The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the open command (or its platform equivalent) to launch a default web browser and display a dynamically generated HTML file (~/.claude/strategy-view.html).
[PROMPT_INJECTION]: The skill modifies the agent's persistent configuration by appending a pointer to ~/.claude/CLAUDE.md. This file governs the agent's behavior across sessions, making it a target for persistent instruction injection.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its data processing cycle.
Ingestion points: Untrusted data enters the agent context via conversational capture (MODE 2) and is stored in ~/.claude/strategy.md.
Boundary markers: None identified; data is stored as markdown lists/paragraphs without isolation or security delimiters.
Capability inventory: The skill possesses file-write access to sensitive directories and the ability to execute system commands (open).
Sanitization: None; the instructions explicitly mandate preserving original user language and quotes verbatim.
[DATA_EXFILTRATION]: The skill accesses and manages files within the ~/.claude/ directory. This is a sensitive location typically housing the agent's internal configuration, instructions, and operational state.
[EXTERNAL_DOWNLOADS]: The visualization template (strategy-view-template.html) fetches styling assets from well-known Google font services (fonts.googleapis.com and fonts.gstatic.com).

strategy-awareness