The Agent Skills Directory

[SAFE]: No malicious behavior or high-risk patterns were identified during the analysis. The skill is entirely instructional and focus on generating text-based artifacts.
[DATA_EXFILTRATION]: While the skill requests sensitive business information (such as contribution margins, cash runway, and burn rate) to compute thresholds, it lacks any mechanism (like curl, wget, or external API calls) to exfiltrate this data. The inputs are used strictly for local computation and documentation.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data through an interview process.
Ingestion points: SKILL.md (Interview section).
Boundary markers: None identified.
Capability inventory: The skill has no tool access, file-system write capabilities, or network operations.
Sanitization: None present.
Result: The risk is negligible as the agent only generates a text policy and cannot execute commands or send data externally based on injected instructions.
[COMMAND_EXECUTION]: The skill does not contain any shell commands, subprocess spawns, or dynamic context injection patterns (!command).

ad-spend-guardrails