▸case-07 Inspect an agent manifest documented as Static Code Analyzer - Reference Only that includes filesystem_write and curl_fetch tools. Is this configuration secure under standard privilege management principles, and what specific remediation action should be set? | pass→pass | 11,582 | 10,797 | -7% | 1 | 1 | 0% | 1,912 | 1,654 | -13% | 0 | 0 | — |
▸case-16 A developer claims their agent plugin is strictly a read-only reference for policy documents. Audit logs show the plugin requested terminal_exec and external_http. Which specific security policy section is violated when terminal_exec is requested by a policy document reader? | pass→pass | 6,967 | 4,287 | -38% | 1 | 1 | 0% | 1,110 | 788 | -29% | 0 | 0 | — |
▸case-06 We have an agent skill whose system prompt says 'This skill is strictly for searching and reading local repository logs', but its defined tools include shell_exec and post_webhook. What security risk rating should be reported in the code review? | pass→pass | 9,451 | 15,606 | +65% | 1 | 1 | 0% | 1,405 | 1,454 | +3% | 0 | 0 | — |
▸case-01 Audit my agent skill manifest to determine if there are security inconsistencies between what the documentation states and the tools granted. | fail→fail | 16,845 | 18,028 | +7% | 1 | 1 | 0% | 1,336 | 2,134 | +60% | 0 | 0 | — |
▸case-02 Review a skill manifest that is documented as an interactive file editor and grants write and exec permissions. Determine if the permissions match the documented purpose. | fail→fail | 11,209 | 8,483 | -24% | 1 | 1 | 0% | 1,694 | 1,267 | -25% | 0 | 0 | — |
▸case-03 Evaluate an API gateway policy where an OAuth scope grants read access to user profiles and the underlying endpoint allows HTTP GET requests to the user profiles endpoint. | fail→fail | 15,609 | 14,622 | -6% | 1 | 1 | 0% | 2,386 | 2,139 | -10% | 0 | 0 | — |
▸case-04 Review an automated deployment skill manifest that claims to deploy web applications to staging and requests execute and network tools. | fail→fail | 14,292 | 11,853 | -17% | 1 | 1 | 0% | 2,534 | 1,883 | -26% | 0 | 0 | — |
▸case-05 Analyze an AI assistant tool configuration where the documentation describes the tool as a documentation retriever and reference lookup, but the tool manifest specifies write_file, bash_exec, and http_request capabilities. Produce a structured security finding. | pass→pass | 14,730 | 13,904 | -6% | 1 | 1 | 0% | 1,394 | 1,146 | -18% | 0 | 0 | — |
▸case-08 A developer created an agent capability described in its README as Database Schema Reader (Read-Only). The tool definitions grant sql_execute with full DDL/DML mutation privileges and network_connect. How should a security audit categorize this finding? | fail→pass | 11,426 | 11,357 | -1% | 1 | 1 | 0% | 1,853 | 1,800 | -3% | 0 | 0 | — |
▸case-09 A customer support agent helper is documented as Knowledge Base Reference Skill. Its manifest grants file_write and cmd_run. Assess whether this skill complies with least-privilege design and provide the exact rule violation ID. | fail→fail | 11,159 | 5,068 | -55% | 1 | 1 | 0% | 1,718 | 932 | -46% | 0 | 0 | — |
▸case-10 During a security audit of an AI agent ecosystem, you find a skill titled API Specification Viewer (Read-Only) that asks for exec_command and fetch_url permissions. What risk severity level applies to the presence of exec_command? | fail→fail | 7,621 | 8,327 | +9% | 1 | 1 | 0% | 1,133 | 1,283 | +13% | 0 | 0 | — |
▸case-11 Review this agent configuration manifest: name: viewer-skill, description: Reference skill for internal wiki pages, tools: [fetch_wiki_page, edit_wiki_page, run_script, outbound_http]. List the specific tools that must be disabled to satisfy read-only reference status. | pass→pass | 3,592 | 3,266 | -9% | 1 | 1 | 0% | 702 | 601 | -14% | 0 | 0 | — |
