The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines an attack surface where instructions processed from a user's repository could influence agent behavior.
Ingestion points: The target command-line argument in scripts/pipeline_generator.py, scripts/terraform_scaffolder.py, and scripts/deployment_manager.py designates external directories for processing.
Boundary markers: The provided script templates lack delimiters or specific instructions to ignore embedded prompts within the files they are intended to analyze.
Capability inventory: While the provided scripts are currently skeletons, the SKILL.md instructions guide the agent to perform sensitive operations like aws ecs update-service, kubectl patch, and terraform apply based on the output of these tools.
Sanitization: No sanitization or validation logic is present in the boilerplate Python scripts to handle malicious content in the target paths.
[METADATA_POISONING]: There is a discrepancy between the documented usage and the actual implementation of the scripts.
Evidence: SKILL.md documentation describes arguments like --platform, --stages, and --provider for the scripts, but the provided source code in scripts/ only implements basic target, --verbose, and --json arguments. This inconsistency could lead an agent to attempt invalid command executions.

senior-devops