The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file contains a dynamic command execution block used during skill initialization to generate a directory listing.
Evidence: !d=${CLAUDE_SKILL_DIR}; command -v skill-toc >/dev/null 2>&1 && exec skill-toc "$d"; s=$HOME/.claude/skills/plugin-utils/bin/skill-toc; [ -x "$s" ] && exec "$s" "$d"; echo "(skill-toc unavailable...)"``
This command is a benign utility intended to help the agent navigate the skill's reference files by executing a Table of Contents generator if the plugin-utils package is installed.
[COMMAND_EXECUTION]: The skill includes a Python script examples/ciz_panel.py that executes SQL commands on the remote WRDS (Wharton Research Data Services) database server.
The script utilizes the psycopg2 library to establish connections to wrds-pgdata.wharton.upenn.edu.
Security best practices are followed, including the use of parameterized queries (%(start)s, %(end)s) to prevent SQL injection and reliance on local configuration files (~/.pgpass) rather than hardcoded credentials.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect injection as it ingests user-supplied parameters to build database queries.
Ingestion points: Command-line arguments for start dates, end dates, and usernames in examples/ciz_panel.py.
Boundary markers: The code uses standard SQL parameterization provided by the database driver to separate user data from the query logic.
Capability inventory: The skill performs network database operations and optional local file writing (Parquet format).
Sanitization: Input is processed through argparse and passed to a mature database driver, mitigating the risk of injection during processing.

crsp-v2