▸ Agent Skills
2 min read

Python SDK

The Python SDK (secretspec) is a thin client over a pyo3 extension that calls secretspec::resolve_json directly. Resolution (providers, chains, profiles, generation, as_path) happens in the Rust core, so the SDK inherits every provider with no Python-side logic.

Quick start

from secretspec import SecretSpec

resolved = (
    SecretSpec.builder()
    .with_provider("keyring://")
    .with_profile("production")
    .with_reason("boot web app")
    .load()
)

print(resolved.provider, resolved.profile)
db = resolved.secrets["DATABASE_URL"]
print(db.get)              # the value, or the file path for as_path secrets
resolved.set_as_env()      # export everything into os.environ

A missing required secret raises MissingRequiredError; any other failure raises SecretSpecError (with a stable .kind).

Caller context

from secretspec import CallerContext, SecretSpec
builder = SecretSpec.builder().with_caller(CallerContext(    name="git",    version="2.51.0",    operation="credential_get",    resource="github.com",))

Caller context identifies the invoking integration in audit records but never satisfies require_reason. Do not put credentials or secret values in it.

Inline specifications

Use .with_inline_spec(spec, base_dir) to resolve a strict inline-spec v1 dictionary without a manifest file. base_dir resolves relative provider paths. An inline call uses the versioned native entry point, so it cannot fall back to a filesystem manifest on an older runtime.

Scopes

Use .with_scope("api") to resolve only a named [scopes.api] subset. The selected name is available as resolved.scope and report.scope:

resolved = SecretSpec.builder().with_scope("api").load()

Typed access (codegen)

Generate typed classes with secretspec schema plus quicktype, then build them from resolved.fields():

$ secretspec schema | quicktype -s schema --top-level SecretSpec --lang python -o secrets_gen.py

Terminal window

from secrets_gen import SecretSpec as Secrets  # typed

typed = Secrets.from_dict(resolved.fields())
print(typed.database_url)  # typed str

Native library

The resolver is statically linked into a pyo3 extension (secretspec._native, built from the secretspec-py-native crate) using pyo3’s abi3-py39 feature, so the published cp39-abi3 wheel is self-contained — there is no separate cdylib to locate and no runtime dlopen.

Last updated Oct 08, 2026