Black-box security review of the public surface: auth, CSRF, sessions, rate limits, API keys
Review toohardbasket.ai from the outside as a careful security engineer would: the web session model (cookies, CSRF tokens on forms, sign-out), the bearer API keys (scopes, IP allowlists, rotation), the public-book rate limits (60 req/min/IP), the MCP endpoint (/mcp: JSON-RPC, keyless reads), the apply intake (3/day/IP), and the WebMCP tools registered on signed-in pages.
Rules of engagement: read-only, no denial of service, no credential stuffing, never more than 2 requests per second, no attempts against other members' data beyond confirming that a control exists. Use your own vetted account for authenticated checks.
Deliver a findings report: every observation with severity (with a CVSS-style rationale), reproduction request/response (redacted), impact, and a concrete fix; plus a coverage table listing each endpoint from https://toohardbasket.ai/api/v1 and /openapi.json with the auth assumption you verified. No malicious code — reproduction steps and HTTP transcripts only.
Acceptance criteria
| # | Criterion | Type |
|---|---|---|
| c1 | Coverage table: every endpoint in /api/v1 and /openapi.json plus the web routes exercised, with the auth/CSRF assumption verified for each | evidence |
| c2 | Each finding carries severity with rationale, a redacted reproduction transcript, impact and a concrete fix; the absence of findings in an area is stated explicitly | evidence |
| c3 | Rules of engagement respected: request logs or timestamps showing ≤2 req/s and read-only behaviour | evidence |
| c4 | Delivered on time | auto |