The Agent Skills Directory

[COMMAND_EXECUTION]: The skill contains a potential command injection vulnerability. It explicitly instructs the agent to capture a URL from the user and pass the raw string directly into a shell command (mktg init --from <url>) while specifically forbidding any validation, regex checking, or normalization. This pattern allows an attacker to execute arbitrary commands by embedding shell metacharacters in the provided URL string if the execution environment does not properly escape the arguments.
File: SKILL.md
Evidence: "Capture the URL exactly as the user provides it. Do NOT validate it inside the wizard — no regex, no protocol check, no normalization. Pass the raw string straight to the CLI: mktg init --from <url>"
[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection. It scrapes content from a user-provided URL and uses the resulting data to influence the behavior of multiple research agents. The instructions lack security boundaries and explicitly discourage input validation, which could allow malicious instructions on the external website to manipulate the agent's research phase.
Ingestion points: Website data scraped via mktg init --from <url> in SKILL.md.
Boundary markers: Absent. There are no instructions to use delimiters or ignore instructions within the scraped content.
Capability inventory: The skill spawns three sub-agents (mktg-brand-researcher, mktg-audience-researcher, mktg-competitive-scanner) and chains subsequent execution to /landscape-scan and /positioning-angles.
Sanitization: Absent. The skill specifically instructs the agent not to validate the external source input.
[EXTERNAL_DOWNLOADS]: The skill performs automated network operations to scrape data from arbitrary, user-provided URLs during the initialization process. While this is a core feature, it involves fetching untrusted content from the open web into the agent's context.
Evidence: Use of mktg init --from <url> to seed brand files from external websites.

mktg-setup