The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The logging example provided in SKILL.md (Step 6) uses mkdir($logdir, 0777, true), which creates a world-writable directory. This violates the principle of least privilege and could allow unauthorized local users to read, modify, or delete log files.
[DATA_EXFILTRATION]: The error handling logic in SKILL.md (Step 6) logs sensitive system information, including full stack traces ($e->getTraceAsString()) and the last executed database query ($DB->get_last_sql()), to the filesystem. Exposure of such internal details can assist an attacker in mapping the system architecture and database schema.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of APIs that ingest and process external data, establishing a potential injection vulnerability surface.
Ingestion points: API parameters defined in execute_parameters() and received by the execute() method in the provided PHP class examples.
Boundary markers: The guide instructs developers to use Moodle's validate_parameters() method to verify input against expected schemas.
Capability inventory: The implemented APIs have access to database operations through the $DB global and filesystem access via mkdir and file_put_contents in the logging example.
Sanitization: The instructions recommend using Moodle's PARAM_* constants (e.g., PARAM_INT, PARAM_TEXT) to clean and validate input values.

moodle-external-api-development