Secrets your agent never sees.
When a coding agent needs API keys, agent-secret-manager opens a localhost form. You paste any values you have, skip the rest, and the tool writes only submitted values to .env. The agent gets names and presence state, never the secret contents.
$ npx agent-secret-manager request OPENAI_API_KEY \ --reason "Run the local OpenAI example"
Why it exists
Built for the way agents actually work.
Coding agents tend to pull secrets into chat, terminal output, screenshots, and logs. agent-secret-manager keeps the value in a local browser form and records only metadata for the agent.
Not in chat
The agent never receives the value, so it cannot echo it back into the conversation.
Not in your terminal
Standard output stays clean. The CLI prints names and saved, skipped, or present state.
Not in screenshots
Inputs are masked by default. Captures of the agent session do not reveal the secret.
Quick start
Three commands. No config required.
Your project gains a local .env file, a blank .env.example, and metadata that never contains values.
Request a secret
The CLI opens a localhost form. The user can submit some values and skip the rest.
Verify presence
Agents call this to confirm a value exists without reading it.
Run with the env loaded
Inject the env file into a child process when the project does not load .env itself.
Agent skill
Bundled instructions for agent workflows.
Install the skill so coding agents request missing secrets through this CLI, verify only presence, and avoid reading or printing .env.
Codex / Claude Code skill
The package ships an agent skill with the expected request, check, and run flow.
Security model
Specific protection, honest boundaries.
This tool prevents routine exposure. It is not a sandbox against local processes that can read files.
Stop routine secret exposure in chat, terminal output, screenshots, shell history, and agent transcripts.
Store submitted values in a local .env file with private permissions.
Sandbox a malicious local process. Anything with read access to your filesystem can read .env.
Stop an agent that is explicitly instructed by a user to read secret files. The skill instructs agents not to.