The Agent Skills Directory

[COMMAND_EXECUTION]: The provided bash scripts (setup-ralph-project.sh, chunk-conversation.sh, and validate-todo.sh) utilize standard shell commands such as mkdir, cat, grep, sed, and chmod to manage project file structures, process text chunks, and set file permissions for local execution.
[EXTERNAL_DOWNLOADS]: The documentation references the installation of the ralph-wiggum-loop skill from the vendor's repository (wojons/skills) as a requirement for the full implementation of the automated build loop.
[PROMPT_INJECTION]: The skill describes an agentic workflow that is susceptible to indirect prompt injection due to its reliance on processing external, potentially untrusted data.
Ingestion points: Agents process data from conversation.txt, SPEC.md, and TODO.md which are sourced from earlier AI-driven steps.
Boundary markers: Absent. The prompt templates provided for the Builder, Verifier, and Planner agents do not include explicit delimiters or instructions to ignore embedded commands in ingested data.
Capability inventory: Agents are empowered to write project files, update task lists, and modify their own configuration files (e.g., MEMORY_RULES.md).
Sanitization: Absent. External content is used to build context without validation or escaping mechanisms.
[COMMAND_EXECUTION]: The setup-ralph-project.sh script dynamically generates prompt files and a placeholder Python script. Additionally, the methodology encourages agents to 'self-expand' their memory bank by updating their own operational rules and decision logs based on project history.

ai-collab-dev