{
  "data": null,
  "kind": "anchor.page",
  "links": {
    "api": "https://www.anchorterminal.com/api/v1/index.json",
    "html": "https://www.anchorterminal.com/niches/secrets.keep-out-of-context/",
    "json": "https://www.anchorterminal.com/niches/secrets.keep-out-of-context/index.json",
    "llms": "https://www.anchorterminal.com/llms.txt",
    "markdown": "https://www.anchorterminal.com/niches/secrets.keep-out-of-context/index.md",
    "mcp": "https://www.anchorterminal.com/mcp",
    "slim": "https://www.anchorterminal.com/niches/secrets.keep-out-of-context/index.min.md"
  },
  "markdown": "One company's docs show a full fit.\n\n| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |\n| --- | --- | --- | --- | --- |\n| 1 | 9 | 14 | 24 | 8 |\n\n## What counts as done\n\nCommands started by the agent get the secrets as environment variables or files.\n\nMeasured, for our own tests: no secret value appears in the agent's transcript, tool output or files it writes across a run that exercises env, cat and an error path.\n\nHow agents ask for it: stop Claude reading my .env; run the tests with real keys but don't show them to the agent; inject secrets for this command\n\n## Does the job\n\nThe docs show it meets the test above, and nothing we read says otherwise.\n\n- **Phase (Phi Security Inc.)**: In AI mode phase run injects secrets, masks secret values, never shows sealed ones and blocks printenv, env and export inside the run. A child script echoing a non-sealed value is not listed as covered.\n\n## Does the job, with a limit\n\nIt does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.\n\n- **1Password service accounts, SDKs and Environments MCP (AgileBits Inc. (doing business as 1Password))**: op run injects references as environment variables and conceals values on stdout and stderr by default, but files a child writes are not covered and an agent holding the token can run op read or --no-masking itself; acc-02 calls it no proof against prying.\n- **Akeyless (SecretlessAI and MCP server) (Akeyless Security Ltd.)**: No run command that puts secrets in a child's environment and no output masking were found; Akeyless Agent writes templates to files the agent could read, and akeyless mcp's get_secret returns values to the model. mcp-runtime-authority avoids values but is a broker, not injection.\n- **AWS Secrets Manager (Amazon Web Services, Inc. (Amazon Web Services EMEA SARL and other regional entities by account location))**: asm-exec resolves {{resolve:secretsmanager:...}} references into a child process and a PreToolUse hook blocks get-secret-value, but AWS calls it best-effort, not a security boundary, and nothing masks a value the child prints.\n- **Bitwarden Secrets Manager (Bitwarden Inc. (terms name 8bit Solutions LLC, wholly owned by Bitwarden Inc.))**: bws run injects secrets as environment variables named after the secrets, but the docs describe no output masking and say the run is not a sandbox, so an agent running env under it sees values.\n- **Doppler (Doppler Technologies, Inc.)**: doppler run injects secrets as environment variables, but no output masking is documented, it writes an encrypted fallback snapshot to disk by default, and the MCP server returns values with no masking. acc-26 dropped Doppler env vars for a broker.\n- **HashiCorp Vault + Vault MCP Server (HashiCorp, Inc. (an IBM company))**: Vault Agent's exec block injects secrets as environment variables into one child process, but nothing masks output, and the beta MCP server warns it may expose secrets to the model.\n- **Infisical (Infisical, Inc.)**: secrets-manager run injects secrets as environment variables but documents no masking, and the MCP server returns values unless INFISICAL_MASK_SECRET_VALUES=true; acc-11 saw a bot print Cloudflare tokens. Agent Vault avoids this by never giving values.\n- **Keeper Secrets Manager (Keeper Security, Inc.)**: ksm exec swaps keeper:// variables for values in the child's environment, with no output masking; the docs warn anything that can read /proc/\u003cpid\u003e/environ sees them. The MCP server masks values by default and asks before unmasking.\n- **Pulumi ESC (Pulumi Corporation)**: pulumi env run injects values and filters secret values from the command's output streams unless -i is passed, but files the child writes are not covered and an agent with open permission can run pulumi env open itself.\n\n- not this job: Azure Key Vault (Microsoft Corporation): No run command injects secrets into a child process, nothing masks output, and the Azure MCP server's secret get returns the value to the model.\n- not this job: Google Cloud Secret Manager (Google LLC (Google Cloud EMEA Limited and other regional entities by billing address)): Secret Manager has no run wrapper or output masking; gcloud secrets versions access prints the value to whoever calls it.\n\n## What stops agents finishing\n\n- the agent can run env or its own script under the wrapper and print values\n- MCP server returns secret values unless masking is switched on\n- wrapper hangs on an interactive login when its session expires\n\n## Around it\n\nBefore: code.open-repo. After: code-sandboxes.run-tests, code.commit.\n\n## Where this comes from\n\nWeight 24 of 100 in Secrets \u0026 credential vaults: 8 of 27 accounts (op run, harness masking, Infisical in Goose, keychain, placeholders, secrets outside .env)\n\nAs data: https://www.anchorterminal.com/api/v1/use-cases/secrets.keep-out-of-context.json, or the MCP tool get_use_case with id secrets.keep-out-of-context. Draft, read 2026-11-15 from public docs and first-hand accounts, not yet reviewed by a person.\n",
  "meta": {
    "attribution": "Anchor Terminal (https://www.anchorterminal.com)",
    "attributionRequired": true,
    "credit": "CC BY 4.0, credit Anchor Terminal and link the page or API address you took it from (https://www.anchorterminal.com/terms/#data)",
    "docs": "https://www.anchorterminal.com/docs/",
    "generatedAt": "2026-11-16",
    "license": "CC-BY-4.0",
    "method": "https://www.anchorterminal.com/benchmark/",
    "methodology": "0.8",
    "openapi": "https://www.anchorterminal.com/openapi.json",
    "preview": false,
    "run": "2026-10-01",
    "runLabel": "October 2026 research run",
    "terms": "https://www.anchorterminal.com/terms/#data"
  },
  "page": {
    "breadcrumbs": [
      {
        "name": "Home",
        "url": "https://www.anchorterminal.com/"
      },
      {
        "name": "For agents",
        "url": "https://www.anchorterminal.com/agents/"
      },
      {
        "name": "Niches",
        "url": "https://www.anchorterminal.com/niches/"
      },
      {
        "name": "Let a coding agent's commands use secrets without the values entering its context",
        "url": ""
      }
    ],
    "description": "Let a coding agent's commands use secrets without the values entering its context: 1 companies fully do it, 9 with a limit, 14 carry the tag. Every listing in Secrets \u0026 credential vaults with the reason and a source.",
    "facts": null,
    "h1": "Let a coding agent's commands use secrets without the values entering its context",
    "image": "https://www.anchorterminal.com/assets/og/default.png",
    "path": "/niches/secrets.keep-out-of-context/",
    "published": "",
    "section": "agents",
    "title": "Let a coding agent's commands use secrets without the values entering…",
    "toc": null,
    "updated": "2026-11-15",
    "url": "https://www.anchorterminal.com/niches/secrets.keep-out-of-context/"
  },
  "tokens": {
    "markdown": 1450,
    "slim": 330
  },
  "version": 1
}
