The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a persistent knowledge management system that ingests session context and stores it in instruction-linked files, creating a vulnerability surface where malicious session content could influence future agent actions.
Ingestion points: The agent scans the current session context for reusable knowledge as described in Step 2 of SKILL.md.
Boundary markers: Captured learnings are formatted as Markdown list items with trailing date-stamp comments (e.g., ).
Capability inventory: The skill uses the rota CLI to write to .rota/KNOWLEDGE.md and updates managed instruction blocks in AGENTS.md or CLAUDE.md.
Sanitization: The skill includes an opt-in verification process in verifier.md that dispatches a high-capability subagent to validate the durability, sharpness, and correctness of new entries.
[COMMAND_EXECUTION]: The skill directs the agent to interact with the project environment using a local CLI tool.
Evidence: Instructions in SKILL.md, topic-size.md, and umbrella-routing.md specify the use of the rota CLI (e.g., rota knowledge add, rota block knowledge, rota knowledge stats) to manage data. Session-derived data is interpolated into these shell commands, which carries a risk of command injection if not properly handled by the agent.

rota-learn