Agent-Skills.md

Agent Skills: Claude Command Dispatcher

Use when the user asks to run or reuse a Claude Code custom command from .claude/commands (for example '/plan', '/tasks', '/specify', '/chat_sum'). Dynamically resolves any command file in .claude/commands without hardcoded command names.

UncategorizedID: tekliner/improvado-agentic-frameworks-and-skills/claude-command-dispatcher
41

Install this agent skill to your local

pnpm dlx add-skill https://github.com/tekliner/improvado-agentic-frameworks-and-skills/tree/HEAD/skills/claude-command-dispatcher

Skill Files

Browse the full folder contents for claude-command-dispatcher.

Download Skill

Loading file tree…

skills/claude-command-dispatcher/SKILL.md

Skill Metadata

Name
claude-command-dispatcher
Description
"Use when the user asks to run or reuse a Claude Code custom command from .claude/commands (for example '/plan', '/tasks', '/specify', '/chat_sum'). Dynamically resolves any command file in .claude/commands without hardcoded command names."

Claude Command Dispatcher

Bridge skill that lets Codex execute command prompts stored in .claude/commands/*.md.

When To Use

Use this skill when the user:

  • references a Claude-style command like /plan or /tasks
  • asks to reuse or run something from .claude/commands
  • asks to adapt a new command that was just added there

Do not hardcode command names. Always resolve the command file dynamically.

Execution Flow

  1. Parse command name and command arguments from the user request:
  • Accept both /name ... and name ...
  • name maps to .claude/commands/name.md
  • the rest of the text is ARGUMENTS
  1. Resolve repository root dynamically:
git rev-parse --show-toplevel

If unavailable, use current working directory.

  1. Build command file path:
  • <repo-root>/.claude/commands/<name>.md
  1. If file does not exist:
  • list available commands with:
ls -1 <repo-root>/.claude/commands/*.md
  • ask the user to choose a valid command name
  1. Read the command markdown file and execute its instructions as the task definition:
  • treat $ARGUMENTS placeholders as the parsed ARGUMENTS
  • preserve instruction order from the command file
  • respect referenced templates/scripts in the command instructions
  1. Return results in normal Codex style:
  • what was executed
  • output artifacts and paths
  • blockers/errors if any

Guardrails

  • Never use absolute user-specific paths in new logic.
  • Prefer repository-relative paths rooted from dynamic repo root.
  • If a command file itself contains absolute paths, use them only as command intent, but keep any new code generic.