The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/inspect-nats.sh uses a shell heredoc to dynamically generate and execute a Python snippet. This is used for a standard network handshake with a NATS broker and follows a safe, hardcoded template pattern.
[INDIRECT_PROMPT_INJECTION]: The skill involves reading and exporting various project files (site profiles, machine profiles, and deployment records) which are populated during the installation process. This creates a theoretical surface for indirect prompt injection from processed data.
Ingestion points: Project configuration files and logs including site-profile.yaml, machine-profile.yaml, and alignment records.
Boundary markers: Instructions explicitly require the agent to 'Redact secrets' and verify evidence at defined 'readiness gates' before moving to subsequent layers.
Capability inventory: Access to docker, puda CLI for hardware interaction, and shell execution for system inspection across multiple scripts.
Sanitization: Instructions include explicit mandates for secret redaction and manual verification of evidence and physical results.

puda-deployment