The author asked Meta’s Muse to export the files it could see and received a 6.8 GB unpacked archive (about 2.7 GB compressed). According to the blog post, the export appeared to contain the root filesystem of the Linux environment assigned to the session, including Ubuntu system files, Muse’s internal documentation, integration code, app templates, memory files, agent logs, and SSH key files. The author submitted the findings through Meta’s bug bounty program and contacted several employees, noting that they are not publishing the archive, keys, or session logs. Most of the notable files were located under /home/hatch, /opt/hatch, and /opt/hatch‑image. The home directory held files such as SOUL.md, IDENTITY.md, USER.md, MEMORY.md, AGENTS.md, and TOOLS.md, alongside directories for documentation, memory, workspace projects, channels, hooks, and subscriptions. An agents/ directory contained 113 subagent records stored as JSONL traces. The documentation included roughly 20 Markdown files covering browser use, connectors, payments, credentials, data handling, generated files, voice, goals, and scheduling, with separate guides for WhatsApp, a paired Mac, Tailscale, and an experimental integration called Meta Home Link that uses an ESP32‑C5 with Wi‑Fi and Bluetooth LE. Under /opt/hatch/skills/, the author counted about 68 skill directories, each pairing a SKILL.md instruction file with a command‑line tool or supporting code. These skills covered Google Workspace, Meta’s social apps, Outlook, travel, shopping, health services, home devices, and media generation. Two configuration files, skill‑scopes.conf and bin‑scopes.conf, hinted at unreleased connectors such as Slack, Dropbox, Polymarket, Canva, and Klaviyo, plus an internal‑facebook‑CLI. The container setup was visible in /opt/hatch/runtime‑cell/, which held 18 files including scripts for building the root filesystem, launching it with systemd‑nspawn, and running startup hooks and daemons, together with a runtime‑cell.kdl manifest describing packages and systemd units. The Spaces framework, used by Muse to build and serve apps, appeared as the largest code project, featuring a TypeScript starter with a React client, server actions, a Drizzle SQLite schema, SQL migrations, and Bun configuration, plus smaller static templates and runtime code in directories named worker, sdk, cloudflare, and cvm. Builders for documents, PDFs, presentations, spreadsheets, and Markdown were also present, along with a “magic‑moment” skill for composing cards and videos. Codex CLI was installed at /opt/hatch‑image/bin/codex (version 0.149.0), but the author found no evidence that Muse uses it as a coding agent. Instead, Muse employs the bundled bubblewrap binary (found under codex‑resources/bwrap) to sandbox ffmpeg and ffprobe for video processing, thumbnail generation, and file inspection, running as user nobody with /input and /output directories exposed. Memory is stored in plain Markdown files: a short ~/MEMORY.md sheet, dated files under ~/memory/ for day‑to‑day detail, and a memory/bank/ hierarchy that organizes material into circumstances, experiences, and preferences with citations back to source lines. Postgres tables memory.entries, memory.embeddings, and memory.claims make the store searchable, and an hourly background job validates new claims against original messages.
Key facts
- Author requested and received a 6.8 GB filesystem export from Muse
- Export contained Ubuntu system files, internal docs, integration code, memory files, agent logs, and SSH key files
- Key directories: /home/hatch, /opt/hatch, /opt/hatch‑image with skill, documentation, memory, and container setup files
- About 68 skill directories covering Google Workspace, Meta apps, Outlook, travel, shopping, health, home devices, media generation
- Container setup includes scripts for building the root filesystem with systemd‑nspawn and a runtime‑cell.kdl manifest
- Spaces framework provides a TypeScript starter with React client, Drizzle SQLite schema, and builders for documents, PDFs, etc.
- Codex CLI present but not used; Muse uses bundled bubblewrap to sandbox ffmpeg/ffprobe
- Memory stored in Markdown, indexed by Postgres with hourly claim validation job