The Agent Skills Directory

[PROMPT_INJECTION]: The skill explicitly directs the agent to follow 'coding level guidelines (levels 0-5)' if they were 'injected at session start'. This is a direct instruction to prioritize potentially malicious content provided by a user or third party over the skill's own logic or safety guardrails.
[PROMPT_INJECTION]: The skill is vulnerable to Indirect Prompt Injection due to its extensive use of external data sources. 1. Ingestion points: Content retrieved via WebSearch, docs-seeker, and ai-multimodal tools. 2. Boundary markers: Absent; there are no instructions to use delimiters or ignore instructions within the fetched data. 3. Capability inventory: The skill can execute database queries via psql and trigger planning workflows via /plan commands. 4. Sanitization: Absent; no validation or filtering is performed on the ingested content before it influences agent behavior.
[COMMAND_EXECUTION]: The skill utilizes the psql command to inspect database structures. While intended for architectural review, this capability provides a path for data exposure if the agent is manipulated via prompt injection to leak schema or data.
[EXTERNAL_DOWNLOADS]: The skill uses WebSearch and docs-seeker to fetch information from unknown third-party websites, introducing untrusted content into the execution context.

brainstorming