菜单
GraphQL node/edge object-level authorization (BOLA/IDOR): Relay global IDs, nested resolvers, connections, mutations, and field-level leaks. Use when node(id), edges, or ID arguments appear under GraphQL and classic REST IDOR coverage is incomplete. Authorized assessments only.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/tomysh1337/openstarry-code --skill idor-graphql-nodes命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
Frontend accessibility (a11y) testing checklist at WCAG high level: semantic HTML, keyboard, focus, names/roles, contrast, forms, media, and ARIA discipline. Use when accessibility, a11y, 无障碍, WCAG, screen reader, keyboard navigation, ARIA, focus trap, color contrast, or accessible UI review. Complements code-quality-standards; does not replace product design systems or legal audits.
Design account lockout and progressive auth-failure controls: attempt counters, lock duration, dual keys (account + IP), unlock paths, enumeration-safe UX, and lockout vs soft rate limits. Use when login lockout policy, failed-attempt thresholds, temporary ban after N passwords, progressive delay, or lockout vs rate-limit design for owned applications.
Authorized account-takeover (ATO) methodology that chains password-reset, session fixation, IDOR/BOLA, JWT/API tokens, and OAuth/OIDC failures into end-to-end impact paths. Use when planning or executing multi-vector ATO assessment with owned test accounts only — not single-class deep dives alone.
| name | idor-graphql-nodes |
| description | GraphQL node/edge object-level authorization (BOLA/IDOR): Relay global IDs, nested resolvers, connections, mutations, and field-level leaks. Use when node(id), edges, or ID arguments appear under GraphQL and classic REST IDOR coverage is incomplete. Authorized assessments only. |
node(id:), nodes(ids:), Relay global IDs, or connection/edge patterns (edges { node { ... } }).id / uuid / slug and nest sensitive fields (user, order, paymentMethod, internalNote).graphql-and-hidden-parameters maps operations but node/edge authZ is unproven.idor-broken-object-authorization covers REST but GraphQL nested resolvers remain untested.api-auth-and-jwt-abuse), injection in args (injection-checking), or WebSocket subscriptions auth after upgrade (websocket-authz-deep / websocket-security).Inventory ID-bearing surfaces
From introspection, client operations, or proxy history, list every place an object is selected:
| Surface | Examples |
|---|---|
| Root query | user(id:), order(id:), node(id:) |
| Batch | nodes(ids:) , aliases a: user(id:"1") |
| Connection | viewer { orders(first:) { edges { node { id } } } } |
| Nested field | order { customer { email ssn } } |
| Mutation | updateOrder(id:, input:), deleteNode(id:) |
| Subscription | orderUpdated(id:) (hand long-lived channel auth to WS skills) |
Tag each ID as owned / shared / admin for account A and B. Decode Relay global IDs (base64("TypeName:dbId")) only to understand type confusion risk — ACL still applies per type.
Dual-session fixtures
Capture honest queries as A for A-owned nodes. Hold B’s session (cookie or Bearer) separately. Never mix tokens. Record baseline field sets so over-fetch is measurable.
Horizontal node swap (core GraphQL BOLA)
Replay A’s ID under B’s auth:
query {
node(id: "A_RELAY_ID") {
... on Order { id total status customer { email } }
}
order(id: "A_RAW_ID") id total
| Need | Skill |
|---|---|
| Classic BOLA methodology, dual-account matrix | idor-broken-object-authorization |
| Schema recovery, hidden args, batching survey | graphql-and-hidden-parameters |
| Subscription transport / post-upgrade authZ | websocket-security, websocket-authz-deep |
| JWT/session broken authN | api-auth-and-jwt-abuse |
| Implement resolver ACL / deny-by-default | code-quality-standards |
nodes(ids:) / alias batch behavior recordedExpect deny (null + auth error, or 200 with null field) — not full foreign data. Test status and body: GraphQL often returns HTTP 200 with partial errors.
Edge and nested resolver authZ
Root may authorize while children leak:
query {
me {
friends(first: 50) {
edges {
node {
email phone privateNotes
orders { id total } # friend → their orders?
}
}
}
}
}
Also try: start from a legit parent B can see, then walk edges to A-only children; cross-link order(id:A) { organization { secrets } }. Every resolver that returns an object must re-check principal vs object.
Type confusion and global ID tricks
AdminUser:1 / Order:A variants the server might resolve without ACL.nodes(ids: [A, B, C]) under B — note which indices return data (batch ACL gaps).... on AdminProfile on a node B should not see.Mutations and partial success
As B, run update*, delete*, addEdge, invite, transfer with A’s IDs. Confirm no silent partial apply (edge created but field error elsewhere). Re-fetch as A to prove integrity. Prefer read-only proof first.
Alias and batch ACL gaps
One HTTP request, many aliased foreign reads; JSON array batching if enabled. Document whether authZ/rate limits are per-operation or per-HTTP-request. Stay within attempt budgets.
Field-level and error oracles
Forbidden fields returning “not found” vs “not authorized” can enumerate IDs. Error extensions that echo internal ids or stack traces are secondary findings. Prefer one clean A/B pair over noisy sweeps.
Evidence and remediation notes
Save operation name, variables, redacted response, and account matrix. Remediation: authorize in resolvers or a central authZ layer per node type and edge; never trust client type labels in global IDs; apply same checks to mutations and subscriptions; complexity limits do not replace ACL.