Reusable AI capability
Recover Durable Agent Execution State
Reconcile checkpoints, durable state, side effects, approvals, and idempotency after interruption to select and gate a safe resume, replay, compensate, or abort path.
This Skill is a reusable set of instructions based on the linked Amo.ng Prompt or Workflow. It tells you what to provide, how to apply the method and how to know when the task is complete. It does not run automatically.
# Recover Durable Agent Execution State Skill ID: AMO-S-000020 Skill URL: https://amo.ng/skills/recover-durable-agent-execution-state Purpose: Give agent platform and service owners a reusable recovery-state method for interrupted long-running executions across workflows, tools, and external systems without claiming that recovery actions occurred. Required inputs: - Interrupted run identity, intended outcome, workflow and configuration versions - Durable state, checkpoints, messages, tool records, approvals, and external side effects - Idempotency keys, transaction or compensation semantics, retry history, and known-good state - Incident timeline, containment status, recovery options, constraints, and accountable owners How to use: When to use: - An agent run stopped after partial execution and its safe continuation state is uncertain. - Duplicate, omitted, or conflicting side effects must be reconciled before restart. When not to use: - Stateless request retries with no durable state or external side effect. - Authorizing production replay, compensation, or rollback. - Treating the latest timestamp as proof of the latest valid checkpoint. Reusable method: 1. Freeze the run, workflow and configuration versions, intended outcome, interruption window, and authority boundary. 2. Build a checkpoint and side-effect ledger from state, messages, tools, approvals, and external-system evidence. 3. Classify each action as committed, pending, failed, duplicated, compensated, ambiguous, or not evidenced. 4. Establish the latest trustworthy recovery boundary and identify stale, corrupt, or conflicting state. 5. Compare Resume, Replay, Compensate, Abort, and Continue investigation paths for idempotency, dependency, data, customer, and reversibility risk. 6. Define prerequisites, smallest safe action, owner authorization, stop condition, verification, rollback, and monitoring for the chosen path. Expected output: A recovery-state record with checkpoint lineage, side-effect ledger, ambiguity and evidence gaps, path comparison, selected disposition, prerequisites, compensation plan, verification, and owner gates. Boundaries: Do not invent state, side effects, commands, approvals, or recovery. Service and incident owners choose the operating path; data, financial, security, and external-system owners authorize actions in their systems; the release owner controls restart. Source: AMO-P-000297. Applicable Workflow: Recover and Requalify an Interrupted AI Agent Runtime. Based on Prompt: Long-Running Agent State Recovery Decision Source ID: AMO-P-000297 https://amo.ng/prompts/long-running-agent-state-recovery-decision Completion criteria: Complete when every material side effect and checkpoint is classified; the latest trustworthy boundary and uncertainty are explicit; one recovery disposition has prerequisites, owner, stop and rollback conditions; and observed recovery is separated from the proposed plan. Use this Amo.ng Skill with your preferred AI tool. Supply the required inputs and follow the usage instructions. # Recover Durable Agent Execution State Skill ID: AMO-S-000020 Skill URL: https://amo.ng/skills/recover-durable-agent-execution-state Purpose: Give agent platform and service owners a reusable recovery-state method for interrupted long-running executions across workflows, tools, and external systems without claiming that recovery actions occurred. Required inputs: - Interrupted run identity, intended outcome, workflow and configuration versions - Durable state, checkpoints, messages, tool records, approvals, and external side effects - Idempotency keys, transaction or compensation semantics, retry history, and known-good state - Incident timeline, containment status, recovery options, constraints, and accountable owners How to use: When to use: - An agent run stopped after partial execution and its safe continuation state is uncertain. - Duplicate, omitted, or conflicting side effects must be reconciled before restart. When not to use: - Stateless request retries with no durable state or external side effect. - Authorizing production replay, compensation, or rollback. - Treating the latest timestamp as proof of the latest valid checkpoint. Reusable method: 1. Freeze the run, workflow and configuration versions, intended outcome, interruption window, and authority boundary. 2. Build a checkpoint and side-effect ledger from state, messages, tools, approvals, and external-system evidence. 3. Classify each action as committed, pending, failed, duplicated, compensated, ambiguous, or not evidenced. 4. Establish the latest trustworthy recovery boundary and identify stale, corrupt, or conflicting state. 5. Compare Resume, Replay, Compensate, Abort, and Continue investigation paths for idempotency, dependency, data, customer, and reversibility risk. 6. Define prerequisites, smallest safe action, owner authorization, stop condition, verification, rollback, and monitoring for the chosen path. Expected output: A recovery-state record with checkpoint lineage, side-effect ledger, ambiguity and evidence gaps, path comparison, selected disposition, prerequisites, compensation plan, verification, and owner gates. Boundaries: Do not invent state, side effects, commands, approvals, or recovery. Service and incident owners choose the operating path; data, financial, security, and external-system owners authorize actions in their systems; the release owner controls restart. Source: AMO-P-000297. Applicable Workflow: Recover and Requalify an Interrupted AI Agent Runtime. Based on Prompt: Long-Running Agent State Recovery Decision Source ID: AMO-P-000297 https://amo.ng/prompts/long-running-agent-state-recovery-decision Completion criteria: Complete when every material side effect and checkpoint is classified; the latest trustworthy boundary and uncertainty are explicit; one recovery disposition has prerequisites, owner, stop and rollback conditions; and observed recovery is separated from the proposed plan.Copy Skill copies the full Skill instructions. Use with AI copies those instructions with a short guide for your chosen AI tool. Both actions only copy text—they do not send anything to an AI provider or run the Skill.
