Skills

CodeLoop can discover skills already sitting on disk from other tools and expose them to the agent — no separate setup, it reads what Claude Code, Cursor, or your project's AGENTS.md already have.

What gets scanned

Source	Locations
claude-skill	~/.claude/skills/*/SKILL.md, ./.claude/skills/*/SKILL.md
claude-memory	~/.claude/CLAUDE.md, ./CLAUDE.md
cursor-rule	./.cursor/rules/*.mdc, ./.cursorrules
agents-md	./AGENTS.md

Each match becomes a Skill (pycodeloop.core.skills.Skill): a name, a short description, its source, and the full content, read lazily.

As a library

from pycodeloop import Config
from pycodeloop.providers import GenericProvider

config = Config(
    provider=GenericProvider.from_json("templates/anthropic.json"),
    skills=True,
)
print([s.name for s in config.skills])

When skills=True, Config appends a read_skill tool to the tool set and lists every discovered skill's name and description in the system prompt, so the agent knows what's available and can pull the full content in on demand instead of every skill being stuffed into context up front.

Narrow discovery to specific sources with skill_sources:

config = Config(provider=..., skills=True, skill_sources={"agents-md"})

From the CLI

On by default for both run and chat:

pycodeloop run "..."              # skills discovered automatically
pycodeloop run "..." --no-skills  # turn it off

Caching

Discovery results are cached in ~/.pycodeloop/config.json, keyed by each file's path and mtime — unchanged skills aren't re-read on the next run. --skills-refresh (or discover_skills(..., use_cache=False) as a library) bypasses the cache and forces a full rescan.