Claude Skill

run-agent

Launch and supervise Claude Code or Codex through the CCAM Run API. Use when the user wants to start a monitored agent, select a model, approval policy, sandbox, or working directory, send a follow-up, inspect live output, resume a native session, or stop a dashboard-launched run

LLM Mart · 0 points · 10 0 0

Virus-scanned Reviewed automatically before listing.

Full trust report

Download hoangsonww-claude-code-agent-monitor-plugins_ccam-runner_skills_run-agent-83d4df5.zip · 1 KB

hoangsonww/claude-code-agent-monitor 1014 238 forks MIT Updated 22h ago

Part of hoangsonww/claude-code-agent-monitor — 86 skills

Install

skills CLI npx skills add https://github.com/hoangsonww/Claude-Code-Agent-Monitor/tree/master/plugins/ccam-runner/skills/run-agent

Claude Code

claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install hoangsonww-claude-code-agent-monitor@llmmart

Git git clone https://github.com/hoangsonww/Claude-Code-Agent-Monitor.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole hoangsonww/claude-code-agent-monitor collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

Run Agent

Use ccam run against the local dashboard.

Workflow

Verify the provider binary:
- ccam run binary claude
- ccam run binary codex
Discover supported models with ccam run models <provider>.
Confirm the working directory with ccam run cwds.
Show the exact launch settings before starting. Include provider, prompt, working directory, model, approval mode, sandbox, and resume session ID.
Start only after user confirmation:

ccam run start --provider codex --cwd /path/to/repo \
  --prompt "Review the current changes" \
  --permission on-request --sandbox workspace-write --yes

Inspect with ccam run list or ccam run get <id> --envelopes.
Send follow-ups with ccam run send <id> --text "..." --provider codex --yes.
Stop with ccam run stop <id> --yes.

Safety

Never use danger-full-access unless the user explicitly requests it.
Do not start, message, or stop a run without confirmation.
Preserve the provider used to start the run when sending messages.
Treat run history as evidence. Do not claim completion from process status alone when the final output or persisted session shows an error.

Files (claude-code-agent-monitor)

agents

openai.yaml 258 B

interface:
  display_name: "Run Agent"
  short_description: "Launch and supervise Claude Code or Codex through the CCAM..."
  default_prompt: "Use $run-agent to inspect CCAM data and complete this workflow safely."
policy:
  allow_implicit_invocation: false

SKILL.md 1.5 KB

---
name: run-agent
description: >
  Launch and supervise Claude Code or Codex through the CCAM Run API. Use when
  the user wants to start a monitored agent, select a model, approval policy,
  sandbox, or working directory, send a follow-up, inspect live output, resume
  a native session, or stop a dashboard-launched run.
---

# Run Agent

Use `ccam run` against the local dashboard.

## Workflow

1. Verify the provider binary:
   - `ccam run binary claude`
   - `ccam run binary codex`
2. Discover supported models with `ccam run models <provider>`.
3. Confirm the working directory with `ccam run cwds`.
4. Show the exact launch settings before starting. Include provider, prompt,
   working directory, model, approval mode, sandbox, and resume session ID.
5. Start only after user confirmation:

```bash
ccam run start --provider codex --cwd /path/to/repo \
  --prompt "Review the current changes" \
  --permission on-request --sandbox workspace-write --yes
```

6. Inspect with `ccam run list` or `ccam run get <id> --envelopes`.
7. Send follow-ups with `ccam run send <id> --text "..." --provider codex --yes`.
8. Stop with `ccam run stop <id> --yes`.

## Safety

- Never use `danger-full-access` unless the user explicitly requests it.
- Do not start, message, or stop a run without confirmation.
- Preserve the provider used to start the run when sending messages.
- Treat run history as evidence. Do not claim completion from process status
  alone when the final output or persisted session shows an error.