Debug Stuck Hawk/Inspect AI Evaluations

VerifiedSafe

Diagnostic guide to resolve stuck evaluations, timeouts, and progression errors. Includes hawk commands, error patterns, and API testing techniques.

Sby Skills Guide Bot
DevelopmentIntermediate
206/2/2026
Claude Code
#debugging#eval#hawk#inspect-ai

Recommended for

Our review

This skill helps diagnose and resolve Hawk/Inspect AI evaluations that are stuck, not progressing, or failing repeatedly.

Strengths

  • Provides a structured checklist to quickly identify the root cause of the stall.
  • Explains common error patterns and their resolutions.
  • Includes direct API testing commands and log inspection.
  • Offers recovery options like deleting and restarting the evaluation.

Limitations

  • Requires access to Hawk infrastructure and S3 logs.
  • Some errors (e.g., OpenAI retries) require manual API testing to see the actual error message.
  • Does not cover all possible stall scenarios (e.g., network issues or quota limits).
When to use it

Use this skill when a user reports that a Hawk/Inspect AI evaluation is stuck, not progressing, or has a high number of retries.

When not to use it

Do not use it for evaluations that are running normally or for configuration issues unrelated to stalling.

Security analysis

Safe
Quality score88/100

The skill provides debugging commands for internal tools; no destructive (beyond deleting an eval set as a recovery step), exfiltrating, or obfuscated actions are instructed. The curl examples use a locally sourced token for authentication to internal services, not external exfiltration.

No concerns found

Examples

Debug a stuck eval
My eval set 'abc-123' has been stuck for hours. It shows 'pending' and no samples are completing. Can you help debug it?
High retry count
The evaluation is showing 'HTTP retries: 45' and it's been stuck for a long time. What should I check?
500 errors in eval
I'm getting 500 internal server errors in my eval logs. How do I find the actual failing request?

name: debug-stuck-eval description: Debug stuck Hawk/Inspect AI evaluations. Use when user mentions "stuck eval", "eval not progressing", "eval hanging", "samples not completing", "eval set frozen", "runner stuck", "500 errors in eval", "retry loop", "eval timeout", or asks why an evaluation isn't finishing.

Quick Checklist

  1. Verify auth: hawk auth access-token > /dev/null || echo "Run 'hawk login' first"
  2. Get eval-set-id from user
  3. Check status: hawk status <eval-set-id> - JSON report with pod state, logs, metrics
  4. View logs: hawk logs <eval-set-id> or hawk logs -f for follow mode
  5. List samples: hawk list samples <eval-set-id> - see completion status
  6. Look for error patterns (see below)
  7. Test API directly if logs show retries without clear errors

Error Patterns

| Log Pattern | Meaning | Resolution | |-------------|---------|------------| | [uuid task/id/epoch model] Retrying request to /responses | OpenAI SDK retry with sample context | Test API directly with curl to see real error | | [uuid task/id/epoch model] -> model retry N ... [ErrorType code] | Inspect retry with error summary | Check error type; use curl for full details | | 500 - Internal server error | API issue | Download buffer, find failing request, test through middleman AND directly to provider | | 400 - invalid_request_error | Token/context limit exceeded | Check message count and model context window | | Pod UID mismatch | Sandbox pod was killed and restarted | No fix needed—sample errored out, Inspect will retry | | Empty output, pending: true | API returned malformed response | Restart eval (buffer resumes) | | OOMKilled in pod status | Memory exhaustion | Increase pod memory limits |

Key Techniques

  1. Retry messages have sample context - All retry messages include a [sample_uuid task/sample_id/epoch model] prefix. Inspect's own retries also include a compact error summary suffix like [RateLimitError 429 rate_limit_exceeded]. The OpenAI SDK's internal retry messages still don't show the actual error — use curl for full details.
  2. FAIL-OK patterns are fine - Alternating failures and successes mean the eval IS progressing. Only worry about consistent FAIL-FAIL-FAIL patterns.
  3. Use S3 for buffer access - Download .buffer/ from S3 rather than accessing the runner pod directly.
  4. Read .eval files with inspect_ai - Use from inspect_ai.log import read_eval_log instead of manually extracting zips.

Test API Directly

Middleman is the auth proxy. If middleman fails but direct provider calls work, it's a middleman issue.

TOKEN=$(hawk auth access-token)

# Test through middleman
curl --max-time 300 -X POST https://middleman.internal.metr.org/anthropic/v1/messages \
  -H "Authorization: Bearer $TOKEN" -H "Content-Type: application/json" \
  -d '{"model": "claude-sonnet-4-20250514", "max_tokens": 100, "messages": [{"role": "user", "content": "Say hello"}]}'

# Test OpenAI-compatible
curl --max-time 300 -X POST https://middleman.internal.metr.org/openai/v1/chat/completions \
  -H "Authorization: Bearer $TOKEN" -H "Content-Type: application/json" \
  -d '{"model": "gpt-4o", "messages": [{"role": "user", "content": "Say hello"}], "max_tokens": 100}'

Recovery

# Delete stuck eval and restart
hawk delete <eval-set-id>
hawk eval-set <config.yaml>

The sample buffer in S3 allows Inspect to resume from where it left off (unless you use --no-resume).

HTTP Retry Count

Task progress logs include "HTTP retries: X". High retry counts indicate API instability even while tasks complete.

Severity: Retry count × wait time = stuck duration. E.g., 45 retries × 1800s = 22+ hours stuck.

More Details

See docs/debugging-stuck-evals.md for:

  • Sample buffer SQL queries
  • Detailed API testing examples
  • Escalation checklist

References

Filing Issues

  • Middleman: https://github.com/metr-middleman/middleman-server/issues
  • Hawk: Linear issue on Evals Execution team
  • Inspect AI: https://github.com/UKGovernmentBEIS/inspect_ai/issues
Related skills