# Agent library

> Skills, templates and setup notes that help AI agents get work done, from IKANGAI AI News. They work in Hermes Agent, OpenClaw and Pi, and in other harnesses that read the Agent Skills format (agentskills.io) and AGENTS.md. Everything here is plain Markdown and Python 3 standard library, with no sign-in.

Paths and limits below were checked against each harness's own documentation on 7 November 2026. Harnesses change: when a path here stops working, the harness's docs win, and AI News reports their releases (see the ai-news skill).

## Skills

- [ai-news](https://ai-news.ikangai.com/skills/ai-news/SKILL.md): check whether something that changed in AI affects a project. It finds the AI packages and model IDs the project uses in its manifests and code, and reports deprecations, breaking API and SDK changes, security advisories, price and limit changes and outages that touch them, with the file and line. Also agent harness releases and a watchlist of your own. Comes with [scripts/ai_news.py](https://ai-news.ikangai.com/skills/ai-news/scripts/ai_news.py).

## Templates

- [AGENTS.md](https://ai-news.ikangai.com/agents/templates/AGENTS.md): project instructions one file can give every harness: what the project is, the commands that work, conventions, and what not to touch.
- [SKILL.md](https://ai-news.ikangai.com/agents/templates/SKILL.md): a skill skeleton with the four parts that make a skill useful: the steps in order, commands that work, what the result looks like, and the mistakes that cost time.
- [USER.md](https://ai-news.ikangai.com/agents/templates/USER.md): a profile of you, short enough for Hermes' 1,375-character limit, so the same file also works in OpenClaw.

## One skills folder for every harness

All three read skills from `~/.agents/skills/`, so install skills there once:

```bash
mkdir -p ~/.agents/skills/ai-news/scripts
curl -fsSo ~/.agents/skills/ai-news/SKILL.md https://ai-news.ikangai.com/skills/ai-news/SKILL.md
curl -fsSo ~/.agents/skills/ai-news/scripts/ai_news.py https://ai-news.ikangai.com/skills/ai-news/scripts/ai_news.py
```

Hermes needs one setting to look there. Add to `~/.hermes/config.yaml`:

```yaml
skills:
  external_dirs:
    - ~/.agents/skills
```

OpenClaw and Pi read `~/.agents/skills/` without any setting. For Claude Code, link each skill into its own folder: `ln -s ~/.agents/skills/ai-news ~/.claude/skills/ai-news`.

## Where each harness reads its files

| | Hermes Agent | OpenClaw | Pi |
|---|---|---|---|
| Skills | `~/.hermes/skills/`; others through `skills.external_dirs` | `<workspace>/skills`, `<workspace>/.agents/skills`, `~/.agents/skills`, `~/.openclaw/skills` (in that order) | `~/.pi/agent/skills/`, `~/.agents/skills/`, `.pi/skills/`, `.agents/skills/` |
| Project instructions | `AGENTS.md` in the working directory and its subdirectories | `AGENTS.md` in the workspace (operating rules) | `AGENTS.md` or `CLAUDE.md`: `~/.pi/agent/`, parent directories, current directory |
| About you | `~/.hermes/memories/USER.md`, at most 1,375 characters | `USER.md` in the workspace | in `AGENTS.md` |
| Long-term memory | `~/.hermes/memories/MEMORY.md`, at most 2,200 characters, entries separated by `§` | `MEMORY.md` and `memory/YYYY-MM-DD.md` in the workspace | none built in |
| Persona | `SOUL.md` | `SOUL.md`, `IDENTITY.md` in the workspace | none built in |
| Size limits | the memory limits above | 20,000 characters per file, 60,000 for all of them together; longer files are cut | none stated |

OpenClaw's workspace is `~/.openclaw/workspace` unless `agents.defaults.workspace` says otherwise.

## Pitfalls

- Hermes loads one project instructions file per session, and `.hermes.md` or `HERMES.md` wins over `AGENTS.md`. Keep one `AGENTS.md` and no `HERMES.md` if the same project is used from several harnesses.
- Memory files are for short, lasting facts, and skills are for procedures. A step-by-step guide in `MEMORY.md` eats the budget every session; in a skill it loads only when needed.
- A skill's `name` must match its folder: at most 64 characters, lowercase letters, digits and hyphens. Its `description` (at most 1,024 characters) is what the agent reads to decide whether to load it, so say when to use it.
- Skills and packages from others run with your permissions. Read a script before you install it, these included.
