The Agent Skills Directory

[PROMPT_INJECTION]: No evidence of prompt injection or instructions designed to bypass agent safety filters was found. The skill uses prescriptive language to define design standards for CLI tools.
[CREDENTIALS_UNSAFE]: No hardcoded credentials, API keys, or secrets were detected. The skill specifically includes a rule ([input-stdin-for-secrets.md]) teaching against passing secrets as flag values to prevent leakage into logs and process tables.
[DATA_EXFILTRATION]: No suspicious network operations or file access patterns indicative of data exfiltration were found. Network examples (e.g., using curl) target well-known domains like api.example.com in instructional contexts.
[REMOTE_CODE_EXECUTION]: The skill provides static code examples in various languages (Python, TypeScript, Go, etc.) for instructional purposes. It does not perform any remote code execution or download untrusted scripts to the agent's runtime environment.
[COMMAND_EXECUTION]: While the skill discusses command execution as a concept for building CLIs, it does not include any malicious subprocess spawning or arbitrary command execution patterns within its own logic.
[OBFUSCATION]: No obfuscation techniques, such as hidden Unicode characters or multi-layer Base64 encoding, were identified. The use of ANSI escape sequences for color formatting in examples is legitimate and well-documented.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for reviewing user-provided code (Cat 8c/8d). However, it provides clear structured guidance and does not introduce exploitable capabilities that would escalate this risk. All examples follow best practices for sanitization and non-interactive safety.

cli-for-agents