The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its core distillation process.\n
Ingestion points: The agent is instructed to read the entire content of uncached files (using the Read tool) identified as missing from the ledger.\n
Boundary markers: The instructions lack delimiters or safety warnings to distinguish between the file's data and the agent's instructions, specifically directing the agent to 'read deeply' for distillation.\n
Capability inventory: The agent possesses Bash and Write capabilities, enabling shell command execution and file system modifications if triggered by content within a summarized file.\n
Sanitization: No sanitization, escaping, or validation is performed on the content of the files before they are processed by the agent.\n- [COMMAND_EXECUTION]: The skill relies on the execution of local Python scripts to manage the summary cache.\n
Evidence: The skill commands the use of inventory.py and inject_summary.py located within the skill's internal directory structure to identify missing files and update the cache ledger.\n- [EXTERNAL_DOWNLOADS]: The skill references external AI models and services for handling large batches of summarization tasks.\n
Evidence: The documentation suggests delegating tasks to an 'agent swarm' using services like Copilot, Gemini, and Claude. These are well-known technology services and represent standard integration for high-throughput tasks.

rlm-distill-agent