The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides workflows that ingest untrusted web text, creating a vector where an attacker-controlled website could influence the agent's behavior.
Ingestion points: Workflows in references/competitive-research.md use the agent-browser tool to fetch text content from competitor landing pages and public ad libraries.
Boundary markers: The instructions do not define clear delimiters or use 'ignore' constraints to safely isolate the ingested external data from the skill's logic.
Capability inventory: The skill defines a environment with filesystem write access and shell execution capabilities, making it a target for malicious input from the web.
Sanitization: There is no mentioned process for sanitizing or filtering the web content before the agent processes it.
[COMMAND_EXECUTION]: The skill directs the agent to use the shell for environmental discovery and tool setup.
Evidence: references/shared-patterns.md instructs the agent to execute local scripts such as tool-discovery.sh and chrome-profiles.sh to verify its environment.
[EXTERNAL_DOWNLOADS]: The skill guides the user or agent to download and install external browser-based tools and dependencies.
Evidence: The setup instructions in references/shared-patterns.md describe using npm install and npx skills add to fetch tools from remote repositories.

paw-mkt-paid-ads