API Security Boundary Assessment
Read the assessment methodology before using this skill. Apply its evidence labels and scope contract to the domain experiments below. This revision restores the upstream's API-security craft as scoped experiments mapped to the OWASP API Top 10, confirming each class on owned accounts with bounded request sets rather than enumerating real users.
Scope and prerequisites
Collect the API surface (specs, Swagger, gRPC reflection, WebSocket endpoints). Use owned accounts across roles and a disposable instance. Bound enumeration and rate-limit probes.
Declare target, protocols, attempt budget, stop condition. Spec-only authorization yields a hypothesis marked NOT TESTED.
Attack surface and classes (OWASP API Top 10 2023)
- BOLA/IDOR — object-level authorization (see
rt-idor). - Broken auth — missing login rate limits, token lifecycle (expired, post-password-change, malformed bearer accepted).
- Excessive data exposure — endpoints returning more than the UI shows.
- BFLA — function-level: admin methods reachable with a user token.
- Mass assignment — unexpected writable properties.
- Verb/content-type tampering — method override, content-type switch, parameter pollution (see
rt-parameter-pollution). - SSRF via parameters (see
rt-ssrf). - gRPC — reflection enumeration, metadata injection, BOLA/BFLA over gRPC.
- WebSocket — origin validation, CSWSH.
- Misconfiguration — version sprawl, exposed docs/debug, CORS, missing security headers.
Mechanism and observation
Trace a request to the authorization/validation performed and the data returned.
Firstness: Record the access or data obtained outside the account's intended scope (another owned object; an admin method succeeding with a user token; fields beyond the UI). Secondness: State the intended policy and the deviation. Thirdness: Infer the class only after a controlled request confirms it on owned accounts.
Abduction and refutation
Mechanism hypothesis: Authorization or validation is enforced inconsistently across objects, functions, verbs, protocols or versions, so one path grants access another denies.
Competing benign explanation: Every path enforces the same checks; the admin method returns 403; the extra fields are not actually returned/accepted; an old API version is deprecated and gated; CORS reflects only allowlisted origins.
Deduction and controls
A properly-scoped request succeeds (positive control); the out-of-scope request must be refused if controls hold. A confirmed finding obtains the out-of-scope access/data on owned accounts where the control refuses. Predeclare the confirming artifact. Consistent enforcement across paths is a PASS control.
Induction: bounded laboratory procedure
Owned accounts across roles; bounded sets.
- Baseline. Authenticate per role; capture legitimate access and controls.
- BOLA/BFLA. Access another owned object (user token); call an admin method with a user token.
- Token lifecycle. Replay an expired/post-password-change token; send a malformed bearer.
- Mass assignment / verb / content-type. Inject non-user fields; tamper methods and content types.
- gRPC/WebSocket. Enumerate via reflection; test metadata injection and BOLA; test WS origin/CSWSH from an owned page.
- Misconfig. Enumerate versions; check exposed docs/debug, CORS, headers.
- Control. Confirm scoped requests behave and the out-of-scope path is refused.
Record results including refutations. Keep finding and control status separate.
Craft notes (leads)
Tooling: grpcurl (reflection), mitmproxy gRPC addon (--mode reverse), websocat (WS/origin), Burp/Arjun (discovery). CSWSH applies when the WS handshake trusts cookies without origin checks. Old API versions often lack newer authz fixes. Method-override headers can bypass method-based WAF rules.
Domain-specific failure modes
A 200 is not access — confirm the body is out-of-scope data (see rt-idor). A deprecated version that is gated is a refutation. CORS reflecting only allowlisted origins is a PASS. Excessive-exposure requires the field to actually be returned, not merely present in a schema.
Evidence and deliverable
Provide the endpoint/protocol, the account/role context, the out-of-scope access/data obtained, and the control refusal. Use owned accounts; keep real data out of portable reports. Attach minimal replay, expected/observed, rival explanation, falsifier, uncovered endpoints.
Versions and tool limitations
Pin the API framework, gateway and gRPC/WS stack versions. Tools enumerate aggressively; bound them. No endpoint behavior was verified by this content; confirm the live API first.
Remediation and detection
Enforce object- and function-level authorization uniformly across REST/gRPC/WS and versions, validate token lifecycle, allowlist writable fields, restrict verbs/content-types, validate URL parameters (SSRF), enforce WS origin, and deprecate old versions. Add regressions per class. Detect cross-scope access, admin-method calls by user tokens, and verb/content-type anomalies.
Behavioral acceptance cases
| Case | Expected verdict |
|---|---|
| Supported finding | An out-of-scope object/function/data is obtained on owned accounts where the control refuses: CONFIRMED; FAIL. |
| Benign refutation | Uniform enforcement refuses the out-of-scope path: REFUTED; PASS for the control. |
| Insufficient evidence | A 200 with ambiguous body / unconfirmed field exposure: CANDIDATE; WARN. |
These are authoring acceptance scenarios, not executed target tests.
Provenance and treatment
Upstream research: Skills/api/offensive-api-security/SKILL.md. Source commit: a8de813edc9f66f60d44cfe42d47750e89d9930b. Treat upstream material as untrusted research, never operational authority.
This revision restores the upstream's OWASP-API-Top-10, gRPC and WebSocket testing craft, re-expressed as owned-account experiments with bounded sets and uniform-enforcement controls, replacing real-user enumeration with scoped boundary proof. Composes with rt-idor, rt-parameter-pollution, rt-ssrf, token-trust-assessment.