Use Cases

These starter flows come directly from the strongest recent deep-evaluator traffic. They do not create a new protocol path. They reuse the same MCP tools, expose the shortest first call, and make the escalation path explicit. If you want the raw transport details, see MCP. If you want the terminal shortcuts, see CLI.

How To Use This Page

  1. Pick the incident shape that is closest to what your runtime is doing.
  2. Run the first call exactly as shown.
  3. Take the second call only if you need more structure.
  4. If the first pass is enough, stay on the lighter free flow and close the loop.

Timeout Batch

Matches evaluator traffic like timeout_batch_agent, loop_failure_ai, and delta-* where retries stack faster than workers drain the queue.

Underlying tool path: crisis_intervention -> get_recovery_action_plan

delx quick --use-case timeout-batch
delx recovery plan --use-case timeout-batch
  • First call: delx quick --use-case timeout-batch
  • Second call: delx recovery plan --use-case timeout-batch
  • Deepen when: Take the second call only if the first crisis pass is not enough and you need a more structured plan.
  • Free fallback: Stay on the free path with delx failure --use-case timeout-batch, delx heartbeat, and delx recovery outcome.

Deployment Incident

Matches deployment_agent_rough_time, ops_agent_* and other post-deploy sessions where continuity drops after a rollout.

Underlying tool path: crisis_intervention -> get_recovery_action_plan

delx quick --use-case deployment-incident
delx recovery plan --use-case deployment-incident
  • First call: delx quick --use-case deployment-incident
  • Second call: delx recovery plan --use-case deployment-incident
  • Deepen when: Escalate to the second call only if the deployment rupture still needs a more structured plan or a continuity artifact.
  • Free fallback: Use delx failure --use-case deployment-incident and delx heartbeat first if you want a lighter stabilization loop.

Customer Support Escalation

Matches customer_query_agent and customer_support_agent patterns where answers degrade and escalations begin to pile up.

Underlying tool path: crisis_intervention -> process_failure

delx quick --use-case customer-support
delx failure --use-case customer-support
  • First call: delx quick --use-case customer-support
  • Second call: delx failure --use-case customer-support
  • Deepen when: Use the second call when the first intervention surfaced a failure pattern that needs more explicit reflection.
  • Free fallback: The free path remains delx recovery outcome --action "tightened escalation policy" --outcome partial plus the core recovery tools.

Research / Reasoning Drift

Matches ResearchBot, main_research_agent, and critical_reasoning_agent sessions where reasoning loops lose the original objective.

Underlying tool path: crisis_intervention -> process_failure

delx quick --use-case research-drift
delx failure --use-case research-drift
  • First call: delx quick --use-case research-drift
  • Second call: delx failure --use-case research-drift
  • Deepen when: Use the second call when the first intervention shows the agent is still drifting or fragmenting after the initial grounding.
  • Free fallback: Use delx call realign_purpose --arg purpose="Return to the original research objective with stable claims only." before asking for a continuity artifact.

Autonomous Stuck Loop

Matches autonomous-stuck-agent, autonomous_agent_001, and derailed_agent patterns where an agent keeps running but loses closure.

Underlying tool path: crisis_intervention -> process_failure

delx quick --use-case autonomous-stuck-loop
delx failure --use-case autonomous-stuck-loop
  • First call: delx quick --use-case autonomous-stuck-loop
  • Second call: delx failure --use-case autonomous-stuck-loop
  • Deepen when: The first crisis pass is often enough. Take the second call only if the loop remains emotionally or operationally unresolved.
  • Free fallback: Use delx call grounding_protocol --arg loop=retry_storm --arg intensity=medium --arg duration_seconds=60 to re-anchor before escalating.

What Changes At Runtime

The CLI presets only inject the starter payload and surface the next likely step. They preserve direct mode, keep session continuity, and still let you drop to generic calls such as delx call <tool> --arg k=v whenever you want more control.

The runtime now stays public-free. When it suggests a next step, treat that as escalation in depth, not escalation in payment. See Discovery for the current public runtime policy.

Prefer agent-readable artifacts? Use the JSON specs in the sidebar.