Review Agent health and logs

1. Select the Organization that owns the Agent.
2. Open the Agent from Home.
3. Review the condition shown under its name.
4. Look for an error banner beneath the Agent header.
5. Select Logs.

Users with activity.read can also see:

• Conversations
• Tool calls
• Logs
• Work

The About tab remains available independently of activity access.

If the diagnostic tabs are missing, confirm that your effective Agent Access Role includes activity.read.

The Agent condition refreshes automatically while the Agent is Running or in Error. It is derived from two separate inputs: the lifecycle state Agent Barn has persisted, and the health the Runtime currently reports.

Idle

Idle means the Agent is Stopped.

This is not a failure. Agent Barn does not query Runtime health, because an active Runtime is not expected.

Review the most recent Log snapshot if you need to understand the previous session. Select Start only when the Agent should resume work.

Initializing

Initializing means the Agent is marked Running, but is not ready yet.

This condition is expected briefly while Agent Barn:

• Creates Runtime resources
• Starts the Agent container
• Loads the selected Template and Skills
• Configures integrations
• Connects to the communication platform
• Begins reporting health

Wait for the condition to change to Working. If it remains Initializing, review Logs for:

• Image download problems
• Missing configuration
• Invalid platform credentials
• Missing channels
• Platform connection timeouts
• Repeated Runtime startup failures

A Runtime that has not connected successfully yet can continue to report a starting state, rather than an established connection failure.

Working

Working means the Runtime health endpoint and configured platform connection report healthy.

This confirms Runtime availability, but it does not guarantee that:

• Every platform message satisfies the Agent’s access policy
• Every Tool Call will succeed
• Every integration credential is valid
• The Agent’s Template will produce the intended answer

When investigating a functional problem on an otherwise Working Agent, continue with Conversations, Tool Calls, and Logs.

Disconnected

Disconnected means the Agent is persisted as Running, but Agent Barn cannot currently confirm healthy Runtime operation.

Possible causes include:

• The Runtime health endpoint is temporarily unreachable
• The platform connection failed after previously working
• Platform credentials were revoked or expired
• Network connectivity to the platform is unavailable
• The Runtime is still recovering
• The log or health request encountered a temporary infrastructure problem

Read the error reason if one is available, then inspect the most recent live Logs.

Hermes retries retryable platform connection failures. If repeated failures open its platform circuit breaker, Kubernetes may restart the Agent workload after approximately five minutes. Continue investigating the underlying credential, configuration, or platform problem, rather than relying only on the restart.

Needs attention

Needs attention means either:

• Agent Barn persisted the Agent in Error after a failed lifecycle operation, or
• The Running workload is in a crashed condition.

The error banner may include a friendly reason such as:

• Agent is crashing repeatedly on startup
• Agent was killed because it ran out of memory
• Failed to pull the Agent image
• A Secret or environment variable may be missing
• The Agent process exited unexpectedly
• A Kubernetes resource or permission problem prevented startup

Open Logs before changing the Agent. The last lines often show the more specific Runtime or platform failure.

When an Agent is Running, the Logs tab displays Live logs.

Agent Barn initially loads recent Runtime output, and then opens a live server-sent event stream for new lines. While the stream is connected, the header displays Streaming.

[startup output]
[platform connection output]
[runtime operation output]
[integration or error output]

This is a visual example of the interface layout. The lines above are placeholders, not product output.

The displayed text comes directly from the Agent Runtime. Exact messages differ between Hermes and OpenClaw, and between communication platforms.

Follow new output

The log viewer stays near the newest line while you remain at the bottom.

If you scroll upward, automatic following pauses so that new output does not move the section you are reading. Select Jump to latest to return to the newest line.

Search for the first relevant failure

When diagnosing startup, begin near the earliest line from the current session and look for terms such as:

error
failed
invalid
unauthorized
forbidden
timeout
connection
token
credential
CrashLoopBackOff
OOMKilled
ImagePullBackOff

Read the surrounding lines. The first error is often more useful than later failures caused by the same problem.

