The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes local project files and configuration, which are subsequently used to inform the generation of other project artifacts.
Ingestion points: Reads project-specific data from beat/config.yaml, as well as manifest files such as package.json, Cargo.toml, and go.mod (Step 4). It also reads existing ADR and feature files.
Boundary markers: The references/config-schema.md defines the injection of context and rules from the user-provided configuration into future prompts but does not specify mandatory delimiters or security-focused instructions to isolate this data from the agent's core behavioral guidelines.
Capability inventory: The skill uses directory creation (mkdir) and file writing (Step 7) across multiple files including SKILL.md.
Sanitization: There is no indication of content validation or sanitization for the data read from project files before it is interpolated into subsequent prompts.
[EXTERNAL_DOWNLOADS]: The skill mentions and provides instructions for installing the superpowers plugin from an external GitHub repository (github.com/obra/superpowers).
[COMMAND_EXECUTION]: The skill utilizes shell commands to establish the project's directory structure.
Evidence: Uses mkdir -p to initialize the beat/, beat/changes/, and beat/features/ directories as part of the project setup workflow.

setup