The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs a tool from an unverified Homebrew tap.
Evidence: brew install silver-flight-group/tap/kakaocli in SKILL.md.
[COMMAND_EXECUTION]: The skill executes multiple commands using the downloaded kakaocli binary.
Evidence: Use of kakaocli status, kakaocli auth, and kakaocli send throughout SKILL.md.
[DATA_EXFILTRATION]: The skill instructs users to grant 'Full Disk Access' and 'Accessibility' permissions to the terminal, enabling the tool to read local messaging databases and interact with the UI. This creates a risk where sensitive data could be accessed or potentially exfiltrated.
Evidence: Instructions in step 2 of SKILL.md for 'Full Disk Access' and 'Accessibility'.
[PROMPT_INJECTION]: The skill ingests untrusted data from KakaoTalk chat rooms, which could contain malicious instructions designed to manipulate the AI agent.
Ingestion points: Reading messages via kakaocli messages and kakaocli search (found in SKILL.md).
Boundary markers: Absent. The skill does not instruct the agent to use delimiters or ignore instructions within the retrieved text.
Capability inventory: The agent can send messages to other users via kakaocli send (found in SKILL.md).
Sanitization: Absent. No validation or escaping is applied to the retrieved message content before processing.

kakaotalk-mac