The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/planctl utility executes shell commands defined in the project's plan/manifest.yaml file. These commands are used as 'quality gates' to verify the success of a phase (e.g., running build scripts or test suites) before advancing the workflow. This functionality is the primary intended use of the skill.\n- [DATA_EXFILTRATION]: The complete command in the scripts/planctl script performs an automated milestone process consisting of git add -A, git commit, and git push. This provides repository-backed persistence for the project's state. While this involves sending project data to a remote server, it is a core feature for collaborative and long-running workflows. The skill includes robust instructions for the agent to maintain the project's .gitignore file to mitigate the risk of accidentally staging sensitive information.\n- [PROMPT_INJECTION]: The skill includes extensive agent instructions (e.g., in references/agent-instructions-template.md) that are synchronized across multiple files (CLAUDE.md, AGENTS.md, .github/copilot-instructions.md). These instructions enforce a 'Three-File Context Law' and other operational constraints designed to prevent state drift and context contamination. This is a legitimate application of instructional steering for workflow stability and does not involve bypassing safety guidelines.

phase-contract-workflow