The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external URLs through the webfetch tool and interpolates extracted values (e.g., page titles, headlines) into shell commands. This creates a vulnerability where a malicious website could attempt to inject command arguments or influence agent behavior via its content.\n
Ingestion points: External web content retrieved by the webfetch tool as defined in SKILL.md.\n
Boundary markers: No delimiters or instructions to ignore potentially malicious embedded content are included.\n
Capability inventory: Shell execution of local Python scripts (scripts/schema_gen.py and scripts/report_publish.py).\n
Sanitization: The instructions lack requirements for validating or escaping the external data before it is used in the command line.\n- [COMMAND_EXECUTION]: The skill requires the execution of local Python scripts via the system shell to generate schema and finalize reports.\n
Evidence: The 'Process' section explicitly directs the agent to run python scripts/schema_gen.py using fields derived from page content.\n
Evidence: The 'Output' section instructs the agent to run python scripts/report_publish.py to generate branded HTML and PDF files.

schema-generator