AI agent runaway action — quick-start

checklist

1. halt the agent process or revoke its API key immediately — stopping further actions takes priority over preserving the action log.
2. preserve the agent's action log before it is overwritten — every tool call, API request, and decision node in the runaway window.
3. preserve the original system prompt and any user instructions that were active at the time of the runaway.
4. list every external system the agent had access to: APIs, databases, file systems, email, messaging — these are all potential blast-radius surfaces.
5. identify what the agent actually did during the runaway: tool calls made, data read, data written, messages sent, payments initiated.
6. identify what the agent was trying to do based on its reasoning trace if available — intent and action may diverge.
7. pull logs from every external system the agent touched — the agent's action log is one view; the target system's log is the ground truth.
8. notify operators of any external services that received requests from the agent during the runaway window.
9. audit any irreversible actions: sent emails, posted messages, initiated financial transactions, deleted files.
10. begin the primary tool path below — AI agent action accountability tracer.

primary tools

1. 01ai agent tool call execution trace reconstructordrop agent run log · reconstruct tool-call sequence + state mutations · runs locally
2. 02ai agent prompt vs action divergence detectordrop agent run log · detect actions taken inconsistent with prompt · runs locally
3. 03ai agent autonomous action accountability tracerdrop agent run log · trace responsibility for each autonomous action · runs locally
4. 04ai agent credential handling auditdrop agent run log · audit credential usage + leakage risk · runs locally
5. 05mcp tool call graph reconstructordrop mcp client + server log set · reconstruct tool-call dependency graph · runs locally
6. 06ai agent persistence mechanism detectordrop agent + system state · detect persistence implanted by agent · runs locally
7. 07ai agent network exfiltration pattern detectordrop agent network log · detect data exfiltration via agent · runs locally

go deeper