The credential store is 0600 under a 0700 directory, which stops other OS users but not the model: tool processes run as the same user, so under the shipped danger-full-access default they read it like any other file. SandboxExecutionPolicy grows readDenyPaths, and sandbox-policy defaults it to $DSH_HOME/.env — the exact file rather than the harness home, so the model keeps its documented access to its own session log. Seatbelt appends a trailing deny (last matching rule wins) and bwrap maps /dev/null over each path after any workspace bind; Landlock grants are a pure allow-list that cannot subtract from its own / read grant, so confine() reports partial enforcement there instead of claiming a boundary the process does not have. A real-kernel Seatbelt e2e proves the shape: the same read succeeds unconfined and fails under the denial, while a sibling file in the same directory stays readable. Both READMEs state the residual boundary plainly — no confining mode means no boundary — and record the OS keychain provider as the real answer.
credentials/
English | 中文
The credential capability seam, as three-package shape dictates (interface / implementation / consumers):
| Package | Role |
|---|---|
credentials/ |
Abstract ctx.credentials: branded CredentialRef references, per-operation resolve, UI-safe describe, fail-loud set/unset, the credentials/updated commit event |
credentials-local/ |
File/environment provider: the live process environment (read-only, wins) layered over $DSH_HOME/.env (writable, byte-preserving line edits, hot-reloaded) |
Configuration files carry references to secrets (apiKeyEnv: DEEPSEEK_API_KEY), never the secrets: the settings document stays safe to sync and render, and rotating a value touches no configuration. The LLM adapters are the first consumers — they resolve their reference once per model request, which is what makes a key stored moments ago reach the very next request without restarting anything.
The seam shape leaves room for keyring-, helper-command-, and KMS-backed providers.