The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data provided by the user, specifically Tinder bios, profile prompt answers, and chat history. A malicious profile could attempt to inject instructions to influence the agent's coaching or modify local memory.
Ingestion points: Profile screenshots, bio text, prompt answers, and pasted chats (SKILL.md).
Boundary markers: Absent. The instructions do not define clear delimiters or warnings for the agent to ignore instructions embedded in the external text.
Capability inventory: The skill can read and write persistent files in the ~/Clawic/data/tinder/ directory (SKILL.md, memory-template.md). It does not have network access or binary execution capabilities.
Sanitization: Absent. There is no evidence of filtering or sanitizing the input data.
[PROMPT_INJECTION]: Instructions in safety-boundaries.md and setup.md regarding information disclosure were reviewed. These guidelines focus on safety (preventing user deception) and user experience (minimizing technical storage details) rather than malicious concealment.
[PERSISTENCE]: The skill maintains persistent state through local files in ~/Clawic/data/tinder/. The design includes a requirement for explicit user confirmation before writing or updating these notes, ensuring user oversight of stored data.

Tinder