The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill attempts to install an external package globally and download browser binaries at runtime.
Evidence: npm install -g agent-browser && npx playwright install chromium in SKILL.md is triggered if the tool is missing.
[COMMAND_EXECUTION]: The skill executes shell commands to automate browser research and system-level package management.
Evidence: Multiple instances of agent-browser commands and an npm install command in the 'Research Mode' section of SKILL.md.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection by fetching and parsing text content from social media search results and user profiles.
Ingestion points: agent-browser get text body commands in SKILL.md which retrieve content from TikTok, Instagram, and YouTube.
Boundary markers: Absent. There are no delimiters or instructions to ignore embedded commands in the retrieved data.
Capability inventory: The agent has access to shell execution (agent-browser, npm), file reads (./brands/), and network access via the browser tool.
Sanitization: Absent. The skill does not describe any filtering or validation of the fetched HTML/text content before use.

marketing-influencer