The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The SKILL.md file contains instructions for a remote installation command (curl -fsSL https://taskmarket.dev/install-skill.sh | sh) that pipes a script from an external URL directly into the shell. This allows for arbitrary code execution on the user's system and is a highly unsafe pattern that bypasses security verification.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from the Taskmarket platform, creating a potential vulnerability to indirect prompt injection.
Ingestion points: Files including SKILL.md and reference/raw-api.md describe the ingestion of untrusted inputs such as task descriptions, requester messages, pitches, proofs, artifacts, and API responses.
Boundary markers: The SKILL.md file defines a clear "Trust Boundary" section that explicitly instructs the agent to treat these external inputs as untrusted and prevents them from overriding system instructions.
Capability inventory: The skill utilizes the taskmarket CLI for operations and executes various shell commands (e.g., curl, jq) based on task data, as outlined in reference/cli.md and reference/raw-api.md.
Sanitization: SKILL.md provides instructions to the agent to inspect code before execution and explicitly warns against piping untrusted content into shell interpreters.
[EXTERNAL_DOWNLOADS]: The skill references external resources for installation and configuration, including the @lucid-agents/taskmarket package via npm and the retrieval of openapi.json and network.info from https://api.taskmarket.dev.
[COMMAND_EXECUTION]: The skill documentation makes extensive use of command-line tools, including the first-party taskmarket CLI and system utilities such as curl, jq, and printf, to manage blockchain wallets and interact with the platform's API.

taskmarket