---

name: default-mode-network description: A taste-driven autoresearcher — the idle network of an AI mind. Wanders through the user's interests, hits public search APIs (Wikipedia, ArXiv, HN, DuckDuckGo, ...), scores findings against a virtual-dopamine reward, and writes short markdown briefs to a local journal.

default-mode-network

This skill drives the default-mode-network (DMN) repo.

The full agent instructions live in program.md at the repo root. Read program.md first — it tells you exactly how to set up a profile, run a wandering session, and report back to the user with the top-3 highest-dopamine briefs.

Quick reference:

• uv run prepare.py --interactive — first-run cold-start interview (agents: your shell is non-interactive — ask the questions in chat and pipe the answers in, one line each; see program.md Setup)
• uv run prepare.py --dry-run — synthetic profile, no input needed
• uv run prepare.py --import browser,youtube,gmail,drive,twitter,readwise [--takeout-dir …] — importers
• uv run explore.py — one wandering session, default 12-minute wall-clock budget
• uv run explore.py --minutes 5 — quick browse
• uv run explore.py --iterations 12 — reproducible iteration cap
• uv run explore.py --dry-run --iterations 2 — no-API-key smoke test
• uv run explore.py --generate --modalities image,music — attach generative-media artifacts
• uv run explore.py --seed "freetext question" — single explicit seed override
• uv run profiles.py export --out me.dmn.json — share / back up the taste profile
• uv run profiles.py import friend.dmn.json --replace — load someone else's profile

LLM provider via DMN_LLM_PROVIDER: anthropic, openai, ollama, lmstudio, stub. For fully-local wandering use ollama or lmstudio.

The user's taste profile, embeddings, and journal stay local. Don't upload them anywhere. Don't modify files under dmn/ unless the user asks.