The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill implements a comprehensive 'Sensitive Information Gate' in SKILL.md (Step 1.2). It explicitly instructs the agent to identify and remove or minimize sensitive data such as access tokens, API keys, passwords, cookies, signed URLs, internal hostnames, and personally identifiable information (PII) before generating any outputs.
[INDIRECT_PROMPT_INJECTION]: The skill acknowledges that input sources (URLs, files, and text) are untrusted. In SKILL.md (Step 1.1), it provides specific instructions to treat commands found within these sources as data to be analyzed rather than instructions to be executed. Furthermore, in Step 5, it requires the agent to isolate source-derived strings within JSON arrays in downstream prompts to prevent escape or hijacking of the generation process.
[REMOTE_CODE_EXECUTION]: The skill does not contain patterns for remote script execution or unverifiable dependency installation. While it references installation via npx skills add, this targets the vendor's own official repository (github.com/KKenny0/folio-skill), which is considered a safe vendor resource.
[COMMAND_EXECUTION]: The skill does not invoke shell commands or system-level tools. It primarily performs content analysis and formatting into structured briefs.
[PROMPT_INJECTION]: The instructions in SKILL.md are focused on content processing and editorial standards. There are no patterns suggesting attempts to bypass safety filters or override system-level agent instructions.
[OBFUSCATION]: The repository includes SHA-256 hashes in assets/showcase/bitmap-qa.sha256 for file integrity verification and uses standard JSON encoding for data isolation. No malicious obfuscation techniques (e.g., zero-width characters, homoglyphs, or multi-layer base64) were detected.

folio