Verified Agent Skill record

Sub Agent

A generic skill for delegating self-contained tasks to a coding sub-agent (Codex, Claude Code, OpenCode, or Pi) via background processes. It serves as a spawn point for any LLM-driven sub-task, such as building features, reviewing PRs, refactoring, document generation, or report synthesis.

Tasks

Build a new application feature by spawning a sub-agent in the target project directory using codex exec --full-auto.
Review a pull request safely by cloning the repository to a temporary directory, spawning a sub-agent to review the diff, and cleaning up the temporary directory afterward.
Fix multiple issues in parallel by creating separate git worktrees for each issue and spawning a background sub-agent in each worktree.

Inputs

Agent prompts and task descriptions.
Working directory (workdir) for the agent to operate within.
CLI commands formatted for non-interactive modes (e.g., codex exec, claude --print, opencode run, pi -p).

Outputs

Modified files or new code created by the sub-agent in the specified workdir.
Foreground command output or background session tracking via session_id.
Background process logs, status updates, and completion messages retrieved via process actions.

Limitations and checks

Not intended for simple one-liner fixes, reading code, thread-bound ACP harness requests, or work inside ~/clawd workspace.
Avoid using a pty parameter or --dangerously-skip-permissions with Claude Code, as it can exit after an interactive confirmation dialog.
Cannot spawn agents safely inside the OpenSquilla state directory ($OPENSQUILLA_STATE_DIR) or live runtime workspace directories.
Verify the target workdir is an explicit project directory or git worktree, and is never the OpenSquilla state directory.
Confirm that the chosen CLI command uses non-interactive modes such as codex exec, claude --print --permission-mode bypassPermissions, opencode run, or pi -p.
If using Codex, ensure the working directory is a git repository (run git init if it is scratch space).
When spawning background agents, confirm the prompt instructs the agent to print a clear completion line for easy log monitoring.