The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests detailed user requirements and architectural descriptions in Phase 1 and utilizes this data to generate a technical decomposition and validation scripts. There are no instructions to sanitize or delimit this user-provided content, creating a risk that malicious instructions embedded in the requirements could influence the generation and execution of commands in subsequent phases.
Ingestion points: SKILL.md Phase 1 (functional/non-functional requirements, current architecture descriptions).
Boundary markers: Absent; there are no specific delimiters or warnings for the agent to ignore instructions embedded in the user data.
Capability inventory: SKILL.md Phase 3 (generates and runs curl/gcloud commands) and Phase 4 (writes code files to the workspace).
Sanitization: Absent.
[COMMAND_EXECUTION]: Phase 3 of the workflow involves generating and executing shell commands (specifically curl and gcloud) to perform validation checks on the proposed architecture. Although the skill requires the agent to obtain user permission before execution, this capability constitutes a significant command execution surface.
[DYNAMIC_EXECUTION]: The skill generates architectural documentation and code files based on user-provided context and architectural designs, which are then written to the local workspace in Phase 4. This runtime creation and deployment of executable artifacts is a form of dynamic execution.

google-cloud-solution-agentic-analytics-spark-knowledge-catalog