The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from diverse external sources, including JSON, CSV, and YAML files, as well as API responses from GitHub and GitLab. This represents a vulnerability surface where malicious instructions could be embedded in data processed by the agent.\n
Ingestion points: The skill utilizes specialized tools like jq, yq, dasel, qsv, and mlr to handle external file and API data.\n
Boundary markers: While the instructions emphasize tool correctness, they do not explicitly mandate delimiters or safety warnings for the agent when processing or outputting external content.\n
Capability inventory: The skill has access to Bash, Read, and Write capabilities, allowing for significant file system and network operations based on processed data.\n
Sanitization: No specific data sanitization or escaping mechanisms are described beyond the inherent validation of structured parsers.\n- [DYNAMIC_EXECUTION]: The references/yq-cookbook.md file describes a workflow for extracting run: steps from GitHub Action YAML configurations into a shell script and executing them. This is a form of dynamic script generation and execution.\n
Evidence: The documentation demonstrates using yq to write a YAML field to a temporary file and then executing it with bash -eo pipefail.\n
Safety context: The skill includes an explicit security warning stating that code from untrusted sources is dangerous and that directory sandboxing is not a complete defense.\n- [DYNAMIC_EXECUTION]: The skill implements a PreToolUse hook that executes a local Python script (scripts/pre_bash_structured_warn.py) to validate Bash commands. This is a functional use of dynamic execution for environment enforcement and safety gating.\n- [DATA_EXFILTRATION]: The skill promotes secure data handling by providing specific recipes for excluding sensitive fields (such as password or secret_key) during transformation and extraction tasks in CSV and JSON formats, helping to prevent unintentional data exposure.

data-tools