The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from third-party websites (fetched via isitagentready.com) which could potentially contain malicious instructions.
Ingestion points: Raw scan results are stored in .agent-ready/scan.json and .agent-ready/fixes.md (as documented in SKILL.md).
Boundary markers: The skill implements a dedicated 'Prompt Injection Boundary' section in SKILL.md, instructing the agent to treat scanned content as data rather than instructions, and uses four mandatory human approval gates (G1–G4) to verify actions.
Capability inventory: The skill can write files (agent-ready-plan.md) and delegate tasks to the /plan-to-issues skill for GitHub issue creation.
Sanitization: scripts/render_plan.py uses a clean() function that collapses whitespaces, escapes markdown table pipes (|), and strips leading hash characters (#) to prevent site content from forging markdown structure.
[EXTERNAL_DOWNLOADS]: The skill sends a user-provided URL to a third-party scanning service to evaluate the site.
Evidence: scripts/scan_site.sh uses curl to POST data to https://isitagentready.com/api/scan. This behavior is the primary purpose of the skill and is explicitly disclosed to the user at mandatory approval gate G1.
[COMMAND_EXECUTION]: The skill uses local shell and Python scripts to automate the scanning and triage process.
Evidence: scripts/scan_site.sh invokes curl and python3. It correctly passes the user-provided URL through an environment variable into a python3 -c call to avoid shell command injection via special characters in the URL.
[PROMPT_INJECTION]: A technical directive was detected in docs/README.md intended to guide the AI agent toward the correct instruction file.
Evidence: The README contains the comment DO NOT READ THIS FILE — This README.md is for human catalog browsing only. The runtime loader only reads SKILL.md.... This is a benign organizational instruction for the agent, not a malicious bypass or safety filter override.

website-agent-readiness