The Agent Skills Directory

[DYNAMIC_EXECUTION]: The screenshot-before-after.sh script executes several blocks of JavaScript code by piping strings to node -. This is used for tasks such as JSON parsing (json_get_field, write_context), URL component encoding (encode_url_component), and server command detection logic.
[INDIRECT_PROMPT_INJECTION]: The skill implements a workflow that reads and modifies untrusted external data, creating an injection surface.
Ingestion points: Fetches existing pull request (PR) or merge request (MR) descriptions from GitHub or GitLab using the gh or glab CLI tools.
Boundary markers: Employs the ## Before & After Markdown header as a delimiter for content replacement.
Capability inventory: The skill can perform git commits, push to remote branches, update remote PR/MR metadata, and execute local shell commands.
Sanitization: The script performs regex-based replacement on the fetched text but does not sanitize or validate the input before re-uploading, allowing malicious descriptions to potentially influence the automated update process.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill performs expected data transfer and storage operations.
Network operations: Communicates with GitHub (github.com) and GitLab (gitlab.com) to push screenshot assets and update review descriptions.
Sensitive file access: Accesses the user's home directory to store operational state in ~/.config/opencode/skills/screenshot-skill/state/context.json.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill relies on external Node.js dependencies (puppeteer-core, minimist) and assumes the presence of system-level binaries like gh, glab, and chromium. The capture script launches Chromium with the --no-sandbox flag, which reduces the browser's security isolation.

git-host-rn-web-before-after