The Agent Skills Directory

[COMMAND_EXECUTION]: The SKILL.md file contains multiple dynamic context markers (!command) that execute shell commands immediately upon loading the skill. These commands access environment variables like CLAUDE_PLUGIN_DATA and PLAYWRIGHT_BROWSERS_PATH to verify the local environment.
[REMOTE_CODE_EXECUTION]: The setup-deps.mjs script performs automated software installation using npm install and playwright install chromium, which downloads and executes binaries and scripts from external registries.
[COMMAND_EXECUTION]: The skill frequently executes system-level commands through spawnSync, including git and gh for repository management, and ffmpeg and magick for video processing and frame analysis.
[EXTERNAL_DOWNLOADS]: The download-resources.js script fetches ZIP files, PDFs, and other course assets from various external domains and Content Delivery Networks (CDNs), such as teachablecdn.com and mux.com.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and processes transcripts and companion code repositories from external course platforms. An attacker controlling these sources could embed malicious instructions to influence the agent's analysis or output.
Ingestion points: extraction/extract-course.js (transcripts), extraction/analyze-code-repo.js (companion repository files).
Boundary markers: Present. The skill utilizes markdown headers and structured templates (e.g., reference/analysis-template.md) to organize its findings.
Capability inventory: Extensive use of spawnSync in extraction/extract-course.js, extraction/analyze-code-repo.js, and extraction/setup-deps.mjs to run system tools (git, ffmpeg, npm). File system operations and network requests are performed throughout the extraction process.
Sanitization: Absent. There are no explicit sanitization routines or instruction-ignoring delimiters applied to content fetched from external platforms before it is processed by the LLM.
[CREDENTIALS_UNSAFE]: The skill manages user authentication by reading credentials from environment variables (COURSE_EMAIL, COURSE_PASSWORD, TEACHABLE_EMAIL, TEACHABLE_PASSWORD) and storing persistent session cookies in JSON files within the plugin's data directory.
[DATA_EXFILTRATION]: The skill performs extensive network operations to fetch course data, transcripts, and video segments from platforms including Dometrain, Teachable, Hotmart, and Mux.

course-digest