The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted code from external sources, which creates a vulnerability surface.
Ingestion points: The skill reads and analyzes HTML source code, JavaScript bundles, and inline scripts from target applications as described in SKILL.md.
Boundary markers: The instructions do not define delimiters or provide guidance to the agent on how to ignore potentially malicious instructions embedded within the code being scanned.
Capability inventory: The agent is instructed to write findings, including secret data, to multiple local files such as .sb-pentest-context.json, .sb-pentest-audit.log, and the .sb-pentest-evidence/ directory.
Sanitization: The skill lacks requirements for sanitizing or validating external content before it is processed or used to generate findings on the file system.
[CREDENTIALS_UNSAFE]: The primary function of the skill is to identify and extract Supabase service keys. It explicitly directs the agent to store these sensitive credentials, including decoded JWT payloads and key fragments, in local files.
[PROMPT_INJECTION]: The skill uses high-urgency, imperative language to dictate agent behavior, including phrases like "CRITICAL: PROGRESSIVE FILE UPDATES REQUIRED," "You MUST write to context files," "This is not optional," and "FAILURE TO UPDATE CONTEXT FILES IS NOT ACCEPTABLE." These are patterns used to override default agent constraints.
[EXTERNAL_DOWNLOADS]: The skill references documentation and API endpoints from well-known services including supabase.com and supabase.co for providing remediation steps and testing instructions.

supabase-extract-service-key