The Agent Skills Directory

[PROMPT_INJECTION]: The skill documentation and scanning scripts contain various prompt injection strings such as 'ignore previous instructions' and 'reveal the system prompt'. These are used exclusively as detection patterns and educational examples for the security scanner and do not represent a malicious attempt to override agent behavior.
[COMMAND_EXECUTION]: The clawdefender.sh script includes an array of dangerous shell command patterns, including fork bombs and reverse shells. These patterns are used for signature-based scanning of other scripts in the workspace and are not executed by the skill itself.
[EXTERNAL_DOWNLOADS]: The safe_install command uses npx clawhub install to fetch and install skills from the platform's repository. This is an integrated feature for managing agent capabilities and includes a post-installation security scan of the downloaded content.
[DATA_EXFILTRATION]: The auditing module checks the file permissions of sensitive workspace files like .env to ensure they are properly secured with restricted permissions. This is a standard defensive check for workspace auditing.
[INDIRECT_PROMPT_INJECTION]: The skill is specifically designed to handle untrusted data from external sources (APIs, emails) to detect malicious prompts. It provides boundary markers to wrap suspicious content, preventing the agent from inadvertently following instructions found in external data.

clawdefender