The Agent Skills Directory

[DATA_EXFILTRATION]: The skill implements a secure handling mechanism for sensitive data. It directs the agent to use local files with restricted permissions (chmod 0600) in the .tigerkit/ directory for secret input, effectively preventing credentials, tokens, or passwords from appearing in chat logs, markdown files, or environment variables. It also enforces Git exclusion checks using git check-ignore to ensure these temporary secret files are not accidentally committed to the repository.
[COMMAND_EXECUTION]: The skill utilizes standard development tools for safety verification. It instructs the agent to run git ls-files and git check-ignore to verify the security of the local environment. It also recommends using bash -n and shellcheck to statically validate any temporary helper scripts created for the user, ensuring they are free of syntax errors or common shell script vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill processes host evidence and UI labels which constitute an external data ingestion surface. The instructions mitigate potential indirect prompt injection by requiring the agent to strictly track internal state, use fixed JSON templates for data collection, and avoid interpreting untrusted data as instructions.
[EXTERNAL_DOWNLOADS]: Reference documentation suggests a default setup for MCP servers using npx -y chrome-devtools-mcp@latest. This involves fetching packages from the npm registry at the user's request for integration purposes, which is documented as a standard and transparent operation for establishing browser-based provider routes.

tk-wizard