skills.rest
ceasarXuu
ceasarXuuCOMMUNITY
Explore all skills by this author →
GitHub

agent-commander

Delegate bounded read-only coding analysis to local coding agents and continue their sessions.

3|Updated Mar 20, 2026
Search Tags:#code review#claude code#codex#delegation#multi-agent#session management#read-only analysis
One-click install
npx skills add https://github.com/ceasarXuu/AstartesSkills --skill agent-commander-ceasarxuu
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: agent-commander
Source: https://github.com/ceasarXuu/AstartesSkills/tree/main/plugins/agent-commander/skills/agent-commander
Command: npx skills add https://github.com/ceasarXuu/AstartesSkills --skill agent-commander-ceasarxuu

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Codex sometimes needs a second opinion or repository-aware analysis from another locally installed coding agent, but manually invoking external CLIs, tracking their session IDs, and continuing the same conversation is error-prone. This Skill standardizes that delegation while keeping Codex in control of scope, authorization, and verification. ## Core Features & Use Cases - Multi-provider delegation: Run bounded read-only tasks on Command Code, Claude Code, OpenCode, or Pi through a single Python adapter that emits structured JSON results. - Session continuity: Preserve the explicit provider session ID and issue follow-up prompts into the exact same external conversation instead of starting over. - Safety boundaries: Enforces read-only operation, forbids permission-bypass flags, requires explicit session IDs, and disables DeepSeek Harness execution in v0.2.1. - Use Case: Ask Codex to have Claude Code review a specific module for potential bugs, then send a follow-up question into that same Claude Code session and compare its claims against Codex's own verification. ## Quick Start Ask Codex to check which external coding agents are available, delegate a bounded read-only review of a specific directory to the best provider, and report the findings with the session ID preserved.

Frequently Asked Questions about agent-commander

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I delegate a coding task from Codex to Claude Code?▼

Run the delegate_agent.py adapter with the run command, specifying --provider claude-code, an absolute workspace path, and a bounded read-only prompt. The adapter returns a JSON object containing the result and a session ID you can reuse for follow-ups.

How do I continue the same external agent session after delegation?▼

Use the adapter's follow-up command with the same provider, the exact session_ref.id from the previous response, and the same workspace path. Never use a 'continue latest' option, since only explicit session IDs guarantee the correct conversation.

Which external coding agents are supported for delegation?▼

Command Code, Claude Code, OpenCode, and Pi are supported for execution. DeepSeek Harness is discovery-only in v0.2.1 because its minimal SDK profile exposes a high-permission shell, so it cannot be executed.

Can the delegated agent edit files or run implementation tasks?▼

No. Version 0.2.1 is strictly read-only and must not be used for implementation, file edits, commits, pushes, or account changes. Permission-bypass flags like --yolo or --dangerously-skip-permissions are explicitly forbidden.

What happens if the delegated agent returns a partial or failed result?▼

A completed status contains usable text, while partial means the process exited without a complete final response. All other statuses must be reported as failures to the user without pretending the delegation succeeded.

Related Skills

Explore More Software Engineering Skills→
harivansh-afkharivansh-afk
3

coding-agent

Run Codex CLI, Claude Code, OpenCode, or Pi Coding Agent in background processes.

Community
Advanced
thegeronimothegeronimo
44

static-review

Identify and prioritize defects in multi-agent tickets via read-only analysis.

Community
Advanced
openvettaopenvetta
154

evidence-review

Verify claims against concrete evidence and flag unsupported assertions before reporting conclusions.

Official
Basic
cajiascajias

q-chat-integration

Delegate specialized tasks to Amazon Q CLI for non-interactive execution.

Community
Advanced