Agent Skills

A small collection of practical AI agent skills.

This repo is now a simple multi-skill repository: each top-level folder is a skill with its own SKILL.md, scripts/, and optional references/ or other support files.

Skills

persistent-memory

Local persistent memory workflow for agents.

• SQLite-backed memory storage and retrieval
• helper commands for init, search, add, recent, and related maintenance
• best for agents that want a database-backed memory layer inside one workspace

telegram-readonly

Read-only access to a personal Telegram account via Telethon/MTProto.

• list dialogs
• read recent messages from a chat
• search messages
• inspect unread chats and DMs
• uses a skill-local bootstrap flow and stores auth state in ~/.config/telegram-readonly/config.json

ddg-search

Lightweight DuckDuckGo search as a no-key fallback or second source.

• text, news, image, and video search
• instant-answer lookups
• DuckDuckGo bang resolution
• local script-based skill with a skill-local bootstrap flow

twitterapi-io

Read-only Twitter/X data access via twitterapi.io.

• fetch tweets, users, timelines, replies, quote tweets, thread context, mentions, and search results
• local script-based skill, no global CLI required
• stores API key in ~/.config/twitterapi-io/config.json

Repo Layout

Each skill is self-contained:

• SKILL.md: agent-facing instructions
• scripts/: local entrypoints and helpers
• references/: optional docs or links

Some skills may also include skill-specific generated local state such as .venv/. Those should generally not be committed.
The skills installer drops dotfiles when installing skills, so some bootstrap scripts recreate a local .gitignore in the installed copy when needed.

Using A Skill

Open the skill folder and follow its SKILL.md.

Examples:

• persistent-memory
• ddg-search
• telegram-readonly
• twitterapi-io

Notes

• The skills in this repo are optimized for real agent workflows, not polished end-user CLIs.
• Keep changes minimal and keep each skill self-contained.