The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest data from untrusted external sources, which creates a surface for potential indirect prompt injection attacks.
Ingestion points: The skill retrieves content from the open web using WebSearch, agent-browser (for web pages), and the GitHub API (for README and SKILL.md files).
Boundary markers: The instructions do not specify the use of clear delimiters or instructions to ignore embedded commands when the agent processes or summarizes the fetched external content.
Capability inventory: The skill has access to shell commands (bash), network downloads (curl), GitHub authentication, and browser automation.
Sanitization: There is no explicit requirement for sanitizing the text or code snippets retrieved from external URLs before they are presented in the generated HTML report or processed by the sub-agent.
[COMMAND_EXECUTION]: The skill makes extensive use of local command-line tools to interact with the filesystem and network.
Evidence: It executes curl to download image assets, gh (GitHub CLI) to pull repository content and metadata, and agent-browser for headless navigation and screen capture.
Context: These operations are core to the skill's purpose of capturing 'real artifacts' from software projects. While the tools are standard, they operate on URLs and repository identifiers derived from external search results.

case-radar