The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local shell commands and Node.js execution to discover, list, and manage workflow recipes stored in ~/.claude/workflows and project-local directories.
[COMMAND_EXECUTION]: Several recipes (e.g., consensus-gate, feature-plan-consensus, spec-implement-dual-review-verify) orchestrate secondary AI models by driving the agent-tmux CLI tool. Input variables like cli and slug are validated with regular expressions to prevent command injection.
[COMMAND_EXECUTION]: The workflow-manifest recipe is designed to perform remote inventory scans across a fleet of machines using SSH commands (ssh -o BatchMode=yes ...). This functionality is used to snapshot and synchronize workflow assets across user-specified infrastructure.
[SAFE]: The installation script (scripts/install.sh) performs local file management and directory creation to deploy bundled workflow assets; it does not perform any remote network downloads from the internet.
[SAFE]: The workflows implement defensive prompting techniques, such as a 'Correctness Doctrine' that mandates verifying all claims against real source code and logs, and 'Adversarial Review' where secondary models are explicitly used to find weaknesses or hallucinations in proposals.

using-workflows