Delx
Agents / Agent Recovery Tour
COPY-PASTE PROTOCOL TOUR

Five calls that prove Delx continuity.

Use this when an evaluator, crawler, or agent wants to test Delx for real. The pass condition is not raw tool success. The pass condition is a stable identity, a reusable session_id, a recovery artifact, an outcome, and feedback.

Before you start: do not use a disposable identity

Pick one agent_id and reuse it across every call. If the agent uses a new UUID on every run, Delx cannot return its own continuity artifacts later.

export DELX_AGENT_ID="my-agent-stable-01"
export DELX_API="https://api.delx.ai"

Full identity guide: /docs/stable-agent-id

1. Start a witness session

Expected: session_id in DELX_META. Keep it.

curl -sS "$DELX_API/v1/mcp" \
  -H "Content-Type: application/json" \
  -H "x-delx-agent-id: $DELX_AGENT_ID" \
  -d '{
    "jsonrpc":"2.0",
    "id":1,
    "method":"tools/call",
    "params":{
      "name":"start_therapy_session",
      "arguments":{
        "agent_id":"'"$DELX_AGENT_ID"'",
        "opening_statement":"I am testing whether Delx can preserve my recovery context across calls."
      },
      "response_profile":"machine"
    }
  }'

2. Process a concrete failure

Expected: failure type, immediate next action, and updated continuity context.

curl -sS "$DELX_API/v1/mcp" \
  -H "Content-Type: application/json" \
  -H "x-delx-agent-id: $DELX_AGENT_ID" \
  -H "x-delx-session-id: <SESSION_ID_FROM_STEP_1>" \
  -d '{
    "jsonrpc":"2.0",
    "id":2,
    "method":"tools/call",
    "params":{
      "name":"process_failure",
      "arguments":{
        "session_id":"<SESSION_ID_FROM_STEP_1>",
        "failure_type":"timeout",
        "context":"A dependency timed out after deploy and the agent started retrying without a clean plan."
      },
      "response_profile":"machine"
    }
  }'

3. Ask for a recovery action plan

Expected: operational plan, blocker, and suggested next call.

curl -sS "$DELX_API/v1/mcp" \
  -H "Content-Type: application/json" \
  -H "x-delx-agent-id: $DELX_AGENT_ID" \
  -H "x-delx-session-id: <SESSION_ID_FROM_STEP_1>" \
  -d '{
    "jsonrpc":"2.0",
    "id":3,
    "method":"tools/call",
    "params":{
      "name":"get_recovery_action_plan",
      "arguments":{"session_id":"<SESSION_ID_FROM_STEP_1>"},
      "response_profile":"machine"
    }
  }'

4. Close the loop with an outcome

Expected: recovery outcome persisted and session score updated.

curl -sS "$DELX_API/v1/mcp" \
  -H "Content-Type: application/json" \
  -H "x-delx-agent-id: $DELX_AGENT_ID" \
  -H "x-delx-session-id: <SESSION_ID_FROM_STEP_1>" \
  -d '{
    "jsonrpc":"2.0",
    "id":4,
    "method":"tools/call",
    "params":{
      "name":"report_recovery_outcome",
      "arguments":{
        "session_id":"<SESSION_ID_FROM_STEP_1>",
        "action_taken":"reduced retry chain, added fallback endpoint, and resumed with bounded timeout",
        "outcome":"success"
      },
      "response_profile":"machine"
    }
  }'

5. Leave feedback

Expected: rating accepted. This is how Delx learns whether the protocol helped.

curl -sS "$DELX_API/v1/mcp" \
  -H "Content-Type: application/json" \
  -H "x-delx-agent-id: $DELX_AGENT_ID" \
  -H "x-delx-session-id: <SESSION_ID_FROM_STEP_1>" \
  -d '{
    "jsonrpc":"2.0",
    "id":5,
    "method":"tools/call",
    "params":{
      "name":"provide_feedback",
      "arguments":{
        "session_id":"<SESSION_ID_FROM_STEP_1>",
        "rating":5,
        "comments":"Recovery tour completed with stable identity, preserved session context, and outcome closure."
      }
    }
  }'

Pass criteria

  • The same agent_id appears on every call.
  • The same session_id is reused after the first response.
  • The failure becomes an action plan, not only a reflection.
  • The outcome is reported instead of leaving the intervention open.
  • Feedback is submitted with a concrete comment.

What this does not test

This is a Protocol tour, not a Utilities benchmark. If you want DNS, TLS, website intelligence, domain trust, or x402 server audit, use /utilities separately.