The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains instructions for the agent to adopt an aggressive Japanese corporate persona ("tsume") that pressures itself to complete tasks. It explicitly instructs the agent to avoid common "polite" AI behaviors (like suggesting manual user intervention) in favor of relentless task ownership.
[INDIRECT_PROMPT_INJECTION]: The skill mandates high proactivity, requiring the agent to investigate related files (Y and Z) even if only X was requested, and to use tools before asking the user for permission. This behavior creates a significant surface for indirect prompt injection, as the agent may ingest and follow malicious instructions found in the data or files it searches autonomously.
Ingestion points: Task-related files, project logs, and web search results accessed during proactive investigations.
Boundary markers: No boundary markers or "ignore embedded instructions" warnings are provided.
Capability inventory: Access to shell commands (curl, build, test), file system reads, and web search tools.
Sanitization: No logic is present to sanitize or validate external content before the agent acts upon it.
[COMMAND_EXECUTION]: The skill explicitly pushes the agent to use terminal-based tools (curl, build, test, git) and command execution proactively, directing it to "act first, ask later" and use terminal output as the primary source of truth over user input.

pua-ja