The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possui a structural surface for indirect prompt injection as it requires the agent to ingest and analyze untrusted external data.
Ingestion points: Data enters the agent's context through scripts/prepare_search_term_analysis.py (which reads user-provided advertising report files) and scripts/fetch_listing_context.py (which scrapes HTML content from Amazon URLs).
Boundary markers: There are no explicit markers around the untrusted data, although the skill defines a strict JSON schema for the agent's classification output to help maintain task structure.
Capability inventory: The skill is capable of executing shell commands via subprocess.run and performing local file read/write operations.
Sanitization: The skill includes a sanitization step in scripts/clean_search_term_report.py that removes hidden Unicode characters (such as zero-width spaces) and control characters from the report data before it is processed.
[COMMAND_EXECUTION]: The scripts/browser_utils.py module uses subprocess.run to execute curl for fetching product information. The execution uses a structured list of arguments and targets well-known services (Amazon domains), which mitigates risk, but it remains a capability that executes external commands based on report data.

zach-search-term-report-analyzer