The Agent Skills Directory

[COMMAND_EXECUTION]: The skill drives the wsh command-line utility to interact with the Wave Terminal workspace. The scripts/wave_delegate.py script uses subprocess.run to invoke wsh commands for creating blocks, sending keys, and reading scrollback.
[COMMAND_EXECUTION]: The scripts/wave_delegate.py script generates shell commands to be executed within Wave Terminal blocks. It utilizes shlex.quote to sanitize user-provided prompt inputs before they are interpolated into the final shell command, which prevents prompt-based command injection.
[PROMPT_INJECTION]: The skill implements a pattern for indirect prompt injection by reading data from external terminal blocks.
Ingestion points: Terminal output is ingested from other blocks using wsh termscrollback within the scripts/wave_delegate.py script and documented in references/output-to-wave.md.
Boundary markers: There are no specific delimiters or instructional guardrails applied to the content read from terminal scrollback before it is processed by the agent.
Capability inventory: The skill has access to extensive system capabilities via wsh, including arbitrary command execution (wsh sendkeys), file system operations (wsh file), and network requests (wsh web open).
Sanitization: Shell-level escaping is applied to outbound command arguments, but no linguistic filtering is performed on ingested terminal output.

waveterm