Hunt API Agent Skill

Hunt: API Security

Assumes hunt-core for the scope gate, two-account rule, confirmation gate, enumeration limits, stop conditions, wiki protocol, FIND output, and Deadends. Do not re-derive any of that here.

Wiki

qmd_query "API REST GraphQL gRPC BOLA BFLA mass assignment excessive data exposure OWASP API Top 10" via wiki-search MCP

Hub: [[web-moc]] (live web index). Primary page: [[api-security]]. Payload arsenal: wiki/payloads/api.md.

Anchors: [[api-testing]], [[graphql-attacks]].

Variants: [[grpc-web-attacks]] (gRPC-Web / protobuf transcoder abuse), [[rsql-injection]] with the [[rsql]] payload (RSQL/FIQL filter-query injection, e.g. Spring Data REST), [[rate-limit-bypass]] (header/race/distributed-source throttle bypass), [[redos]] payload (catastrophic-backtracking regex DoS in an input validator), [[jwt-attacks]] (token flaws).

For object-level authorization (BOLA/IDOR) see hunt-idor.

Attack surface signals

/api/, /v1/, /graphql, /rest/, gRPC (application/grpc, HTTP/2), Swagger UI (/swagger, /api-docs, /openapi.json), mobile/SPA backends.

Rank before testing. Not all surface is equally likely to be broken:

• Undocumented endpoints present in the spec (Swagger/introspection) but never called by the UI - the classic BFLA surface; nobody tested the route the client does not exercise.
• GraphQL introspection and the mutations/fields it reveals that the client never invokes.
• Batching and alias endpoints - one request, many operations; per-item authorization and rate limiting are frequently applied to the request, not each operation.
• Bulk and export endpoints - one call, many objects; much higher severity per finding.
• Non-GET verbs on read-looking routes - GET /orders/123 authorized, PATCH/DELETE not.
• Newest features - authorization middleware lags new code; a route written outside the conventions misses it.

Methodology

Setup: two accounts per hunt-core (A owns, B attacks, separate profiles). Get the spec if any - Swagger/OpenAPI, GraphQL introspection, .proto.

Drive it through Burp for operator visibility. Push the load-bearing requests (the BOLA cross-account swap, the mass-assignment body, the BFLA verb/route call) into Repeater via Skill(hunt-burp) / the native Burp MCP (mcp__burp__*) so the operator can replay and inspect them; brute/fuzz belongs in Intruder (send_to_intruder), not a hand-rolled loop. A quick throwaway curl per account for the writeup PoC is fine.

• Enumerate. Parse Swagger/OpenAPI for every endpoint + param; GraphQL introspection (__schema); gRPC server reflection (grpcurl -plaintext <h> list, then list <svc> / describe). No spec? Discover endpoints with ffuf -w <api-wordlist> -u https://HOST/FUZZ and fingerprint hosts with httpx, not a hand curl loop. This is service / endpoint DISCOVERY, not object enumeration - it is bounded by the engagement RoE (no_dos, scan-rate caps), NOT by the 5-to-20 object cap. Do not clamp the wordlist to 20.
• BOLA / IDOR (API #1). Swap object IDs across accounts (numeric, UUID, in body/path/header) - the dominant API bug. Swapping IDs to pull other accounts' records IS object enumeration: bounded to 5 identifiers by default, 20 ceiling with operator approval, 0 under no_bruteforce, per hunt-core. Two or three adjacent IDs prove the sequential pattern; cite a total/pagination count for scale, never a sweep. Full bounded-sample loop and the trusted-identifier test live in hunt-idor. -> overlaps hunt-idor.
• Broken function-level auth (BFLA). Call admin/privileged methods as a low-priv user; swap the HTTP verb (GET -> PUT/DELETE); hit the undocumented endpoints from the spec.
• Mass assignment. Add fields the client never sends (role, isAdmin, verified, balance) to JSON bodies; look for the privilege/state change to actually take effect (confirm from a fresh read, step below).
• Excessive data exposure. The API returns more than the UI shows (full objects, other users' fields, internal flags) - inspect the raw response, not the rendered page.
• Auth. JWT flaws ([[jwt-attacks]]: alg:none, weak secret, kid injection), API-key reuse, missing auth on some routes, OAuth scope creep.
• GraphQL specifics. Introspection, batching/aliases (rate-limit/brute bypass), nested-query DoS, field suggestion leaks. Payloads: [[graphql-attacks]].
• gRPC specifics. Reflection to enumerate; grpcurl to call methods; tamper protobuf fields; run the same authz tests as REST.
• Rate limit / resource. Unbounded pagination, no throttle on sensitive actions (OTP/login). Prove the throttle is absent with a bounded burst; honor no_bruteforce/no_dos - do not actually brute credentials or OTPs and never run this at volume against a live auth endpoint. See [[rate-limit-bypass]].

When a direct call 403s or the extra field is stripped, it is not closed. Parameter pollution (?id=A&id=B - check and fetch may read different occurrences), array/nested wrapping of the field ({"user":{"role":"admin"}}), casing and separator variants of a mass-assign key (isAdmin/is_admin/admin), alternate content-type (form vs JSON vs XML), verb-override headers (X-HTTP-Method-Override), version downgrade (/v1/ predates the middleware /v2/ has - the most reliable), and batching to skip per-item authorization.

Chain: mass assignment -> privilege escalation. A role:admin/isAdmin:true that actually takes effect turns the whole BFLA admin surface reachable - re-run step 3 as the escalated principal. For the broader workflow/state-tampering angle hand off to hunt-bizlogic.

Distill a confirmed, GENERIC pattern (product + endpoint + impact, no client host): python3 scripts/wiki-stage.py --kind technique --slug <slug> --target-page techniques/web/api-security.md

Confirmation gate

NOT confirmation: a 200 echoing your own request back; an empty or shell response; B seeing an object that is shared, public, or org-visible; a body you have not compared against A's baseline; a BFLA endpoint returning 200 without the privileged action actually performed; a mass-assignment write returning 200 with the extra field accepted but the privileged field never verified in effect; any result not re-verified in a clean session.

IS confirmation:

• BOLA - B's low-priv session returns A's actual *data*, matching A's own baseline response, with A's legitimate/shared access ruled out, reproduced in a clean session.
• BFLA - a low-priv token reaching AND executing an admin/privileged function (perform the action, do not just get a 200).
• Mass assignment - the privileged field (role/isAdmin/balance) verified in effect from a fresh authenticated read, not merely accepted in the request body.

Severity

• CRITICAL - unauthenticated admin action, or cross-tenant data access.
• HIGH - BOLA/BFLA reaching other users' data or functions; mass-assignment privilege escalation.
• MEDIUM - excessive data exposure; missing or weak rate limiting.

Deadends

Append: - [ ] API on <host> <endpoint> -- BOLA/BFLA/mass-assign all enforced;
              extra fields ignored; JWT validated; introspection off

Record what you tried (pollution/array/verb/version-downgrade/batch), not just that it failed.