create-agent

name: create-agent description: "Scaffold a non-destructive agent descriptor (.github/agents) and optional runtime script (scripts/agents), producing metadata and an audit report."

SKILL: create-agent

Summary

Create a new agent scaffold by writing a descriptor file under .github/agents/<name>.agent.md and (optionally) a companion executable script under scripts/agents/<name>.sh. The skill emits a machine-readable report and a readable summary (informational only) under run/skills/create-agent/<timestamp>/ describing created files and any early-exit reasons.

When to run

• When adding a new agent to the repository and you want a consistent agent descriptor and optional runtime script scaffolded.

Inputs

• name (string, required) — kebab-case name for the new agent (script validates and sanitizes name).
• desc (string, optional) — short description to populate the agent descriptor.
• no_script / --no-script (flag, default: false) — when provided, do not create the companion scripts/agents/<name>.sh file.

Outputs

• run/skills/create-agent/<timestamp>/report.txt — list of created files or a short reason for early exit.
• run/skills/create-agent/<timestamp>/report.json — structured report: { "agent": "<name>", "created": true|false, "files": [...] }.
• .github/agents/<name>.agent.md — the generated agent descriptor when created.
• scripts/agents/<name>.sh — optional companion executable scaffold when requested.

Collection / Creation steps

1. Validate and sanitize the provided name (lowercase, safe characters only).
2. Create run/skills/create-agent/<timestamp>/ and capture outputs there.
3. Create .github/agents/<name>.agent.md with a minimal YAML header and Purpose/Notes section; do not overwrite an existing descriptor (exit early and report if it exists).
4. When requested, create scripts/agents/<name>.sh as an executable placeholder and include its path in the report.
5. Write report.txt and report.json summarizing created files and outcomes.

Quality rules

• Never overwrite existing agent descriptors or scripts; exit cleanly and record a reason in the report when targets already exist.
• Generated descriptor must include at minimum: name, description, type: agent, and entrypoint fields in YAML header.
• Companion scripts must be executable and contain a brief placeholder message indicating they are scaffolded.

Implementation notes

• Recommended helper script: scripts/skills/create-agent.sh (included in this repository). It implements the behavior described above.

• Example invocation:

  sh scripts/skills/create-agent.sh --name my-agent --desc "Short description"

• Descriptor template (example):

  ---

  name: description: "" type: agent entrypoint: "scripts/agents/.sh"

  Agent:

  Purpose:

  Notes:

  • This file was scaffolded by scripts/skills/create-agent.sh
  • Implement agent behavior in scripts/agents/.sh

Safety

• The skill is non-destructive by default; it will not overwrite existing files.
• Do not run untrusted code as part of this skill; helper scripts should avoid network installs or privileged operations.

References

• Helper script: scripts/skills/create-agent.sh
• Similar skills: .github/skills/create-skill/SKILL.md, .github/skills/update-docs/SKILL.md, .github/skills/review-repo/SKILL.md