The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends the installation of the agbrowse tool from the NPM registry.
Evidence: Documentation mentions npm install -g agbrowse for standalone use.
[COMMAND_EXECUTION]: The workflow relies on executing various shell commands for browser control, JSON processing, and archive management.
Evidence: Commands include cli-jaw, agbrowse, jq, and unzip.
[DATA_EXFILTRATION]: The skill is designed to package and upload local project files and source code to external AI services such as ChatGPT, Gemini, and Grok.
Evidence: Uses --context-from-files and --context-file to transmit data to AI providers via browser sessions.
[PROMPT_INJECTION]: The skill ingests local file content as context for AI models, which presents a surface for indirect prompt injection attacks.
Evidence:
Ingestion points: Local files and directories specified via --context-from-files in SKILL.md.
Boundary markers: Employs a structured prompt shape with [USER], ## Context, and ## Question delimiters.
Capability inventory: Capable of writing files to the local system (zip artifacts) and performing browser mutations via cli-jaw.
Sanitization: There is no explicit mention of sanitizing or escaping the content of ingested files.

web-ai