The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, potentially untrusted content from files (e.g., draft-cloud-migration.txt and ./content/*.md) and sends this data to the Originality.ai API. This represents an ingestion point for untrusted data.
Ingestion points: draft-cloud-migration.txt (via jq --rawfile), ./content/*.md (via path.read_text() in Python).
Boundary markers: The skill uses jq to safely wrap text in JSON for shell execution and Python's requests JSON serialization, which helps prevent command injection but does not prevent the LLM or API from being influenced by the content of the text itself.
Capability inventory: File reading (path.read_text()), network operations (requests.post, curl), and shell command execution (curl, jq).
Sanitization: Standard JSON serialization is used, but there is no specific filtering or sanitization of the content itself before it is sent to the API.
[COMMAND_EXECUTION]: The skill provides examples that execute shell commands (curl, jq) and run Python scripts that interact with the local filesystem and network.
[EXTERNAL_DOWNLOADS]: The skill performs network operations (POST requests) to api.originality.ai. These are documented as official vendor API endpoints and are treated as safe interactions for the skill's stated purpose.

originality-ai