The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted Markdown files and GitHub issues from external sources to propose and apply changes to its own codebase.
Ingestion points: The scripts/harvest_lessons.py script reads files from docs/lessons/*.md in directories specified by the user, and the skill also fetches external GitHub issues via the gh CLI.
Boundary markers: The instructions tell the agent to "Show the user a summary before touching any skill file" and "Only apply changes after explicit confirmation", providing a human-in-the-loop checkpoint.
Capability inventory: The agent has the capability to write to the file system (modifying SKILL.md and reference files) and potentially commit changes via git.
Sanitization: There is no evidence of sanitization or filtering of the content within the external files or issues to prevent embedded instructions from influencing the agent's behavior during the "harvesting" process.
[COMMAND_EXECUTION]: Executes a bundled Python script scripts/harvest_lessons.py to parse lesson files and the GitHub CLI (gh) to fetch lesson-related issues from the repository ronjunevaldoz/kmp-agent-skills.

kmp-skill-harvester