The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external websites (HTML, robots.txt, sitemaps), creating an attack surface for indirect prompt injection.
Ingestion points: External content is fetched using curl and written to an evidence directory in references/intake.md.
Boundary markers: The instructions in SKILL.md and references/edge-cases.md explicitly command the agent to treat fetched bytes as untrusted data and never execute instructions found within them. Additionally, references/merge-format.md mandates stripping embedded instructions from data before report generation.
Capability inventory: The skill has access to networking (curl), repository management (git), and dependency management (asm).
Sanitization: Content is parsed using the Python standard library with size and timeout constraints, and the agent is instructed to validate target URLs.
[COMMAND_EXECUTION]: The skill uses various command-line utilities to perform its orchestration and auditing tasks.
Evidence: It uses curl to fetch web assets, git (stash, fetch, pull) to synchronize repository state, and asm (agent-skill-manager) to manage and acquire sub-skill dependencies. These tools are used for their intended purposes with appropriate quoting and error handling.
[EXTERNAL_DOWNLOADS]: The skill retrieves metadata and site assets from user-provided URLs.
Evidence: references/intake.md fetches robots.txt, sitemap.xml, and llms.txt. It also facilitates the use of the external service isitagentready.com through a sub-skill, but only after the user approves the action via a G1 consent gate.
[DYNAMIC_EXECUTION]: The skill executes small, static Python scripts to assist with data parsing.
Evidence: In references/intake.md, python3 is used in isolated mode (-I) to extract URL origins and parse HTML header information using standard library modules. The scripts are provided as fixed templates within the skill's reference files.

search-optimizer