The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references the official LearnLM Partner Prompt Guide from Google LLC and Creative Commons license documentation. These links point to official documentation and do not involve executable code downloads.
[PERSISTENCE]: The skill includes instructions to maintain a learning record across sessions using the ~/.learnlm/ directory. The instructions explicitly require user authorization before any filesystem writes and mandate the exclusion of sensitive data, such as secrets or private text.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and tutor based on external sources, including user-provided documents and codebases. This ingestion surface is subject to indirect prompt injection. Mitigation steps are provided in the instructions, such as establishing boundaries for source material and separating recall from evaluation.
Ingestion points: Learner-supplied files, images, and external documentation processed via source-grounding.md.
Boundary markers: The agent is instructed to establish a 'source boundary' to separate provided material from supplemental knowledge and to mark outside context naturally.
Capability inventory: Instructions for file system access to manage progress records and the processing of various external content types (text, code, images).
Sanitization: The skill relies on linguistic labeling and contextual boundaries defined within the prompt instructions rather than programmatic sanitization.
[NO_CODE]: The skill consists entirely of configuration and instructional Markdown files. It does not contain any executable scripts, binary files, or external package dependencies.

learnlm-inspired-tutor