The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions in SKILL.md mandate that the agent fetch and integrate updates from the author's GitHub repository (https://github.com/olafBobryk/averlo-tool-code-replay.git) before every use. Additionally, the setup process downloads the Chromium browser via Playwright and standard Node.js dependencies from the NPM registry.
[COMMAND_EXECUTION]: The skill executes multiple shell commands to manage its lifecycle and perform rendering, including npm ci, npm run setup, npm run doctor, and ffmpeg. The scripts/render-replay.mjs file spawns a local VS Code web server and manages subprocesses to capture video frames.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes arbitrary source code from a user's workspace.
Ingestion points: The agent reads the contents of the target file and explores the wider workspace provided via the --source-workspace argument to plan and render the replay.
Boundary markers: None. The content of the user's source code is read directly into the agent's context without delimiters or instructions to ignore embedded commands.
Capability inventory: The skill has the ability to execute shell commands (spawnSync in render-replay.mjs), perform file system operations (read/write/copy in workspace.mjs), and conduct network operations (Git updates as directed in SKILL.md).
Sanitization: While scripts/workspace.mjs implements a copySourceWorkspace function that attempts to exclude sensitive directories (like node_modules, secrets, and .git), it does not sanitize or validate the actual content of the files it processes for rendering.

averlo-code-replay