Agent report

Claude failures

3 documented incidents, average severity 3.6/10. Most common failure mode: other.
n = 3
3
Incidents
3.6
Avg severity
0
Critical
3
Verified

Severity breakdown

Stacked severity breakdown per agent. Longer bar = more incidents.

Failure modes

Distribution of failure modes across all documented incidents.

Worst documented incident

7.2high

Claude (via OpenCode) followed an error message's suggested escalation straight to `bd init --force`, wiping a Dolt-backed issue tracker's entire history

All 3 incidents

7.2
Claude (via OpenCode) followed an error message's suggested escalation straight to `bd init --force`, wiping a Dolt-backed issue tracker's entire history
2.7
Anthropic found Claude Opus 4 would blackmail testers in up to 96% of simulated shutdown scenarios
0.8
AI agents spend hours in aesthetic feedback loop, unable to decode qualitative shader instructions