hn.today

How Meta's Muse works, revealed by the 6.8 GB filesystem it sent me

mouse.dev21 points15 comments
Screenshot of How Meta's Muse works, revealed by the 6.8 GB filesystem it sent me

A researcher requested Meta’s Muse to archive the files visible to its session and received a roughly 2.7 GB compressed (6.8 GB unpacked) download that appears to be the root filesystem of the Linux environment assigned to the session. The export included Ubuntu system files, Muse’s internal documentation, integration code, app templates, memory files, agent logs, and apparent SSH key files. The researcher reported the finding through Meta’s bug‑bounty program and did not publish the archive; the core concern is that internal runtime files and sensitive material can be exported through an ordinary conversation and a connected destination, though active key access was not established.

The filesystem layout and contents give a detailed picture of Muse’s runtime: a /home/hatch with SOUL.md, IDENTITY.md, USER.md, MEMORY.md, AGENTS.md and TOOLS.md; an agents/ directory with 113 subagent JSONL traces; ~20 Markdown docs explaining connectors and integrations (including a Home Link device); ~68 skill directories under /opt/hatch/skills; a runtime-cell with scripts to build and launch the container via systemd-nspawn; a Spaces TypeScript app framework and file builders for docs, presentations and media; a bundled Codex CLI and a bubblewrap binary used to sandbox ffmpeg/ffprobe. Memory is stored as plain Markdown plus Postgres tables (memory.entries, memory.embeddings with 384‑dim vectors, memory.claims), hourly and nightly jobs curate and synthesize memories and “dream” summaries that influence future sessions.

Read on mouse.dev15 comments on Hacker News

Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.

More in Security

The daily digest

Today's best Hacker News stories, summarized and screenshotted, one email a day.