The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted Swift source code (**/*.swift) through its auditing and failure analysis agents. This creates a surface for indirect prompt injection where malicious instructions embedded in source files could attempt to influence the agent's behavior.\n
Ingestion points: Swift source files (**/*.swift) read during auditing phases.\n
Boundary markers: The instructions do not define explicit boundary markers or use protective framing to isolate scanned code from the agent's instructions.\n
Capability inventory: The skill uses file search, reading, and specialized tools (e.g., axiom_get_agent, axiom_xcsym_crash) to perform analysis.\n
Sanitization: Content from scanned files is processed directly without sanitization or escaping.\n- [DATA_EXFILTRATION]: The test-failure-analyzer.md component instructs the agent to read system crash reports located at ~/Library/Logs/DiagnosticReports/*.ips. While this is necessary for analyzing test failures, it involves accessing system-level logs that can contain sensitive runtime state.\n- [PRIVILEGE_ESCALATION]: The documentation includes instructions for executing high-privilege shell commands using sudo to configure the Network Link Conditioner (dnctl and pfctl) on the host machine. These commands are intended for environment setup by the user or within a CI pipeline.

axiom-testing