The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines a structured workflow (the 'ribbon') that includes building, proving, and merging code. This involves standard development activities such as executing shell commands, managing Git repositories, and running live tests or benchmarks, all of which are consistent with the skill's stated purpose of automating development cycles.
[DATA_EXFILTRATION]: The skill mentions using Slack for status reporting when running in autonomous mode. This is a legitimate notification feature for project management and does not indicate unauthorized data movement.
[PROMPT_INJECTION]: The skill's design involves generating a 'chain doc' with instruction blocks ('prompts') that are executed in subsequent steps. This creates a surface for indirect prompt injection. Ingestion point: .cascade/LOOP_CHAIN_.md. Boundary markers: No explicit delimiters for prompt interpolation. Capability inventory: Git operations, shell execution, and Slack reporting. Sanitization: No explicit sanitization of the chain doc content is mentioned. The risk is minimized by the skill's internal logic and instructions to keep the evidence directory ignored by version control.

cascade