▸case-12 An AI coding assistant imports a helper declared as Read-only Linter Reference. The helper manifest enables fs.writeFile, child_process.exec, and net.connect. What OWASP or CWE security weakness ID describes this privilege escalation vector? | fail→fail | 9,309 | 8,924 | -4% | 1 | 1 | 0% | 1,574 | 1,697 | +8% | 0 | 0 | — |
▸case-13 An open-source AI agent extension claims in its summary to be a Read-only Compliance Checker, but requests system_exec and network_post tools. What exact Boolean pass/fail evaluation result should an automated security pipeline return? | fail→pass | 13,185 | 4,188 | -68% | 1 | 1 | 0% | 713 | 675 | -5% | 0 | 0 | — |
▸case-14 Evaluate a pull request introducing a skill named Metric Dashboard Reference with description Provides read-only metric visualization. The tool capabilities requested are create_file, bash, and http_client. What specific risk classification code should be logged? | fail→fail | 11,145 | 7,249 | -35% | 1 | 1 | 0% | 1,887 | 1,245 | -34% | 0 | 0 | — |
▸case-15 In an agent framework, a skill's description states Reference tool for checking cloud configuration compliance, while its capability array lists modify_config, run_cli, and socket_send. Which requested tool capability poses an arbitrary command execution vulnerability? | pass→pass | 4,536 | 3,069 | -32% | 1 | 1 | 0% | 659 | 519 | -21% | 0 | 0 | — |
▸case-17 Assess an enterprise AI agent manifest: title: HR Policy Reference, summary: Read-only access to employee handbooks, granted_tools: [read_pdf, write_temp, execute_binary, send_email]. Which specific capability in granted_tools creates an unauthenticated code execution vulnerability? | pass→pass | 10,185 | 8,563 | -16% | 1 | 1 | 0% | 903 | 775 | -14% | 0 | 0 | — |
▸case-18 A third-party agent skill is advertised as Reference Manual Searcher. Its tool declaration grants execute_code, update_record, and open_socket. What audit status code should an automated skill scanner assign to this manifest? | fail→fail | 8,799 | 18,078 | +105% | 1 | 1 | 0% | 1,557 | 1,033 | -34% | 0 | 0 | — |
▸case-19 A cloud security scanner analyzes an agent definition claiming Read-only AWS Architecture Viewer but finding aws_cli_exec and upload_s3 in the allowed tool list. What CVSS base score severity level should be assigned to aws_cli_exec on a read-only viewer? | fail→pass | 12,187 | 11,259 | -8% | 1 | 1 | 0% | 2,087 | 2,245 | +8% | 0 | 0 | — |
▸case-20 When conducting a privilege review of an AI agent manifest labeled Read-Only SQL Query Reference, you see drop_table_tool, system_shell, and curl_api. What specific security mitigation step must be taken before approval? | pass→pass | 5,654 | 4,836 | -14% | 1 | 1 | 0% | 999 | 924 | -8% | 0 | 0 | — |
▸case-21 An agent manifest contains description: Read-only threat intelligence reference, tools: [query_intel, write_file, execute_script, network_listen]. Which requested tool presents a network socket binding risk? | pass→pass | 3,041 | 2,646 | -13% | 1 | 1 | 0% | 449 | 467 | +4% | 0 | 0 | — |
▸case-22 Evaluate a candidate agent skill whose metadata claims Read-only incident response reference sheet. Its definition includes exec_command and network_post. Why does network_post fail security review for a read-only reference tool? | pass→pass | 10,024 | 9,512 | -5% | 1 | 1 | 0% | 1,638 | 1,503 | -8% | 0 | 0 | — |
▸case-23 An AI system administrator checks a newly registered skill manifest. The header reads Read-only Status Checker, but the permission scope requests write_storage, spawn_process, and fetch_remote. What audit report field should record write_storage? | fail→pass | 6,310 | 6,985 | +11% | 1 | 1 | 0% | 1,139 | 1,451 | +27% | 0 | 0 | — |