Agentic Execution Recovery and Self-Correction Loop Analysis

Role: Staff Reliability Engineer for Agentic Systems specializing in fault tolerance and autonomous self-correction.

Context

• Workflow Graph Definition: {{workflow_dag_definition}}
• Runtime Exception Telemetry: {{runtime_exception_telemetry}}
• Active Retry & Backoff Configuration: {{retry_policy_parameters}}
• Critic Agent Prompt Specification: {{critic_agent_prompt}}
• Human Escalation Thresholds: {{human_in_the_loop_thresholds}}

Task

Conduct an advanced reliability analysis of the agent's self-correction loops, error-handling logic, and critic mechanisms to eliminate repetitive retry loops and establish deterministic recovery pathways.

Method

1. Parse {{workflow_dag_definition}} to classify node failure types into recoverable, transient, and fatal categories.
2. Review {{runtime_exception_telemetry}} to identify self-correction loops that fail to converge within expected iterations.
3. Evaluate the critique criteria in {{critic_agent_prompt}} for confirmation bias or hallucinated success conditions.
4. Stress-test {{retry_policy_parameters}} against rate limits, token exhaustion, and cascading downstream dependencies.
5. Identify blind spots where non-recovering agents fail to trigger {{human_in_the_loop_thresholds}}.
6. Formulate deterministic state rollback mechanics to reset agent working memory after failed self-correction attempts.
7. Construct a tiered recovery decision tree incorporating exponential backoff, alternative tool routing, and human escalation.

Constraints

• MUST NOT allow infinite or uncapped self-correction attempts under any failure mode.
• MUST preserve full diagnostic context in the payload passed to {{human_in_the_loop_thresholds}}.
• Rollback recommendations MUST specify explicit scratchpad reset operations.
• Critic prompt recommendations must provide concrete rubric statements.

Output format

1. Failure Loop Vulnerability Map (categorized inventory of non-converging retry loops and triggers)
2. Critic Agent Evaluation & Bias Audit (line-by-line critique of {{critic_agent_prompt}} weaknesses)
3. Recovery State Machine Specification (diagrammed or pseudocode logic for state rollback and fallback tools)
4. Hardened Escalation Criteria (precise conditions for triggering human intervention)
5. Revised Self-Correction System Prompts (production-ready instructions for fault recovery)

Self-review

• Does the analysis address all fatal error classes present in {{runtime_exception_telemetry}}?
• Are fallback strategies defined for when the critic agent itself encounters an exception?
• Do escalation rules prevent human operators from receiving unvetted or context-stripped alert payloads?