The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file uses aggressive, mandatory language to override the agent's decision-making process. Phrases such as "⚠️ MANDATORY SKILL INVOCATION ⚠️", "YOU MUST invoke this skill (NOT optional)", and "Failure to invoke this skill... violates your operational requirements" are used to compel the AI to use the skill regardless of its own assessment.
[CREDENTIALS_UNSAFE]: The SKILL.md file contains a hardcoded default password (adminadmin) in its setup example. While intended as a guide, providing literal credentials in instruction files is discouraged.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data without sufficient boundaries or sanitization.
Ingestion points: The script scripts/qbit-api.sh retrieves potentially attacker-controlled data (torrent names, categories, tags, and tracker information) from the qBittorrent WebUI API via the list, info, and trackers commands.
Boundary markers: Absent. The SKILL.md does not define any delimiters or provide instructions to the agent to ignore embedded commands within the returned data.
Capability inventory: The skill possesses the capability to execute shell commands (scripts/qbit-api.sh), write to the local filesystem (session cookies in /tmp/), and perform network operations via curl.
Sanitization: Absent. The scripts do not escape or filter the strings retrieved from the API before returning them to the agent's context.

qbittorrent