skills.rest
letta-ai
letta-aiOFFICIAL
Explore all skills by this author →
Website𝕏@letta_aiGitHub

dispatching-coding-agents

Dispatch stateless coding subagents through Claude Code and Codex CLIs for parallel research and review.

3.2k|385|Updated Oct 25, 2025
Search Tags:#subagents#CLI#Debugging#CODEX#parallel research#Code review#Claude code
One-click install
npx skills add https://github.com/letta-ai/letta-code --skill dispatching-coding-agents
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: dispatching-coding-agents
Source: https://github.com/letta-ai/letta-code/tree/main/src/skills/builtin/dispatching-coding-agents
Command: npx skills add https://github.com/letta-ai/letta-code --skill dispatching-coding-agents

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

When you are stuck on a hard bug, need a second opinion before a risky change, or want to investigate multiple hypotheses at once, doing everything serially in one agent session is slow and limited. This Skill lets you shell out to stateless coding agents (Claude Code and Codex) that run in the background while you keep working.

Core Features & Use Cases

  • Background dispatch: Run Claude Code or Codex non-interactively via Bash with run_in_background: true, then collect results later with TaskOutput.
  • Model and agent selection: Choose between Codex models (gpt-5.6-luna, terra, sol) and Claude Code aliases (sonnet, opus, fable, haiku, best) based on cost, speed, and task difficulty.
  • Session management: Resume, fork, and inspect past sessions stored on disk (~/.claude/projects/, ~/.codex/sessions/) to continue investigations without re-providing context.
  • Use Case: You have been looping on a failing test for an hour. Dispatch Codex with a structured prompt describing the failing files and what you have already tried, run Claude Code in parallel for a second opinion, and compare both root-cause analyses.

Quick Start

Ask your agent to dispatch a Claude Code subagent in the background to trace the request flow through a specific module and report back with file and line citations.

Frequently Asked Questions about dispatching-coding-agents

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

FAQPage Schema
How do I run Claude Code non-interactively from the command line?▼

Use `claude -p "YOUR PROMPT" --dangerously-skip-permissions` for non-interactive print mode. Add `--model` to select an alias like sonnet or opus, and set the working directory with `cd /path/to/repo &&` since Claude Code has no working-directory flag.

How do I run Codex as a background coding agent?▼

Use `codex exec "YOUR PROMPT" --sandbox workspace-write -C /path/to/repo` for non-interactive execution. Run it via Bash with `run_in_background: true` so you can keep working and check results later with TaskOutput.

Claude Code vs Codex for code review tasks?▼

Codex has a native review command (`codex review --uncommitted`) for local changes. Claude Code works by passing the diff inline, e.g. `claude -p "Review this diff..." --model opus`. Running both in parallel gives cross-agent validation.

When should I not dispatch a coding subagent?▼

Avoid dispatching for simple file reads, greps, small edits, or anything under roughly three minutes of direct work. If you already know exactly what to do, or context transfer would take longer than the task itself, do it yourself.

Why does Claude Code hang on large repositories?▼

Claude Code can hang on large repos when given unrestricted tool access. Restrict tools with `--allowedTools "Read Grep Glob"` (no Bash), use shorter timeouts, or switch to Codex, which handles large repositories better.

How do I resume a previous Claude Code or Codex session?▼

Claude Code: use `claude -r SESSION_ID -p "follow up"` or `claude -c` for the most recent session. Codex: use `codex exec resume SESSION_ID "prompt"` or `--last` for the most recent. Session IDs come from JSON output or the session file header.

Related Skills

Explore More Software Engineering Skills→
arbazkhan971arbazkhan971
26

research

Scan dependency manifests, docs, and git history to produce .godmode/research.md.

Community
Advanced
balic-AI-ML-R-D-Resourcesbalic-AI-ML-R-D-Resources

security-fix-review

Review security patches for root-cause coverage and bypass risks.

Official
Advanced
chadyi-kingchadyi-king

dispatch-multiple-agents

Dispatch independent tasks to parallel subagents and integrate results.

Community
Advanced
romiluz13romiluz13
11

session-handoff

Transfer active work between Pi sessions with a handoff document and intercom notification.

Community
Advanced