The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process potentially malicious code from third-party repositories. It implements a strong defense-in-depth strategy to mitigate this risk.
Ingestion points: Source code, documentation, and metadata from local directories or remote Git URLs.
Boundary markers: The skill uses a structured 'Evidence Ledger' with explicit status labels (Observed, Inferred, Unknown) and isolated 'evidence cards' to encapsulate repository excerpts.
Capability inventory: The skill is granted capabilities for filesystem reads, remote Git clones, temporary filesystem writes, and artifact verification via browser automation.
Sanitization: A comprehensive 'Artifact Contract' mandates strict HTML and SVG sanitization, specifically removing scripts, event attributes, and foreign objects, while enforcing 'strict' security levels for diagram rendering.
[EXTERNAL_DOWNLOADS]: The skill performs remote Git clones to analyze repositories. These operations are limited to temporary staging directories and utilize shallow clones to minimize the local footprint and exposure.
[COMMAND_EXECUTION]: The skill manages the execution of Git commands and potentially uses browser automation and Mermaid rendering engines during the creation and verification of its educational artifacts. These executions are strictly controlled by the skill's internal workflow rules and targeted at the sanitized output.

repo-tutor