The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from conversation history and external feedback sources, which presents an attack surface for indirect prompt injection.\n
Ingestion points: SKILL.md identifies conversation-history and skill-feedback as consumed data types; references/chat-evidence.md provides specific instructions for extracting evidence from historical chat transcripts.\n
Boundary markers: The instructions do not define technical delimiters (such as XML tags or unique markers) to segregate untrusted user data from the agent's instructions, relying instead on high-level procedural guidance to identify different types of evidence.\n
Capability inventory: The skill possesses the capability to modify local files (local edits) and perform network-facing actions through the gh command-line interface.\n
Sanitization: references/feedback.md contains explicit instructions to manually redact credentials and private information from any feedback summary or issue report before submission.\n- [COMMAND_EXECUTION]: The skill uses the gh (GitHub CLI) tool for managing remote feedback and bug reports.\n
Evidence: references/feedback.md advises using the gh tool to create GitHub issues.\n
Note: This functionality is intended for use with the vendor's repository (agentjido/kata) and is protected by a requirement for explicit user opt-in before execution.

kata-maintain-skill