Init

SkillAI & models

Initializes Agent Skills in a project by checking prerequisites and verifying the validator configuration. Use when the user says "init", "set up agent-skills", or "initialize agent-skills".

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Init skill

What this skill tells your AI

The instructions your AI receives, as published by codagent-ai/agent-skills in skills/init/SKILL.md and read by ahel’s review.

Set up Agent Skills in the current project. This skill is idempotent — safe to re-run.

Announce at start: "Initializing Agent Skills in this project."

Steps

1. Check Prerequisites

Check that the Agent Validator CLI is installed. If missing, tell the user what to install and stop.

  • agent-validator — run agent-validator --version.
    • If not found: "Agent Validator CLI is required. Install with npm install -g agent-validator, then run agent-validator init in your project, then re-run /codagent:init."
    • If found, extract the version number and verify it is ≥ 0.15. If too old: "agent-validator 0.15 or higher is required (found <version>). Upgrade with npm install -g agent-validator@latest, then re-run /codagent:init."

Use this shell snippet to compare versions:

version_gte() { [ "$(printf '%s\n' "$2" "$1" | sort -V | head -1)" = "$2" ]; }

Example: version_gte "$installed_version" "0.15" returns true if $installed_version ≥ 0.15.

If the CLI is missing or out of date, stop. The user must resolve it first, then resume /codagent:init.

Validator config (check after CLI passes):

  • .validator/config.yml must exist. If not found: "Validator config not found. Run agent-validator init in your project first, then re-run /codagent:init." Stop.

2. Print Success

Print a summary:

Agent Skills initialized successfully.

Available skills:
- /codagent:propose — evaluate an idea and write a proposal
- /codagent:spec — interview-driven requirement discovery
- /codagent:design — brainstorm architecture and write a design doc
- /codagent:test-plan — plan automated and acceptance testing
- /codagent:plan-tasks — break a change into scoped task files
- /codagent:implement-and-validate — implement a single task and verify with the validator
- /codagent:finalize-pr — push PR, wait for CI, fix failures

3. Commit

Invoke /agent-validator:validator-commit skip to commit any scaffolding changes. Checks are skipped because init only writes boilerplate.

Guardrails

  • Stop on missing or outdated prerequisites — tell user what to install/run and resume with /codagent:init
  • Never overwrite .validator/config.yml — only add/update entry points and reviews
  • Use the hosting runtime's native mechanism to locate the plugin's root directory

Signals

GitHub stars
30
Forks
1
Last commit
Sep 2026

ahel review

  • K1binfo
    installs-packages

Automated review, not a security audit. Ruleset v1+k2.

Advanced
Catalog kind
skill
Gateway key
init-codagent-ai
Source
github.com/codagent-ai/agent-skills