The Agent Skills Directory

[COMMAND_EXECUTION]: The skill mandates the execution of a curl shell command to a local port (8888) for voice notifications. The command template in SKILL.md uses placeholders (WORKFLOWNAME, ACTION) which are susceptible to shell command injection if the agent populates them with unvalidated user input.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through the loading of user-defined customizations from ~/.claude/skills/PAI/USER/SKILLCUSTOMIZATIONS/Aphorisms/. The instructions explicitly state that these files "override default behavior," allowing malicious configurations to hijack the agent.
Ingestion points: ~/.claude/skills/PAI/USER/SKILLCUSTOMIZATIONS/Aphorisms/PREFERENCES.md and related files.
Boundary markers: Absent.
Capability inventory: curl (shell), Edit (file-write), Read (file-read), WebSearch, WebFetch.
Sanitization: Absent.
[PROMPT_INJECTION]: The FindAphorism.md workflow ingests untrusted content from external URLs via WebFetch. This content is processed without boundary markers or sanitization, creating a surface for indirect prompt injection.
Ingestion points: User-provided URLs via WebFetch.
Boundary markers: Absent.
Capability inventory: curl (shell), Edit (file-write), Read (file-read), WebSearch, WebFetch.
Sanitization: Absent.
[EXTERNAL_DOWNLOADS]: The skill utilizes WebSearch and WebFetch tools to retrieve external data for quote verification, research, and newsletter content analysis.

Aphorisms