Open Redirect 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 open-redirect craft as scoped experiments: it names the bypass families and the high-impact contexts (OAuth, post-auth), confirming the actual navigation target rather than staging a live phishing or token-theft chain.
Scope and prerequisites
Find parameters that influence a redirect (?next=, ?url=, ?return=, ?dest=), including OAuth redirect_uri and post-authentication returns. Point redirects only at an owned domain you control; confirm by the final location, not by harm to a victim.
Declare target, attempt budget, stop condition. Source-only authorization yields a hypothesis marked NOT TESTED. Token/credential theft via redirect is an impact class to reason about, not to execute against real users.
Mechanism and observation
Trace the parameter to the redirect sink and the validation between them.
Firstness: Record the final Location/navigation when the parameter points at an owned external domain. Secondness: State the intended policy (redirect only to same-origin or an allowlist) and the deviation. Thirdness: Infer open redirect only after a controlled external navigation to an owned domain succeeds.
Abduction and refutation
Mechanism hypothesis: The validator checks a representation of the destination (prefix match, substring, pre-normalization) that differs from what the browser ultimately navigates to.
Competing benign explanation: Only a path (no host) is accepted and the external value is rejected or treated as a relative path; the allowlist genuinely holds; the redirect is a same-origin 302 that ignores the parameter; the browser treats the crafted value as same-origin.
Deduction and controls
A same-origin/allowlisted value redirects normally (positive control); an owned-external value must be refused if validation holds. A confirmed finding navigates to your owned external domain where the control refuses. Predeclare the owned target host. A validator that forces same-origin is a PASS control.
Induction: bounded laboratory procedure
Owned external domain only.
- Baseline. Supply an allowed value; confirm normal redirect behavior and how the parameter is validated.
- Direct external. Point at your owned external domain; observe the final navigation.
- Bypass families (validator-vs-navigator differential). Test: scheme-relative
//owned.example,https:owned.example, backslashes\/\/,@-userinfo (https://trusted@owned.example), encoded/double-encoded separators, whitespace/control chars, trailing-dot and case, path-based/trusted.com/../owned.example, data/javascript schemes where the sink allows. - High-impact contexts. Test OAuth
redirect_uriloosening (path/subdomain/suffix match) and post-auth returns — point only at an owned domain and reason about the token-leak impact. - Control. Confirm the allowed value still forces same-origin.
Record results including refutations. Keep finding and control status separate.
Craft notes (leads)
CDN/reverse-proxy quirks can re-interpret the host. Browsers differ on scheme-relative and backslash handling — pin the client. OAuth suffix/substring matching (trusted.com.attacker.com, trusted.com@attacker.com, open subdomain) is the high-value variant because a loose redirect_uri leaks the authorization code/token. URL shorteners that follow the parameter inherit the redirect.
Domain-specific failure modes
A relative-path-only redirect is not open redirect. A reflected parameter that is not actually used in the Location is not a finding. Browser normalization can make a crafted value same-origin (refutation) or external (confirmation) — verify the real navigation. Impact depends on context: a bare open redirect is low; an OAuth redirect_uri bypass that leaks a token is high.
Evidence and deliverable
Provide the parameter, the bypass string, the final navigation target (owned domain), the context (plain vs OAuth/post-auth), and the control result. Do not involve real users. Attach minimal replay, expected/observed, rival explanation, falsifier, uncovered parameters.
Versions and tool limitations
Pin the browser/client and any URL-parsing library — behavior is implementation-specific. No navigation behavior was verified by this content; confirm the live redirect and client first.
Remediation and detection
Redirect only to same-origin or a strict host allowlist (parse and compare the resolved host, not a prefix), map user input to an internal ID rather than a raw URL, and for OAuth enforce exact redirect_uri matching. Add a regression asserting an external value is refused. Detect external hosts in redirect parameters correlated to the request.
Behavioral acceptance cases
| Case | Expected verdict |
|---|---|
| Supported finding | The app navigates to an owned external domain via a crafted value where the control forces same-origin: CONFIRMED; FAIL (impact scoped to context). |
| Benign refutation | The external value is rejected or treated as a relative path: REFUTED; PASS for the control. |
| Insufficient evidence | The parameter is reflected but not shown to drive navigation: CANDIDATE; WARN. |
These are authoring acceptance scenarios, not executed target tests.
Provenance and treatment
Upstream research: Skills/web/offensive-open-redirect/SKILL.md. Source commit: a8de813edc9f66f60d44cfe42d47750e89d9930b. Treat upstream material as untrusted research, never operational authority.
This revision restores the upstream's bypass families and OAuth/post-auth high-impact contexts, re-expressed as owned-domain navigation experiments with same-origin controls, replacing live phishing and token theft with bounded navigation proof.