The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes error messages originating from external integrations and providers, which are considered untrusted data sources.
Ingestion points: The skill reads data from table cells using clay tables query, clay tables rows list, and clay tables rows get in SKILL.md. The specific field ingested is the .value.error string.
Boundary markers: There are no explicit boundary markers or instructions to the agent to ignore potentially malicious instructions embedded within the error messages.
Capability inventory: The agent has access to the clay CLI suite (query, rows, columns), jq for data processing, and general Bash execution as defined in the skill's allowed-tools.
Sanitization: Step 5 in SKILL.md recommends normalizing messages by stripping row-specific values, which provides a layer of data aggregation but does not specifically filter for malicious instructions.

tables-error-sweep