The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides explicit instructions for the agent to utilize shell commands to interact with local project tools within the workspace.
Evidence: The 'Codex adapter note' instructs the agent to use 'shell commands for local tools' when the source instructions mention generic tool surfaces.
Evidence: The instructions reference a specific local directory ~/ai-berkshire/tools/ for tool access.
[SAFE]: The core content of the skill is an educational and roleplay-based prompt focusing on investment principles and corporate culture.
The instructions explicitly discourage recommending specific buy/sell actions or providing precise price targets.
No evidence of obfuscation, credential harvesting, or attempts to bypass system safety guidelines was found.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for processing user-supplied data through functional tools, though it does not contain malicious patterns itself.
Ingestion points: User requests are passed to the agent via the $ARGUMENTS variable.
Boundary markers: None present to distinguish user input from instructions during tool interpolation.
Capability inventory: The skill allows access to shell commands (Bash), web search, and file read/write operations.
Sanitization: No explicit sanitization of $ARGUMENTS is defined within the skill text.

dyp-ask