resenv resolves values in a process environment. It does not provide a vault, access-control service, key rotation, or isolation between programs running with the same credentials. The application can read its resolved secrets and any credentials supplied to it.

Doppler mode uses a fixed HTTPS API origin, disables redirects and fetches only referenced secret names. It does not execute shell expressions. Credentials come from an explicit token, resenv's OS credential store or a fixed Doppler CLI helper. An API rejection never triggers a switch to another identity. Native browser login uses endpoints observed in Doppler's official MCP client rather than a public third-party OAuth contract. See Doppler details for fallback rules, credential scope and platform limits. The canonical resolve preload and run command select Doppler only for doppler:// values. They do not infer shell execution from $() values. Startup does not initiate login; the separate login command rejects CI and noninteractive sessions.

The local credential store does not prevent code running as the same user from requesting credentials. resenv does not automatically remove an explicitly supplied DOPPLER_TOKEN from the application environment. Interactive login displays a temporary human authorization code; tokens and the polling code are not logged. The Node preload's private worker exchanges credentials and results over captured pipes, never process arguments or dotenv files.

Command mode executes trusted configuration with the current user's permissions. Enable it only for configuration you have reviewed. Commands can access files and the network; failed resolution cannot undo their external effects. Commands are bounded by timeout and output limits, and their output is not included in resenv diagnostics.

Node's child_process debug category prints spawn options, including stdin and environment values. resenv refuses its own subprocess and credential-helper operations when that category is enabled, including through NODE_DEBUG=*. Use NODE_DEBUG=ciqol:resenv for diagnostics that omit values. This guard does not control application logs, other instrumentation, or a debugger attached to the process.

All values are resolved before resenv updates the caller's environment or starts the application. Provider failures, command failures and invalid environment values stop startup. Fetched values are never evaluated recursively. Application logs are outside resenv's control.

Different automatic preload modes cannot be combined in one process. Otherwise the second mode could execute literal data fetched by the first. Combining a preload with the CLI is rejected for the same reason. Do not add another resolver to the child through NODE_OPTIONS or other launch hooks. Use one resenv run --commands invocation or one library call with commands enabled to resolve cloud references and commands together. Repeated manual library calls are separate operations; only pass original configuration to them.

Provider responses must be valid UTF-8 and JSON, and resolved strings must not contain lone Unicode surrogates or null bytes. In-place replacement preflights all changed properties; read-only properties, setters and Proxy targets are rejected before any values are committed. Use resolve() to obtain a copy when the input is not a writable environment record.

This experimental package has not had an independent security audit. Validation evidence and platform limits are recorded in docs/validation.md.

Do not open a public issue containing a vulnerability, credentials or sensitive output. Open the repository's Security tab and use "Report a vulnerability" when available. If that option is unavailable, contact the maintainer through an existing private channel. Before the public transition is complete, maintainers must enable GitHub private vulnerability reporting and verify that it is available.

Include affected versions, platform, impact and a minimal reproduction using dummy data. Never include production tokens, private keys or environment dumps. There is no guaranteed response time or long-term support policy yet.