The Agent Skills Directory

Indirect Prompt Injection Surface: The skill processes untrusted data from external sources, creating a potential surface for indirect prompt injection.
Ingestion points: The agent reads content from GitHub PR comments, CI job summaries, and CI job logs to extract screenshot hashes (SKILL.md).
Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded commands within the ingested text.
Capability inventory: The agent has access to gh api, gh pr, curl, git, and python3 for image processing and repository updates.
Sanitization: There is no explicit sanitization or filtering of the content retrieved from GitHub before it is processed.
Context: The skill includes a manual verification step (image comparison) which serves as a functional mitigation against unintended updates.
External Resource Fetching: The skill uses curl to download image files from hediet-screenshots.azurewebsites.net (a Microsoft Azure domain) for comparison. These resources are treated as data for visual verification and are not executed.
Subprocess Command Execution: The skill utilizes shell commands (gh, curl, git) and a Python script using the Pillow library to perform its tasks. The use of specific, structured commands minimizes the risk of arbitrary command injection.

update-screenshots