The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external data from the ./case_files/ directory, creating a surface for indirect prompt injection. 1. Ingestion points: The agent reads documents from the input directory for granular review and metadata extraction as described in references/ralphing_skill.md. 2. Boundary markers: The skill employs strict XML schemas (e.g., <fact> tags) defined in references/definitions.md to isolate extracted data, which helps mitigate but does not eliminate injection risks. 3. Capability inventory: The skill has the ability to read from the filesystem, write analysis results to disk, and execute the agent-cli tool. 4. Sanitization: There is no evidence of explicit sanitization or filtering of the input file content to remove potential embedded instructions.
[COMMAND_EXECUTION]: The skill utilizes a bash orchestrator script (ralph_loop.sh) to automate the repetitive execution of the agent. This script invokes the agent-cli command to process files in a stateless loop, which is a core architectural feature of the provided workflow.

case-file-analyzer