atlasBLOCK
Grow a knowledge vault by talking to your agents. Read it like a book.
Overview
From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.
A knowledge vault you grow by talking to your agent, and read in the browser.
You ask a question in Atlas's chat. Your agent researches it, and you watch the first captures come in. They get distilled into insights; once a few exist, Atlas suggests maps; "I have an idea" becomes a sketch. Everything is plain Markdown in vault/, in your own git repository. No database, no account, no onboarding.
Quickstart
- Install the skill once, for every agent harness on this machine:
npx skills add a-ratz/atlas -g -s atlas.
- In Claude Code, Codex or any other harness, say "start atlas". The first
time, it asks where Atlas should live (default ~/atlas) and installs it there; this folder is then your Atlas. It also offers demo notes, so you can try a full vault at once; "clean the demo" removes them again.
- Open http://localhost:8765 and type what you are curious about, or
/
for commands.
The agent needs uv to run the engine and installs it after asking you if it is missing. uv brings Python and the engine's packages along; nothing else is required.
Optional extras:
- git, for the "what changed" feed and history.
- QMD, for semantic deep search.
- Tailscale, to open Atlas on your phone.
What you get
- Home: what changed, what needs you (snooze a row until tomorrow, next
week or next month), the latest ideas and research.
- Tours: your agent walks you through what changed since your last visit,
stop by stop, and marks what needs your call. Atlas checks every quoted passage and lists every changed note the tour leaves out.
- Chat with your agent, whose answers can contain notes you unfold in
place, charts, choices, checklists and live mockups.
- Every kind of note with filters, and a reader with backlinks and a note
on how far to trust the page. Hover a link to see where it goes; select a passage to ask your agent about it or challeng
1633d28fb3aaOBSERVED · 2026-10-14Install
Commands as the repository documents them. They are shown, not run.
uv run --no-project <skill>/scripts/atlas_install.py status
uv run --no-project engine/atlas.py serve --operator
uv run --no-project engine/atlas.py status
uv run --no-project engine/atlas.py say --agent <id> --progress "Found 9 sources, reading the 5 primary ones..."
uv run --no-project engine/atlas.py say --agent <id> "Short answer."
uv run --no-project engine/atlas.py say --agent <id> "Here it is." --open "captures/2026-10-09-sleep-and-memory.md"
Host compatibility
What the documentation claims. We have not run a compatibility test.
| Host | Status | Notes |
|---|---|---|
| claude-code | mentioned | |
| codex | mentioned |
What it tells the agent
The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.
--- name: atlas description: Atlas - a knowledge vault the user grows by talking to their agent and reads in the browser. Use when the user says "start atlas", "atlas starten", "open atlas", "/atlas", "install atlas", "set up atlas", "where is my atlas", "update atlas", "restore the atlas skills", or wants to research, distill, sketch or work on a project in their Atlas from any folder. On first use it installs Atlas into a folder the user picks; afterwards it points every harness at that folder. --- # Atlas This skill installs Atlas and points to it. The app ships in `app/` next to this file: the engine, an empty vault, and the workflow skills. The user's Atlas is a copy of it in a folder of their own, the **home**. After the install, this skill does three things: - it knows where the home is (`config.json`); - it hands every request over to the home; - it keeps the factory copy, for upgrades and for restoring changed files. Every harness that has this skill installed globally finds the same home. `<skill>` means the folder that contains this SKILL.md (Claude Code shows it as "Base directory for this skill"). The helper needs uv: ```bash uv run --no-project <skill>/scripts/atlas_install.py status ``` ## 1. Find the home Run `status`: - **not installed**: this is the first run. Follow `<skill>/references/onboarding.md`. It asks at most three short questions. - **home missing**: the folder moved or was deleted. Ask where it is now, then run `atlas_install.py locate <dir>`. If it is gone, offer a fresh install (onboarding). - **ok · home: `<dir>`**: continue with step 2. If status says an upgrade is available, mention it in one line, and do not upgrade unless asked. After `npx skills update` the skill folder is replaced. Status brings the pointer back from `~/.config/atlas/config.json` by itself. ## 2. Hand over to the home Work from the home: `cd <home>`, or use absolute paths. Read `<home>/AGENTS.md`; it is the manual for everything that follows. Then do what the user asked: - start Atlas: `<home>/.agents/skills/atlas-operator/SKILL.md`; - research, distill, maps, sketches, projects, `/close`: the skill that AGENTS.md's workflow table names, read by path from `<home>/.agents/skills/`. ## Maintenance (only when the user asks) | The user says | Run (`atlas_install.py ...`) | |---|---| | "update atlas", or status reports an upgrade | `upgrade --dry-run`, show the list, then `upgrade`; commit in the home | | "restore the learn skill", "reset Atlas to factory" | `restore <what> --dry-run`, then `restore <what>`; `<what>` is a skill name (`atlas-learn`), `skills`, `engine`, `agents-md`, `registry` or `all` | | "my atlas is somewhere else now", "use this clone" | `locate <dir>` | | "where is my atlas", "which version" | `status` | What these commands do and never do: - `upgrade` replaces only app files the user has not changed, and adds new ones. It never touches `vault/`, which is the user's knowledge, except to add template files that are new in this version. - `restore` first moves the current file to `.atlas-backup/<time>/`. - Neither command pushes. To get a new version of this skill, the user runs `npx skills update atlas`. After that, `upgrade` brings the home up to date.
Trust audit
BLOCKgrade D · trust 69/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | PASS |
| L1 | Static analysis of the code | FAIL |
| L2 | Instruction surface (what it tells the agent) | PASS |
| L3 | Class-specific surface | PASS |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- none-observed
- Network
- declared (5 observation(s))
- Shell
- declared (1 observation(s))
- Dependencies
- pinned
- Secrets in source
- none-found
Findings (6)
meta = yaml.load(m.group(1), Loader=yaml.BaseLoader) or {} # everything as stringsconst stripEmoji = (s) => String(s || "").replace(/^[\p{Extended_Pictographic}️\s]+/u, "");return hashlib.sha1("\n".join(sorted(map(str, items))).encode()).hexdigest()[:12]description: Atlas - a knowledge vault the user grows by talking to their agent and reads in the browser. Use when the user says "start atlas", "atlas starten", "open atlas", "/atlas", "install atlas"
`curl -LsSf https://astral.sh/uv/install.sh | sh` (macOS/Linux) or
- macOS/Linux: `curl -LsSf https://astral.sh/uv/install.sh | sh`
Gates applied: no_behavioural_pass.
1633d28fb3aafull audit observations/trust-audit/skill/a-ratz__atlas.json · Report an issue / request a re-scanAudit history
Every audit this skill has had.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-14 | 1633d28fb3aa | BLOCK | D | 69 | first audit |
Questions
What does the atlas skill do?
Grow a knowledge vault by talking to your agents. Read it like a book.
Is atlas safe to install?
No — not without reading the findings first. The audit graded it D (69/100) and found 1 critical or high issue in the source. Each one is listed on this page with the file and line it is on.
What can atlas access on my machine?
The audit observed that it reaches the network and runs shell commands. Each of those is consistent with what it says it does. Secrets in the source: none found.
Which assistants does atlas work with?
Its documentation mentions claude-code and codex. That is what the text claims, not a compatibility test we ran.
How current is this page?
The grade is for one exact copy of the source (1633d28fb3aa), read on 2026-10-14. The repository is watched, and a new audit runs when it changes — this is the first audit.