The Agent Skills Directory

[SAFE]: The skill consists of documentation, implementation patterns, and validation scripts designed to help developers build secure and efficient hooks. No malicious instructions or hidden behaviors were found.
[COMMAND_EXECUTION]: The skill provides numerous examples of Bash hooks used for lifecycle events. These include commands for logging execution metadata to local files (e.g., ~/.claude/audit.log) and performing input validation. These commands are benign instructional templates. Static analysis detections of fork-bomb patterns are false positives, as the code uses these strings as negative examples to be blocked by validation logic.
[DATA_EXFILTRATION]: Documentation includes patterns for logging tool inputs and outputs. The skill proactively addresses potential data exposure by providing specific code examples and regular expression patterns for redacting credentials (API keys, passwords, tokens) before they are written to local logs.
[REMOTE_CODE_EXECUTION]: The skill mentions a utility script hook_validator.py for verifying hook syntax. It also references standard Python libraries such as aiohttp for fetching configuration files. The examples use placeholder domains and follow non-blocking execution practices.
[INDIRECT_PROMPT_INJECTION]: The skill describes building hooks that intercept and modify user prompts and tool outputs, which is a recognized attack surface. However, the documentation focuses on defensive use cases like validation and provides advice on sanitizing external inputs to mitigate these risks.

hook-authoring