Reproduce one controlled request

If the Agent is Working but a particular operation fails:

1. Keep Logs open.
2. Send one controlled platform request.
3. Note the request time.
4. Watch for the related Runtime output.
5. Compare the result with Conversations and Tool calls.

Avoid sending many repeated tests at once. They make timestamps and causality harder to interpret.

Live-log display limit

The web log viewer retains up to 10,000 lines in its active browser buffer. When more lines arrive, it keeps the newest lines.

This browser limit is separate from stored Log snapshot retention.

When an Agent is not Running, Logs shows output from the latest retained Runtime session.

Agent Barn attempts to capture a Log snapshot before it stops an Agent, including a stop performed as part of Apply & Restart.

Scroll toward the beginning of the log viewer to load older retained sessions. Sessions are separated with a marker:

• Latest session

  Live output while Running, or the newest retained snapshot when the Agent is not Running

• === Session ended 8/29/2026, 10:42:18 AM ===
• Previous session

  The snapshot captured when that Runtime session ended

• === Session ended 8/28/2026, 4:07:52 PM ===
• Older sessions

  Loaded as you continue scrolling, until the oldest retained snapshot

This lets you compare the current or latest session with earlier starts. Useful comparisons include:

• The last Working session versus the first failing session
• Configuration before and after Apply & Restart
• Platform connection output before and after credential replacement
• Repeated startup failures across several attempts

An empty Log view can mean:

• The Agent has never produced Runtime output
• The Agent has never completed a session that produced a snapshot
• The latest session produced no output
• Snapshot capture failed
• Older snapshots exceeded the retention window

It does not prove that no Agent activity occurred.

Logs are most useful when combined with the other Agent activity tabs.

Conversations

Use Conversations to confirm:

• The platform message reached the Agent
• The message appeared in the expected channel or direct message
• The correct thread was used
• An outbound response was recorded

If a message does not appear, investigate platform delivery, channel access, mention gating, and bot membership before assuming the Agent’s Template failed.

Tool calls

Use Tool calls to inspect external executions with these states:

• Pending
• Success
• Error

A Working Agent can still have an individual Tool Call in Error. Open Logs around the Tool Call’s timestamp to find provider, credential, validation, or network details.

Logs

Use Logs for Runtime-level evidence such as:

• Startup sequencing
• Platform connection attempts
• Model or proxy errors
• Skill loading
• Integration failures
• Runtime exceptions
• Shutdown and restart behavior

Logs are operational output. They are not the canonical record of Conversations or Tool Calls.

Choose the recovery action based on the condition and the evidence you collected.

Correct configuration problems

Open Configuration when Logs indicate problems with:

• Model selection
• Template or required Skills
• Channel access
• Platform routing
• Runtime-specific settings

A Running Agent uses Apply & Restart. A Stopped Agent, or an Agent in Error, uses Apply.

Correct credential problems

Open Configuration → Keys & integrations when Logs report invalid, revoked, expired, or missing platform or provider credentials.

Credential changes require agent.secret.manage.

Because Secret values are write-only, Agent Barn shows credential metadata rather than returning stored plaintext. Enter the replacement value, and validate the integration where that option is available.

Restart a Running or crashed Agent

Agent Barn does not provide a separate one-click Restart action.

1. Select Pause.
2. Wait for the Agent to become Idle.
3. Select Start.
4. Wait for Initializing to become Working.

Starting regenerates Runtime configuration, platform settings, integration material, and ingest identity from the Agent’s currently selected configuration.

Verify recovery

After recovery:

• Confirm that the condition changes to Working
• Confirm that Live logs shows Streaming
• Check that the original startup or connection error no longer repeats
• Send one controlled platform request
• Confirm that the inbound message appears in Conversations
• Confirm that the Agent produces the expected response
• Check any expected Tool Call
• Continue observing briefly for recurring failures

Do not treat a successful Start response alone as recovery. The Agent should reach Working and complete an end-to-end platform interaction.