The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it processes data from untrusted external sources.
Ingestion points: The youtube-video command retrieves user-generated YouTube comments, and the nlp command extracts entities from arbitrary URLs provided during analysis.
Boundary markers: There are no explicit instructions or delimiters in the markdown templates or the command descriptions to warn the agent about potentially malicious instructions embedded in the retrieved data.
Capability inventory: The skill executes several local Python scripts via the ${CLAUDE_PLUGIN_ROOT}/scripts/claude-seo run wrapper for its core functionality and writes audit reports to local markdown files.
Sanitization: There is no evidence of sanitization or filtering of the ingested content before it is presented to the agent.
[COMMAND_EXECUTION]: The skill relies on executing local Python scripts to perform its functions.
Evidence: Multiple commands in SKILL.md invoke ${CLAUDE_PLUGIN_ROOT}/scripts/claude-seo run with scripts such as pagespeed_check.py, gsc_query.py, and ga4_report.py.
[EXTERNAL_DOWNLOADS]: The skill documentation references external dependencies and makes requests to remote services.
Evidence: references/keyword-planner-api.md mentions pip install google-ads. The skill also communicates with official Google API endpoints like googleapis.com, which are well-known and trusted services for this use case.

seo-google