The Agent Skills Directory

[PROMPT_INJECTION]: The skill documentation and templates in references/system-prompt-and-guardrails.md contain common injection strings like 'ignore your rules' and 'reveal your prompt'. These are used exclusively as examples of adversarial behavior that the agent is explicitly instructed to detect and refuse. These patterns function as defensive guardrails rather than active injection attempts.
[CREDENTIALS_UNSAFE]: The scripts/verify.sh file includes regular expressions (e.g., sk-[a-z0-9]{8,}, api_key, bearer) used to scan other files for potential secrets during development. A placeholder key (sk-live-9f2...) is also present in SKILL.md within a 'Bad Example' code block intended to teach users to avoid hardcoding secrets. These are diagnostic and educational tools, not credential leaks.
[COMMAND_EXECUTION]: The skill includes a local utility script, scripts/verify.sh. This script is dependency-free and uses standard POSIX commands (grep, sed, printf) to perform read-only static analysis on user-provided text files. It does not execute remote code, access sensitive system paths, or perform network operations.

chatbot