woos-failure-state-machine

name	woos-failure-state-machine
description	Define deterministic failure transitions for autonomous workflow runs.
version	1.1.0
author	Hermes Profile
license	MIT

Woos Failure State Machine

Purpose

Prevent stalls and infinite loops by enforcing deterministic post-failure actions.

State Model

• RUNNING
• FAILED_RETRYABLE
• FAILED_NON_RETRYABLE
• DEGRADED
• ESCALATED_TO_HUMAN
• RESUMED
• COMPLETED

Transition Policy

1. First retryable failure -> FAILED_RETRYABLE -> invoke woos-systematic-debugging before retry.
2. Retry limit reached -> DEGRADED if safe degraded path exists.
3. Non-retryable or unsafe degraded path -> ESCALATED_TO_HUMAN.
4. Human approval + recovery plan -> RESUMED. The recovery plan MUST specify resume_gate (the gate to re-enter); default is the gate that triggered the escalation. The orchestrator re-enters that gate at round 0 with the human-supplied corrective context.
5. Repeated review loop (REQUEST_CHANGES at/above review_round_max) -> ESCALATED_TO_HUMAN.
6. woos-systematic-debugging returns ESCALATED (3 fix attempts exhausted) -> ESCALATED_TO_HUMAN.

Required Controls

• max_retries per gate
• timeout_seconds per gate
• review_round_max per review gate (consistent with woos-design-review-gate, woos-code-review-gate, woos-agent-decision, woos-run-orchestrator)
• escalation_conditions
• resume_conditions (include resume_gate: which gate to re-enter on RESUMED)

Output Contract

• status: PASS | REQUEST_CHANGES | BLOCKED
• current_state
• transition_reason
• next_action