The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of the @sigcli/cli package from the NPM registry to function. While this is a standard dependency for the tool, it introduces a supply chain dependency on the @sigcli scope.
[COMMAND_EXECUTION]: The skill facilitates the execution of arbitrary commands through sig run, which injects authentication credentials into the subprocess environment. This is a core feature but allows the agent to execute shell commands with elevated API access.
[DYNAMIC_EXECUTION]: The configuration format supports a validateRule field, which executes JavaScript expressions at runtime. These expressions are evaluated against data received from external HTTP responses (res.body, res.status). This allows for dynamic code execution triggered by external server content.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external websites and APIs through sig proxy and sig request. There is a risk that maliciously crafted responses (e.g., hidden instructions in JSON or HTML comments) could be passed to the AI agent's context. The analysis found:
Ingestion points: Data enters via sig request output and sig proxy traffic interception.
Boundary markers: None identified in the prompt templates to distinguish between authenticated data and instructions.
Capability inventory: The skill has the capability to write to the file system (sig init), execute shell commands (sig run), and perform network operations (sig request).
Sanitization: No explicit sanitization or filtering of external content is mentioned before the data is returned to the agent.

sigcli-auth-proxy