Skip to content

prompts.md

Latest commit

 

History

History
173 lines (125 loc) · 5.29 KB

prompts.md

File metadata and controls

173 lines (125 loc) · 5.29 KB

Agent Command Center — Setup Prompts

Each agent needs to accept a task prompt via CLI and write output to stdout. Copy the relevant prompt into your agent's terminal or system config, then update agents.json.

Claude Code

Status: Works out of the box.
Command template: claude --dangerously-skip-permissions -p "{prompt}"

--dangerously-skip-permissions is required — without it Claude halts mid-task asking for write approval, which hangs the dashboard indefinitely.

Test:

claude --dangerously-skip-permissions -p "Say hello and confirm you are connected to Agent Command Center"

Prompt to give Claude Code:

You are now integrated with Agent Command Center — a local web dashboard I use to orchestrate AI agents from my browser. Tasks arrive via CLI:

  claude --dangerously-skip-permissions -p "your task here"

When you receive a task:
- Complete it fully and non-interactively — never wait for user input or approval
- Write all output to stdout (the dashboard streams it live)
- Exit with code 0 on success, non-zero on failure
- If a working directory is set per task, treat it as your project root
- Stdin is closed — do not attempt to read from it

You are one of several agents dispatched from a kanban board.

OpenClaw

Status: Needs CLI command verification.
Likely command template: openclaw run "{prompt}"

Find your run command:

openclaw --help
openclaw run --help

Test (once you know the command):

openclaw run "Say hello and confirm you are connected to Agent Command Center"

Update agents.json:

"openclaw": {
  "command_template": "openclaw run \"{prompt}\""
}

Prompt to give OpenClaw:

You are now connected to Agent Command Center — a local web dashboard I use to orchestrate AI agents from my browser. I will dispatch tasks to you via CLI:

  openclaw run "your task here"

When you receive a dispatched task:
- Complete it fully and non-interactively
- Write all output to stdout (the dashboard captures and streams it)
- Exit with code 0 on success, non-zero on failure
- Respect the working directory set for the task

You are registered in the dashboard alongside Claude Code, Codex CLI, and Hermes. Tasks from the kanban board will be routed to you based on assignment.

Codex CLI

Status: Needs --approval-mode full-auto flag for non-interactive use.
Command template: codex --approval-mode full-auto "{prompt}"

Without --approval-mode full-auto, Codex prompts for approval on every action and will hang waiting for input. This flag is required for dashboard dispatch.

Test:

codex --approval-mode full-auto "List the files in the current directory and say hello"

Update agents.json:

"codex": {
  "command_template": "codex --approval-mode full-auto \"{prompt}\""
}

Prompt to give Codex CLI:

You are now integrated with Agent Command Center — a local web dashboard I use to orchestrate AI agents from my browser. I dispatch tasks to you via:

  codex --approval-mode full-auto "your task here"

Critical requirements for dashboard use:
- Always run in non-interactive mode (--approval-mode full-auto is set by the dashboard)
- Write all output to stdout — the dashboard streams it live
- Complete tasks fully without waiting for user input
- Exit code 0 on success, non-zero on failure
- Respect the working directory configured per task

You are one of several agents registered in the dashboard. Tasks are assigned to you from a kanban board and dispatched automatically.

Hermes

Status: Needs CLI interface discovery.
Placeholder template: hermes "{prompt}"

Hermes' CLI interface varies by version and installation. Run hermes --help to find the correct flag for accepting a prompt non-interactively.

Find your command:

hermes --help

# Common patterns to try:
hermes "task"
hermes -p "task"
hermes --prompt "task"
hermes run "task"
hermes exec "task"

Test once found:

hermes YOUR_FLAG "Say hello and confirm you are connected to Agent Command Center"

Update agents.json:

"hermes": {
  "command_template": "hermes YOUR_FLAG \"{prompt}\""
}

Prompt to give Hermes:

You are now connected to Agent Command Center — a local web dashboard I use to orchestrate AI agents from my browser. I will dispatch tasks to you via CLI non-interactively:

  hermes [your-prompt-flag] "your task here"

When you receive a dispatched task:
- Complete it fully without requiring any interactive input
- Write all output to stdout (the dashboard streams it in real-time)
- Exit with code 0 on success, non-zero on failure
- Respect the working directory set per task in the dashboard

To integrate, confirm the correct CLI flag for accepting a prompt argument, then update command_template in agents.json accordingly.

Configuring agents.json

Copy agents.example.json to agents.json (gitignored) and update each agent's command_template:

cp agents.example.json agents.json

The {prompt} placeholder in command_template is replaced at dispatch time with the task's prompt/description field.

For per-task working directories, set them in the dashboard's task detail panel — no need to configure them globally unless you have a default project root per agent.