review-http-security
/review-http-security
Przegląda serwery HTTP i API pod kątem braków hardeningu: uwierzytelniania, uprawnień, nagłówków, limitów, TLS i ekspozycji endpointów.
- Autor
- yuri-xyz ↗
- Repozytorium
- skills ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 03.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 3
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 03.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Przegląda serwery HTTP i API pod kątem braków hardeningu: uwierzytelniania, uprawnień, nagłówków, limitów, TLS i ekspozycji endpointów.
Oryginalny opis (EN)
Use when reviewing HTTP servers, APIs, or their hosting configuration for hardening gaps, such as authentication, authorization, input parsing, uploads, CORS, headers, cookies, limits, SSRF, redirects, TLS, proxy trust, or exposed endpoints and metadata.
Jak działa
- Objective
- Workflow
- Prefer Fixes Like
- Avoid
- Output
Podgląd SKILL.md
--- name: review-http-security description: Use when reviewing HTTP servers, APIs, or their hosting configuration for hardening gaps, such as authentication, authorization, input parsing, uploads, CORS, headers, cookies, limits, SSRF, redirects, TLS, proxy trust, or exposed endpoints and metadata. --- # Review HTTP Security ## Objective Review the current or proposed changes to an HTTP server, API, or the configuration that hosts it (reverse proxy, CDN or edge worker, container, systemd unit, or platform config) for hardening gaps a remote client could exploit. Find them by walking each route and each layer between the internet and the handler, naming the request an attacker sends and what it gets them, and prove each one with a failing test or a specific request before fixing it. Judge everything by least trust and least exposure: treat every request as hostile until parsed and authorized, return only what the client needs, with errors that are short, generic, and free of internals, and advertise nothing about the server's software, versions, or topology. This is the HTTP specialization of `/review-security`, which still covers general injection, memory safety, and file handling inside handlers. Use `/review-cryptography` for token and session cryptography. ## Workflow 1. Inspect the recent diff, pending plan, or relevant files, and map the surface: every route and method, who may call it, what it reads and changes, and the layers in front of it: TLS termination, reverse proxy, CDN or edge, load balancer, and the platform's own settings. 2. Note the deployment facts that change the answers: public or internal, browser clients or machine clients, cookies or bearer tokens, single or multi-tenant, and which layer owns TLS, limits, and headers. Read the actual proxy and platform config instead of assuming defaults. 3. Walk the surface through the classes in `references/http-classes.md`. For each candidate, name the attacker (anonymous internet client, signed-in user of another tenant, malicious site in the victim's browser, or a client on the internal network), the request, and the impact. Drop candidates no attacker can reach, such as a header the proxy already sets and the app cannot bypass. 4. Check the tests: unauthenticated and cross-tenant requests to every protected route, oversized, malformed, and type-confused bodies and uploads, hostile origins, and redirect and SSRF targets. Note the missing ones. 5. Fix a gap when it is confirmed, local, and the right behavior is clear: write a failing test first when the project has tests, fix it at the layer that owns the concern, and watch the test pass. Prefer the framework's or proxy's built-in mechanism over hand-written middleware. 6. Report, without fixing, gaps whose fix changes authentication, the public API, client compatibility (such as tightening CORS or CSP), or deployed infrastructure; those need the user. Never probe live production endpoints or change live hosting settings; report what to change instead. 7. Run the focused tests and checks for the changed code. ## Prefer Fixes Like - Layered checks on auth and other critical paths: authorization at the route and again where data is read or changed, assertions for states that should be impossible, and fail-closed errors that are logged. - Default-deny routing: authentication and authorization applied by the router or a middleware every route passes through, with explicit public exceptions. - Limits set once, at the layer that sees the request first, and matched or tightened behind it: body size, header size, timeouts, rate limits, and concurrency. - Parsing every request once, at the edge, with the project's schema or validation library into typed values, rejecting unknown, duplicate, and out-of-range input instead of coercing it. - Allowlists for CORS origins, redirect targets, outbound hosts, and content types. - Security headers and cookie flags set in one place for every response. - Trusting forwarded headers only from known proxies, configured explicitly. - Least exposure: generic, concise error responses with a stable code and a request ID, details kept in server logs, and explicit output schemas that return only the fields the caller needs. - No fingerprints: banner and version headers removed, and default error and status pages replaced at every layer. ## Avoid - Duplicating a configured limit or header at every layer with different values, so nobody knows which one applies. Layered authorization and invariant checks on critical paths are different: keep them, and share one policy between the layers. - CSP, CORS, or cookie changes that break existing clients without the user's agreement. - Security through obscurity, such as hidden routes or unusual ports, in place of authentication. - Logging credentials, tokens, cookies, or full request bodies while fixing. ## Output If no gap was found, say so and name the routes, layers, and classes reviewed. Otherwise list findings, most severe first: ```markdown - 🔴 <file:line or config> <class>: <attacker> sends <request> → <impact>. Fixed; regression test `<name>`. - 🟡 <file:line or config> <class>: <attacker> sends <request> → <impact>. Not fixed: <why>. - 🔵 <file:line or config> Hardening: <missing defense or test>. Added / not added: <why>. ``` 🔴 unauthenticated access to data or actions, cross-tenant access, account takeover, SSRF to internal services, or an exposed admin or debug surface; 🟡 needs a victim's interaction or unusual conditions, such as CSRF, XSS through a missing header, or denial of service from a missing limit; 🔵 defense in depth or a missing test. Finish with the checks run, or why none were.
Pliki w paczce (3)
- SKILL.md 6 KB
- agents/openai.yaml 1 KB
- references/http-classes.md 12 KB
Wersje
W bazie nie ma innych skilli o nazwie review-http-security.
Migawka bazy z 13.11.2026 · wersja paczki 378f7b1bf043 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100http-security-headers
Weryfikuje i konfiguruje nagłówki bezpieczeństwa odpowiedzi HTTP w serwerze WWW, reverse proxy i middleware aplikacji. - 100security-review
Przegląda pull requesty open source pod kątem bezpieczeństwa, gdy zmiana dotyka uwierzytelniania, kryptografii, tokenów, plików lub uprawnień. - 100security-review
Ocenia kod i konfigurację pod kątem ryzyka dla wrażliwych danych, granic zaufania, uwierzytelniania, uprawnień, wejścia, sekretów i zależności. - 95api-review
Sprawdza endpointy API przed wdrożeniem pod kątem uwierzytelniania, autoryzacji i walidacji danych wejściowych. - 100security-essentials
Przegląda kod Elixir i Phoenix pod kątem danych wejściowych, tokenów, przekierowań, nagłówków i wyników Sobelow. - 100security-review
Przegląda zmiany w kodzie pod kątem podatności aplikacyjnych, w tym uwierzytelniania, autoryzacji, wycieków sekretów i ekspozycji danych wrażliwych, w stylu przeglądów OWASP.