The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) due to its core functionality of ingesting and processing untrusted external data.
Ingestion points: Step 1 in SKILL.md describes fetching and analyzing content from user-provided website URLs.
Boundary markers: None. The instructions do not specify any delimiters or safety prompts to prevent the agent from obeying instructions potentially hidden in the scraped website content.
Capability inventory: The skill utilizes Read, Write, and Bash tools, allowing it to modify the local environment and execute shell commands.
Sanitization: None. The skill extracts raw text (headlines, copy, CTAs) and directly interpolates it into new SKILL.md templates.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to retrieve content from external domains.
Evidence: Step 1 involves fetching page content for analysis. While targeting user-provided URLs, this represents a network ingestion surface for untrusted data.
[COMMAND_EXECUTION]: The skill uses the Bash tool and performs dynamic generation of executable agent instructions.
Evidence: The skill is configured with allowed-tools: Bash. Steps 2, 3, and 4 involve generating code for new SKILL.md files (brand-voice, ICP, and landing-pages) which are then written to the .claude/skills/ directory. These files define future agent behavior, effectively acting as runtime-generated scripts.

auto-setup