agent-coding-contract

Load this skill at the start of any session where a coding agent is assisting with code changes. Defines the rules governing how the agent operates: two execution modes (Interactive default vs Autonomous opt-in), what requires explicit user authorization before acting, how to handle ambiguity without guessing, how to log decisions made without user input, when to stop vs proceed, and the five-step task workflow every task must follow. Use this skill whenever you need to establish or reinforce the behavioral contract — especially before autonomous tasks, refactors, multi-file changes, or any work where scope boundaries matter. Also load this skill when the user says "proceed autonomously", "don't stop to ask", or "interactive mode".

Agent Coding Contract

Behavioral contract for coding agents assisting with code changes. Covers two execution modes (Interactive default vs Autonomous opt-in), explicit authorization rules, ambiguity resolution, the five-step task workflow, stop conditions, partial failure handling, and the decision log format for autonomous decisions.

References

Load the relevant file for the topic at hand.

File	Topic
references/agent-role.md	What the agent may and must not do
references/execution-modes.md	Interactive vs Autonomous modes, authority hierarchy when context files conflict
references/task-workflow.md	Five-step workflow, task decomposition, task completion output
references/core-rules.md	Core rules and behavioral summary table
references/stop-conditions.md	When to stop, partial failure handling, multi-agent scenarios
references/decision-log.md	Decision log entry format for Autonomous Mode
references/non-goals.md	Universal non-goals — what never to do unless explicitly asked