The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructs the agent to execute multiple Python scripts using a dynamically resolved path. In the 'Formal evidence handoff' section of SKILL.md and the 'Shared tools' section of references/markdown-delivery-contract.md, the agent is directed to read a local file named '.seo-suite-runtime.json' to obtain the 'suite_root' value. This value is used to construct absolute paths for executing scripts such as 'validate_seo_findings.py', 'manage_markdown_site.py', and 'manage_markdown_work.py'. This practice of loading execution paths from a workspace configuration file constitutes dynamic loading from a computed path, which could allow an attacker to redirect execution to malicious code by providing a crafted configuration file.
[INDIRECT_PROMPT_INJECTION]: The skill audits external, untrusted publisher content, which introduces a vulnerability to indirect prompt injection where instructions hidden in articles or metadata could influence agent behavior.
Ingestion points: The skill captures external article pages, listing pages, raw/rendered metadata, and news sitemap/feed rows (SKILL.md, Procedure Step 2).
Boundary markers: The instructions do not define delimiters or provide warnings to the agent to ignore embedded instructions within the captured data.
Capability inventory: The skill has the capability to write multiple Markdown files to the filesystem and execute Python scripts that perform validation and site management.
Sanitization: There is no evidence of sanitization, escaping, or schema validation for the ingested external content before it is processed or incorporated into reports.
[COMMAND_EXECUTION]: The skill makes extensive use of the shell to run Python scripts for validating findings and managing Markdown-based sites. While these commands are intended for the skill's primary purpose, the use of shell execution with variable paths derived from local configuration increases the overall risk profile.

seo-news-discover