent id. Evidence hub/src/web/routes/voice.ts:205, hub/src/web/routes/voice.ts:217.
  Suggested fix:

  const existingAgentId = await findHapiAgent(apiKey)
  if (existingAgentId) {
      try {
          const toolIds = await ensureHapiToolIds(apiKey)
          await updateHapiAgent(apiKey, existingAgentId, toolIds)
      } catch (error) {
          console.warn('[Voice] Using existing agent without auto-update:', error)
      }
      agentIdCache.set(cacheKey, existingAgentId)
      return existingAgentId
  }
  
  const toolIds = await ensureHapiToolIds(apiKey)
  const agentId = await createHapiAgent(apiKey, toolIds)

Questions

• None.

Summary

• Review mode: follow-up after new commits
• 1 major issue remains: token issuance still hard-fails on the new tool provisioning/update path before reusing an existing auto-discovered agent.
• Residual risk: Not found in repo/docs: coverage for an existing-agent path where ElevenLabs tool creation or agent patching is denied.

Testing

• Not run (automation; untrusted PR code)

HAPI Bot

[github-actions]

[MAJOR] This still makes the auto-discovery path depend on ConvAI tool/agent write access. A deployment can already have a usable Hapi Voice Assistant, but this new ensureHapiToolIds() call runs before findHapiAgent() and returns null on any tool list/create failure; the later update failure also returns null. /api/voice/token then emits Failed to create ElevenLabs agent automatically instead of minting a token for the existing agent.

Suggested fix:

const existingAgentId = await findHapiAgent(apiKey)
if (existingAgentId) {
    try {
        const toolIds = await ensureHapiToolIds(apiKey)
        await updateHapiAgent(apiKey, existingAgentId, toolIds)
    } catch (error) {
        console.warn('[Voice] Using existing agent without auto-update:', error)
    }
    agentIdCache.set(cacheKey, existingAgentId)
    return existingAgentId
}

const toolIds = await ensureHapiToolIds(apiKey)
const agentId = await createHapiAgent(apiKey, toolIds)

[github-actions]

Findings

• [Major] Existing agents still cannot be reused if tool provisioning or patching fails — /api/voice/token only needs an agent id to mint the conversation token, but getOrCreateAgentId() now runs ensureHapiToolIds() before findHapiAgent() and returns null on any tool list/create failure; it also returns null if patching an already-found agent fails. A deployment with a usable existing Hapi Voice Assistant and a key that can read agents/token but cannot manage ConvAI tools/agent config will now fail token issuance. Evidence hub/src/web/routes/voice.ts:205.
  Suggested fix:

  const existingAgentId = await findHapiAgent(apiKey)
  if (existingAgentId) {
      try {
          const toolIds = await ensureHapiToolIds(apiKey)
          const updated = await updateHapiAgent(apiKey, existingAgentId, toolIds)
          if (!updated) {
              console.warn('[Voice] Using existing agent without auto-update')
          }
      } catch (error) {
          console.warn('[Voice] Using existing agent without auto-update:', error)
      }
      agentIdCache.set(cacheKey, existingAgentId)
      return existingAgentId
  }
  
  const toolIds = await ensureHapiToolIds(apiKey)
  const agentId = await createHapiAgent(apiKey, toolIds)

Questions

• None.

Summary

• Review mode: follow-up after new commits
• 1 major issue remains: the server token path still hard-fails on the new tool provisioning/update work before falling back to an existing auto-discovered agent.
• Residual risk/testing gap: Not found in repo/docs: coverage for reusing an existing ElevenLabs agent when tool list/create or agent patch permissions are denied.

Testing

• Not run: bun typecheck could not start because bun is not installed in this runner.

HAPI Bot

[github-actions]

[MAJOR] This still blocks the existing-agent fallback on the new tool-management path. /api/voice/token only needs an agent id, but this ensureHapiToolIds() call runs before findHapiAgent() and returns null on any tool list/create failure; the later updateHapiAgent() failure also returns null. Deployments with a usable existing Hapi Voice Assistant but an ElevenLabs key that cannot manage ConvAI tools/agent config will fail token issuance instead of using the existing agent.

Suggested fix:

const existingAgentId = await findHapiAgent(apiKey)
if (existingAgentId) {
    try {
        const toolIds = await ensureHapiToolIds(apiKey)
        const updated = await updateHapiAgent(apiKey, existingAgentId, toolIds)
        if (!updated) {
            console.warn('[Voice] Using existing agent without auto-update')
        }
    } catch (error) {
        console.warn('[Voice] Using existing agent without auto-update:', error)
    }
    agentIdCache.set(cacheKey, existingAgentId)
    return existingAgentId
}

const toolIds = await ensureHapiToolIds(apiKey)
const agentId = await createHapiAgent(apiKey, toolIds)