Purpose
Give agent platform and service owners a reusable recovery-state method for interrupted long-running executions across workflows, tools, and external systems without claiming that recovery actions occurred.
Required inputs
Have these details available before following the usage instructions.
- Interrupted run identity, intended outcome, workflow and configuration versions
- Durable state, checkpoints, messages, tool records, approvals, and external side effects
- Idempotency keys, transaction or compensation semantics, retry history, and known-good state
- Incident timeline, containment status, recovery options, constraints, and accountable owners
How to use this Skill
When to use:
- An agent run stopped after partial execution and its safe continuation state is uncertain.
- Duplicate, omitted, or conflicting side effects must be reconciled before restart.
When not to use:
- Stateless request retries with no durable state or external side effect.
- Authorizing production replay, compensation, or rollback.
- Treating the latest timestamp as proof of the latest valid checkpoint.
Reusable method:
1. Freeze the run, workflow and configuration versions, intended outcome, interruption window, and authority boundary.
2. Build a checkpoint and side-effect ledger from state, messages, tools, approvals, and external-system evidence.
3. Classify each action as committed, pending, failed, duplicated, compensated, ambiguous, or not evidenced.
4. Establish the latest trustworthy recovery boundary and identify stale, corrupt, or conflicting state.
5. Compare Resume, Replay, Compensate, Abort, and Continue investigation paths for idempotency, dependency, data, customer, and reversibility risk.
6. Define prerequisites, smallest safe action, owner authorization, stop condition, verification, rollback, and monitoring for the chosen path.
Expected output:
A recovery-state record with checkpoint lineage, side-effect ledger, ambiguity and evidence gaps, path comparison, selected disposition, prerequisites, compensation plan, verification, and owner gates.
Boundaries:
Do not invent state, side effects, commands, approvals, or recovery. Service and incident owners choose the operating path; data, financial, security, and external-system owners authorize actions in their systems; the release owner controls restart. Source: AMO-P-000297. Applicable Workflow: Recover and Requalify an Interrupted AI Agent Runtime.
Based on an Amo.ng Prompt
Long-Running Agent State Recovery Decision
Open the linked Prompt to read the instructions this Skill is based on.
Completion criteria
Complete when every material side effect and checkpoint is classified; the latest trustworthy boundary and uncertainty are explicit; one recovery disposition has prerequisites, owner, stop and rollback conditions; and observed recovery is separated from the proposed plan.
Was this useful?
Explore related Workflows
Browse WorkflowsRecover and Requalify an Interrupted AI Agent Runtime
Reconcile interrupted durable state and side effects, conditionally recertify tool access and fallback behavior, then calibrate escalation controls before resuming an AI agent runtime.
Related Prompts
Browse PromptsBuild an AI Support Triage Integration from an Approved Pilot
Implement a bounded AI-assisted support-triage integration in a sandbox or shadow environment, with abstention, grounded drafting, routing, injection tests, manual fallback, and disablement evidence.
Implement and Verify an n8n Workflow in Staging
Convert an approved n8n blueprint into a disabled importable workflow or staging implementation, with node-level traceability, synthetic execution evidence, retry tests, and rollback instructions.
Automate CRM Lead Capture and Routing in a Sandbox
Implement and verify CRM lead capture, consent, deduplication, qualification, routing, and SLA behavior in an authorized sandbox using synthetic test leads and disabled production activation.
Agent Escalation Threshold Calibration
Tune agent escalation triggers using incident severity, uncertainty, false-positive and false-negative evidence, queue capacity, delay, and owner authority.
Model Fallback Failure Analysis
Reconstruct a failed model fallback decision, test contract compatibility across routes, and determine whether to repair, restrict, or disable fallback behavior.
Tool Permission Drift Investigation
Compare approved and effective agent tool permissions over time, reconstruct permission drift, contain excess access, and define evidence-based recertification actions.