The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to clone an external repository (github.com/disler/infinite-agentic-loop.git) to obtain configuration files and custom command definitions.
[INDIRECT_PROMPT_INJECTION]: The skill uses specification files (specs/*.md) to drive the behavior of parallel sub-agents, creating an attack surface for indirect instructions.
Ingestion points: Specification markdown files stored in the project's specs/ directory.
Boundary markers: None identified in the provided instructions; sub-agents appear to follow the specification content directly.
Capability inventory: The system is granted permissions for file writing, shell command execution, and recursive agent deployment.
Sanitization: No evidence of sanitization or instruction filtering for the content of specification files.
[COMMAND_EXECUTION]: The provided configuration (.claude/settings.json) explicitly enables allow_shell_commands and allow_file_operations, granting the agent broad access to the local environment to manage generation waves.
[DYNAMIC_EXECUTION]: The skill utilizes custom slash commands (/project:infinite) defined in external markdown files (.claude/commands/infinite.md) which execute logic to parse arguments and coordinate agent tasks at runtime.

infinite-agentic-loop-claude