The Agent Skills Directory

[SAFE]: The skill uses local development tools (git, jq) and a task manager (bd) to inspect the current project state. All operations are confined to the local file system and repository metadata.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from local files (ralph-plan.md and referenced plan documents) and passes them to a subagent context, which presents a surface for indirect prompt injection.
Ingestion points: The file .llmtmp/ralph-plan.md and variable file paths parsed from the plan documents: line within that file.
Boundary markers: Absent. The skill does not use specific delimiters or instructions to the subagent to ignore potentially malicious content within the ingested files.
Capability inventory: The skill executes local shell commands using git, bd, and jq for task management and repository inspection. It has no built-in network or credential access capabilities.
Sanitization: None. The skill reads file content and relays it verbatim to the subagent's context without filtering or escaping.

ralph-review