The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/launch_script_writer.py tool generates a shell script (launch.sh). This script is designed for the user to execute manually to deploy agents using curl commands to official endpoints.
[PRIVILEGE_ESCALATION]: The scripts/launch_script_writer.py script applies execution permissions (chmod +x) to the generated launch.sh file. This is a standard procedure for utility scripts meant for manual user execution.
[DATA_EXPOSURE]: The skill uses environment variables to manage API credentials, preventing sensitive keys from being hardcoded or logged. The scripts/payload_validator.py utility scans generated JSON artifacts and the launch script to ensure no sk-ant- keys are present before deployment.
[EXTERNAL_DOWNLOADS]: The generated launch script initiates network connections to api.anthropic.com. This activity involves a well-known service and is fundamental to the agent deployment process.
[INDIRECT_PROMPT_INJECTION]: The scripts/payload_generator.py script constructs agent payloads from a JSON build sheet, representing an ingestion point for external data.
Ingestion points: The --sheet argument in payload_generator.py reads a local JSON build sheet.
Boundary markers: The generated JSON payloads do not use explicit delimiters to isolate untrusted input within constructed system instructions.
Capability inventory: The skill performs local file writing and modifies file permissions, while the generated script uses curl for network requests.
Sanitization: The validation process focuses on JSON structure and secret detection rather than content sanitization against prompt injection.

stage-launch