The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from external web pages and search engine results pages (SERPs). This data is used to generate core outputs, which could allow an attacker to influence the agent's behavior via content found on target URLs or within search results.
Ingestion points: The target URL is fetched in Step 1 (scripts/fetch_page.py), and external content from search results (titles, snippets, PAA questions, and ad copy) is retrieved in Step 2 using the WebSearch tool.
Boundary markers: The prompt templates for User Story Derivation (Step 4) and Persona-Based Scoring (Step 6) do not include explicit instructions to ignore embedded commands or clear delimiters for the ingested content.
Capability inventory: The skill can execute local Python scripts, perform web searches, and write JSON summaries to the local file system (.seo-cache/).
Sanitization: There is no evidence of filtering or sanitization of the content extracted from target pages or search results before it is interpolated into templates.
[COMMAND_EXECUTION]: The skill explicitly invokes external scripts to handle data processing, which represents a standard but relevant execution capability.
Evidence: The execution pipeline in SKILL.md triggers scripts/fetch_page.py for target acquisition and scripts/parse_html.py for content extraction.

seo-sxo