SafeClaw Docs
A systematic tour of SafeClaw, one module per page. Concepts explain how the
product thinks; guides get a task done; security answers the questions that
decide trust. reference/ holds the lookup layer; contributor design docs
live outside this tree in design/.
Pages are deliberately short and will deepen over time.
New here? Read Overview, then do the Quickstart.
Concepts
| Page | The module |
|---|---|
| Overview | The whole model on one page: phantoms, the broker, passkey approvals |
| Vault | Where credentials live: sealed under your passkey, synced blind |
| Connections | The catalog, host anchors, and how a secret becomes a phantom |
| Broker | sc run: one local egress point that swaps phantoms for values |
| Approvals & policy | Who decides what runs: levels, rules, single-use grants |
| Agents | Agent identities: attribution without possession |
Guides
| Page | The task |
|---|---|
| Quickstart | Zero to first brokered call in five minutes |
sc run and phantoms | The patterns, and the sc get anti-pattern |
| For your agent | Handing SafeClaw to an agent: skill file, paste-ready prompt |
Security
| Page | The question |
|---|---|
| Design principles | The rules the whole system is built on |
| Security model | Where keys live, what the cloud sees, what compromise costs |
Reference
For users and service authors, in reference/:
services.md (writing a service.toml),
policy.md (per-action decisions),
diagnostics.md (every error code, with fixes),
consent-templates.md (approval-copy grammar).
Design docs
For contributors reading the source, outside this tree in
design/:
protocol.md (SUDP cryptographic profile),
credential-broker.md (the broker architecture),
connection-schema.md (vault data shapes),
sync.md / sse-sync.md (cloud sync),
request-scope.md (approval scope binding),
stores-and-items.md (stores and vault content model).