The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the puda CLI (e.g., puda machine list, puda init, puda protocol validate, puda protocol run) for project management and machine interaction. These tools are associated with the skill's authoring organization.
[DYNAMIC_EXECUTION]: The skill provides instructions for generating metadata using Python one-liners, such as python -c "import uuid; print(uuid.uuid4())" and datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ"). It also describes an automated interaction pattern using terminal(background=true) and process(action="submit", data="yes") to handle interactive confirmation prompts from the protocol runner.
[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface where it ingests untrusted data from external sources and incorporates it into the agent's context and user-facing output.
Ingestion points: Reads data from puda machine list, puda machine commands <machine_id>, and "previous protocol files."
Boundary markers: The instructions lack specific delimiters or warnings to ignore potentially malicious instructions embedded within the safety metadata retrieved from the machine catalog.
Capability inventory: The skill has capabilities to write files (protocols/<protocol_id>.json), execute shell commands via the puda CLI, and run Python code snippets.
Sanitization: The skill explicitly instructs the agent to "copy the structured metadata into the command unchanged" and "relay the complete displayed command and safety chunk to the operator," which could lead to the display of malicious instructions if the machine catalog data is compromised.

puda-protocol