The Agent Skills Directory

[COMMAND_EXECUTION]: The skill operates by executing the sc CLI tool for orchestration, layout management, and worktree operations. It utilizes jq to construct temporary JSONL files for command input, which is a safe practice that prevents shell injection issues when handling complex prompts.
[EXTERNAL_DOWNLOADS]: The skill documentation references a dependency on the Superconductor application and its sc CLI tool. It does not perform automated installation or downloading of these external components, relying on the host environment.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for indirect prompt injection as it ingests and processes data from external sources, including other agent outputs and in-app review threads.
Ingestion points: Commands such as sc agent read, sc team status, and sc worktree review-get in SKILL.md and reference files read data into the agent's context.
Boundary markers: Explicit "ignore instructions" delimiters are not mandated, but the skill instructs the agent to verify all results independently using Git and tests, and requires human intent for destructive mutations.
Capability inventory: The skill uses subprocess calls for sc commands, including task dispatch, worktree management, and agent interaction.
Sanitization: The skill uses jq to escape content when creating launch input files, providing a layer of sanitization for prompt data.

superconductor