The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands to interface with system utilities and project tools.
Evidence: Utilizes git, gh, curl, magick/convert, ffmpeg, and gifski to analyze repositories and process images.
Evidence: Executes package managers including npm, brew, apt-get, and dnf for dependency management, with explicit instructions to seek user confirmation for commands requiring sudo.
[EXTERNAL_DOWNLOADS]: The skill fetches dependencies and external utilities from remote sources.
Evidence: Installs the agent-browser package from the NPM registry.
Evidence: References the share-file utility from the author's repository at github.com/Vesely/skills for image hosting.
[DATA_EXFILTRATION]: The skill manages the capture and transmission of screenshots of the user's local application.
Evidence: Captures visual data via agent-browser screenshots and GIF recordings.
Evidence: Uploads media to external services through configured uploaders or GitHub repository branches, supported by privacy constraints and consent gates.
[PROMPT_INJECTION]: The skill processes repository data which represents a potential indirect injection surface.
Ingestion points: git diff, git log, and git status output accessed in SKILL.md.
Boundary markers: The skill lacks explicit delimiters or instructions to ignore embedded content within the diffs it processes.
Capability inventory: Includes shell command execution, package installation, and browser automation.
Sanitization: No specific filtering or escaping is applied to git output before it is used to determine UI capture targets.

pr-handoff