The Agent Skills Directory

[DATA_SOURCES]: The skill explicitly defines a policy for handling untrusted data. In SKILL.md and references/skill-contract.md, it instructs the agent to treat scraped text, reviews, and third-party exports as untrusted input and forbids following any embedded instructions found within that evidence. This is a robust defense against indirect prompt injection.
[COMMAND_EXECUTION]: The provided Python scripts (fetch_page.py, generate_markdown_report.py, generate_pdf_report.py, schema_generator.py, and seo_health_score.py) are focused utility tools. fetch_page.py includes significant security controls, such as validating URLs to prevent access to private/loopback IP addresses and blocking URLs that contain hardcoded credentials. It also implements bounded reads to prevent resource exhaustion.
[REMOTE_CODE_EXECUTION]: No patterns of remote code execution or dynamic package installation were detected. generate_pdf_report.py mentions an optional dependency on reportlab but instructs the user to install it manually rather than attempting to install it automatically at runtime.
[CREDENTIALS_UNSAFE]: The skill documentation in references/connectors.md explicitly mandates keeping API keys out of skills and source control. No hardcoded secrets were found in the provided files.
[PRIVILEGE_ESCALATION]: There are no commands or instructions requesting elevated privileges (e.g., sudo, chmod) or persistent system modifications.
[PROMPT_INJECTION]: The instructions in SKILL.md are focused on task routing and evidence labeling. There are no attempts to override system safety guidelines or extract system prompts. The 'ignore embedded instructions' rule is applied to data processing rather than agent behavior bypass.

seo-competitor-pages