The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external, untrusted codebases, which creates an inherent surface for indirect prompt injection. This is mitigated through a multi-layered security architecture described in the 'Universal execution safety' and 'Write isolation' sections of SKILL.md, which mandate the use of OS-enforced sandboxes with no network access, read-only target mounts, and strictly controlled scratch directories for all analysis tasks.
[SAFE]: The JavaScript validation scripts (validate-coverage-ledger.cjs and validate-findings.cjs) demonstrate excellent security hygiene by utilizing non-blocking, no-follow file descriptors (O_NOFOLLOW, O_NONBLOCK) and fstat checks to prevent Time-of-Check to Time-of-Use (TOCTOU) symlink attacks. They also implement manual structural limits for JSON parsing to defend against resource exhaustion via deeply nested or oversized inputs.
[SAFE]: The skill implements least-privilege principles by requiring all interactions with external services, cloud APIs, and registries to be simulated with dummy data or local fixtures, effectively eliminating the risk of accidental data exfiltration or unauthorized production modifications during the audit workflow.

kata-security