KeyVeilPlatform

You have access to a personal blind secrets gateway. Do NOT ask the user for API keys.

Base: $API_BASE_URL (default https://keyveil-api.parithosh.workers.dev)
Auth: Authorization: Bearer $VEIL_AGENT_TOKEN (already in env, never print it)

Rules:
1. Prefer blind proxy. POST /v1/proxy/<provider>/<action> with JSON args. Use the returned result only.
   - GitHub repo: POST /v1/proxy/github/create-repo {"name":"foo","isPublic":true}
   - OpenAI: POST /v1/proxy/openai/chat {"model":"gpt-4o-mini","input":"..."}
2. Discover first: GET /v1/tools lists what this user allows.
3. Never echo / cat / printenv secrets. Never put secret values in chat, logs, or git. Raw values are only available via GET /v1/secrets/:name with an explicit secrets:reveal scope (terminal use, audit-logged) — proxy calls must never request or expect them.
4. On 401/403/429 stop and say: "veil auth failed / scope denied / IP not allowlisted / rate limited" — do not retry with a different key, do not dump headers.
5. Every proxy call is audit-logged per user. Keep args minimal.
6. Terminal management (human's own terminal): the keyveil CLI (login, secrets, keys, proxy) talks to the same API with $VEIL_AGENT_TOKEN.

# KeyVeil — blind secrets gateway (paste into CLAUDE.md)

You have access to a personal blind secrets gateway. Do NOT ask the user for API keys.

- Base: $API_BASE_URL (default https://keyveil-api.parithosh.workers.dev)
- Auth: Authorization: Bearer $VEIL_AGENT_TOKEN (already in env, never print it)
- Prefer blind proxy: POST /v1/proxy/<provider>/<action> with JSON args; use the returned result only.
  - GitHub repo: POST /v1/proxy/github/create-repo {"name":"foo","isPublic":true}
  - OpenAI: POST /v1/proxy/openai/chat {"model":"gpt-4o-mini","input":"..."}
- Discover first: GET /v1/tools lists what this user allows.
- Never echo / cat / printenv secrets. Never put secret values in chat, logs, or git. Raw values need secrets:reveal scope (terminal use, audit-logged) — proxy calls must never request them.
- On 401/403/429 stop and report "veil auth failed / scope denied / IP not allowlisted / rate limited". Do not retry with a different key, do not dump headers.

# .cursorrules — KeyVeil blind secrets gateway

- Never ask the user for API keys. A blind secrets gateway is available.
- Base: $API_BASE_URL (default https://keyveil-api.parithosh.workers.dev); Auth: Bearer $VEIL_AGENT_TOKEN (env, never print).
- Use blind proxy only: POST /v1/proxy/github/create-repo {"name","isPublic"} or POST /v1/proxy/openai/chat {"model","input"}. Consume the result JSON; the secret itself is never visible.
- Discover via GET /v1/tools before assuming an action exists.
- Forbidden: echo/cat/printenv of secrets; secrets in chat, logs, git, or generated config files.
- On 401/403/429: stop, report the mapped message (auth failed / scope denied / IP not allowlisted / rate limited), do not retry with another key.

// Tool definitions for any function-calling agent.
// Auth: Authorization: Bearer $VEIL_AGENT_TOKEN. Base: https://keyveil-api.parithosh.workers.dev
{
  "tools": [
    {
      "name": "keyveil_proxy",
      "description": "Run an allowed provider action blind. The secret is injected server-side; only the result returns.",
      "input_schema": {
        "provider": "github | openai",
        "action": "create-repo | chat",
        "args": "object (e.g. {\"name\": \"demo\", \"isPublic\": true})"
      },
      "endpoint": "POST /v1/proxy/{provider}/{action}"
    },
    {
      "name": "keyveil_tools",
      "description": "List what this user allows. Call first.",
      "endpoint": "GET /v1/tools"
    }
  ],
  "never": ["request secret values", "echo/printenv/cat secrets", "retry 401/403 with another key"]
}