The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script via the command line: python3 scripts/analyze_blog.py <blog-root> --batch --format json. This allows the agent to run code within the host environment to perform content analysis.
[DYNAMIC_EXECUTION]: The audit process relies on the execution of scripts/analyze_blog.py, an external script provided with the skill. This represents a dynamic execution vector where the agent's behavior depends on the integrity and safety of the local script file.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from blog files across various formats (.md, .html, .astro, .tsx, etc.).
Ingestion points: Recursive file discovery across common blog directories and CMS export folders as defined in SKILL.md (Step 1).
Boundary markers: No explicit instructions or delimiters are provided to the agent to treat the content of these files as data rather than instructions, potentially leading to the agent following malicious prompts embedded in the blog posts.
Capability inventory: The skill possesses the ability to execute shell commands (python3) and perform file system writes (reports/ directory).
Sanitization: While the skill includes logic to filter out sensitive paths like .env and .git/, it does not implement sanitization or validation of the actual content within the blog files before they are processed by the LLM or the analysis script.

blog-audit