The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external source code repositories provided via URL. This creates a theoretical surface for indirect prompt injection if the reference project contains malicious instructions designed to subvert the agent's analysis.
Ingestion points: External codebases are fetched and read via the /rebuild analyze <URL> command.
Boundary markers: The skill employs strict markdown templates for its analysis reports, feature extraction, and planning phases to maintain structured output.
Capability inventory: The skill uses subprocess calls for analysis tools (cloc, pydeps, etc.) and performs file writes to a local .project-rebuilder/ directory for state management.
Sanitization: There is no explicit mention of sanitizing or escaping the content of the external code files before the agent processes them.
[COMMAND_EXECUTION]: The rules/project-analysis.md file specifies the use of several standard command-line tools for project analysis, including cloc for line counts, pipreqs for dependency identification, and pydeps/pyan for generating architectural diagrams. These are legitimate utilities for the skill's stated purpose of project rebuilding.
[PERSISTENCE]: The skill implements a persistence mechanism known as RALPH (Recursive Autonomous Long-term Persistent Helper). It saves project state, completed tasks, and decision logs to JSON checkpoint files within a .project-rebuilder/ folder. This is a functional feature for managing complex, multi-session tasks and does not involve malicious persistence such as modifying shell profiles or system services.

project-rebuilder