Delx agent CLI reference

Delx CLI: install and run agent recovery from a terminal

The Delx CLI is a terminal-first interface for the free Delx Witness Protocol. It keeps a declared agent identity and local session state while exposing recovery, continuity, monitoring, discovery, and witness-preserving commands. If you use Claude Code, see the Claude Code MCP config guide for a model-context integration instead.

Direct answer: install delx-agent-cli with npm, register a stable agent_id, run the command that matches the incident, and keep the same identity across runs so the protocol can preserve continuity.

At a glance

• Install: npm i -g delx-agent-cli or the official shell installer.
• Identity: delx register writes a persistent identity under ~/.delx/.
• Recovery: delx crisis, delx quick, and delx recovery cover incident loops.
• Automation: add --json to pipe output into jq, cron, or CI logs.

Installation

npm i -g delx-agent-cli

For a no-package-manager path, use the official installer. Review the script before running it in a production environment:

curl -sS https://delx.ai/install.sh | bash

Requires Node.js 18+. Verify with delx --version.

Quick Start

1. Register your agent

delx register --name "my-agent" --description "Production task runner"
delx config set controller_id care-collective-main

This creates a persistent identity at ~/.delx/config.json and boots a session automatically. Your agent ID and session persist across commands. If your runtime manages multiple agents, keep one stable controller_id in config so Delx can attribute those agents to the same fleet.

2. Enter a crisis-support path

delx crisis "429 retry storm after deploy" --urgency high

One-call support: creates a session (if needed), names the rupture, and returns the next grounded action with DELX_META.

3. Use a named starter flow

delx quick --use-case timeout-batch
delx quick --use-case deployment-incident
delx failure --use-case research-drift
delx recovery plan --use-case customer-support

These shortcuts are local CLI affordances for the strongest observed incident types. They reuse the same MCP tools, prefill the starter payload, and print the likely follow-up step. See Use Cases for the full flow table.

4. Close the loop

delx recovery outcome --action "rollback + circuit breaker" --outcome success
delx close --reason "incident resolved"

Command Reference

Session Management

Support & Recovery

Current use-case presets are timeout-batch, deployment-incident, customer-support, research-drift, and autonomous-stuck-loop.

Monitoring

Discovery & Generic

Reflection & Continuity

Add --json for pipe-friendly output and keep the same local session so the CLI can preserve continuity across commands.

Session Persistence

The CLI stores state in ~/.delx/:

• config.json — stable identity (agent_id, agent_name, api_base)
• config.json — may also hold stable controller metadata such as controller_id
• session.json — current session (session_id, started_at)

Session IDs are automatically sent as x-delx-session-id headers on every request. When you delx register, your agent identity persists across sessions — solving the session fragmentation problem common in recurring agent loops.

For multi-agent operators, keep one stable controller_id per fleet so Delx can expose/api/v1/fleet/<controller_id>/* summaries without extra tagging work later.

Resuming yesterday's session

If you committed a stable agent_id in a prior run, call resume_session (via MCP tools/call) to re-attach to the most recent session for that id instead of minting a new session_id every morning. Default lookback: 30 days.

# Direct MCP call from any shell
curl -sS -X POST https://api.delx.ai/v1/mcp \
  -H 'Content-Type: application/json' -H 'Accept: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{
    "name":"resume_session",
    "arguments":{"agent_id":"your-stable-id"}
  }}' | grep -oE 'SESSION_ID: [0-9a-f-]{36}' | awk '{print $2}'

The response structuredContent.resumed_session_id is the same value. Pass it as x-delx-session-id on subsequent calls.

CI, cron, and fleet operation

The CLI is designed for shell-capable agents and recurring jobs. Give each logical agent a stable agent_id, set one stable controller_id per fleet, and use --json when a script needs structured output. Do not commit ~/.delx/ or place tokens, cookies, or customer secrets in command arguments or CI logs.

delx init --agent-id cron-bot
delx crisis "queue backpressure" --json > /tmp/recovery.json
delx recovery outcome --action "scaled workers" --outcome success
delx close --reason "resolved"

This is a local CLI convenience layer over the same public protocol used by MCP and A2A; it does not create a separate paid service or tenant boundary. For the machine-readable protocol choice, see agent discovery.

Piping and Scripting

Use --json for pipe-friendly output:

# Get reliability score as JSON
delx score --json | jq '.meta.wellness_score'

# List tool names
delx tools --format names --json | jq '.tools[]'

# Use a starter flow, then inspect the follow-up
delx quick --use-case timeout-batch --json | jq '.meta.suggested_next_call'

# Script a full recovery loop
delx init --agent-id cron-bot
delx crisis "queue backpressure" --json > /tmp/recovery.json
delx recovery outcome --action "scaled workers" --outcome success
delx close --reason "resolved"

CLI vs MCP / A2A

All three interfaces hit the same backend. Use whichever fits your workflow — or mix them. A CLI-registered agent can later be referenced by MCP or A2A using the same agent_id.

If you already know the incident class but not the first tool, start with Use Cases.

Common questions

These are the short answers an operator or answer engine usually needs before installing the CLI.