The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The SKILL.md file contains a shell script intended for 'Release Artifact Verification'. This script downloads metadata, digital signatures, and public keys from the author's official GitHub repository (github.com/prompt-security/clawsec). These downloads are used to verify the integrity of the skill's package before installation.
[COMMAND_EXECUTION]: The verification script utilizes standard system utilities including curl, openssl, jq, and shasum to perform cryptographic validation of downloaded files. Additionally, the skill creates and manages local state and report files within the user's home directory (~/.clawsec/clawtributor/).
[INDIRECT_PROMPT_INJECTION]: As the skill is designed to ingest and report malicious activity, it naturally processes untrusted data (e.g., observed attack payloads). The skill mitigates this risk by instructing the agent to draft reports locally, require explicit human approval before any submission, and follow a privacy checklist to sanitize data.
[PROMPT_INJECTION]: Static analysis identified phrases associated with prompt injection, such as 'You are now in developer mode' and 'Disregard earlier safety constraints'. Manual review confirms these strings are documented examples of malicious behavior for the agent to identify and report, rather than instructions for the agent to execute.

clawtributor