The Agent Skills Directory

[SAFE]: The skill provides a structured and well-documented interface to the sc CLI for workflow orchestration. It incorporates safety best practices, such as requiring human-in-the-loop for destructive operations and independent verification of agent-generated content.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests untrusted data from external agent outputs (sc agent read), team reports (sc team status), and review threads (sc worktree review-get).
Ingestion points: Commands like sc agent read and sc worktree review-get bring external data into the agent's context.
Boundary markers: The SKILL.md file defines explicit boundaries, requiring human intent for significant actions and cautioning against inferring success without verification.
Capability inventory: The skill uses the sc CLI for orchestration, which includes capabilities for file system interaction, agent control, and process management.
Sanitization: The instructions emphasize manual verification and the use of stable identifiers to prevent confusion or unauthorized actions.
[COMMAND_EXECUTION]: The skill utilizes the sc CLI and standard utilities like jq and mktemp to manage session state and configuration. These operations are restricted to the intended purpose of the skill and do not involve arbitrary shell injection or unauthorized command execution.

superconductor