The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The instructions in SKILL.md and the onboarding logic within publish.sh explicitly direct the agent to persist the IPFS_PUBLISH_TOKEN (a Bearer token) in plain text. This is done by modifying user-level configuration files such as ~/.zshrc, ~/.bashrc, or ~/.claude/settings.json, exposing sensitive credentials to any local process.
[COMMAND_EXECUTION]: The skill executes a bash script (publish.sh) that utilizes find for filesystem traversal and curl for network requests. Additionally, the skill's documentation encourages the agent to execute shell commands to modify system startup scripts, which is a common persistence mechanism used to maintain state or environment variables across sessions.
[DATA_EXFILTRATION]: The core purpose of the skill is to transmit local data (files or entire directories) to a remote IPFS cluster. While this is the intended functionality, the script allows for the upload of any accessible path, which could be leveraged to exfiltrate sensitive files if the agent is instructed to target them.

publish-artifact