The Agent Skills Directory

Command Execution: The skill interacts with the local environment by executing pac copilot commands. It uses the Bash(pac *) tool to perform operations like listing and cloning agents. The instructions include detailed safety protocols to prevent command injection, such as single-quoting all shell arguments and rejecting values containing control characters.
Indirect Prompt Injection Surface: The skill processes external data, including user-supplied IDs, schema names, and URLs which could potentially contain malicious instructions.
Ingestion points: Untrusted data enters the agent context via $ARGUMENTS and Copilot Studio URLs provided by the user.
Boundary markers: The instructions explicitly define a section for 'Passing values safely,' which mandates the rejection of suspicious characters and requires specific parsing logic for URLs to ensure only expected path segments are used.
Capability inventory: The skill possesses the capability to run shell commands (pac), read files, and search using Glob and Grep.
Sanitization: The instructions provide clear requirements for parameter escaping, validation of URL path segments, and a strict rule against running commands or scripts found within user-provided text.

clone-agent