The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill uses npx -y @anthropic-ai/claude-code in scripts/ask-claude.mjs and scripts/verify-setup.mjs to execute the Claude CLI. This pattern downloads and runs code from the NPM registry at runtime. The package is published by the @anthropic-ai organization, which is a trusted vendor.
[COMMAND_EXECUTION]: The script scripts/ask-claude.mjs manages subprocesses to interact with the Claude CLI.
It utilizes child_process.spawn with shell: false to execute the tool, mitigating common shell injection vectors.
On Windows, it spawns taskkill to handle process termination when timeouts occur.
[PROMPT_INJECTION]: The skill contains an indirect prompt injection surface due to how it interpolates user data.
Ingestion points: Untrusted prompt text is accepted via command-line arguments or stdin in scripts/ask-claude.mjs.
Boundary markers: The input is interpolated directly into the -p argument of the CLI command without the use of delimiters (such as XML tags) or instructions for the model to ignore potential instructions embedded within the data.
Capability inventory: The skill invokes the secondary Claude session with the --dangerously-skip-permissions flag, which allows the headless session to execute tools without user confirmation.
Sanitization: There is no escaping or validation of the input prompt before it is passed to the execution wrapper.
[EXTERNAL_DOWNLOADS]: The use of npx results in the download of the @anthropic-ai/claude-code package from a remote registry if it is not present locally. This is a functional requirement for the skill and targets a trusted source.

omega-claude-cli