verification-before-completion

Verification Before Completion is an essential engineering skill designed to eliminate false positives and speculative status reports in automated software development. It functions as a mandatory quality gate that prevents agents from asserting that tasks are finished until empirical evidence is produced. By mandating that agents execute specific verification commands and parse their outputs—checking for exit codes, test success counts, or clean linter results—this skill ensures that assertions are tethered directly to the current repository state rather than agent confidence or assumptions. It is intended for autonomous coding agents operating in environments where rigorous testing, regression verification, and reliable delivery are critical. Typical use cases include verifying Red-Green-Refactor cycles in TDD, confirming build success via exit codes, validating specific bug fixes against original symptoms, and ensuring all requirements are met through checklist verification before creating pull requests or committing code. This skill prevents common failures such as relying on partial checks, trusting outdated success reports, or making premature satisfaction claims like 'Done' or 'Perfect' without substantiating proof. It serves as a behavioral safeguard to maintain developer trust and system stability.

• Enforces the 'Iron Law': No completion claims without fresh verification evidence.

• Requires systematic execution of validation commands (tests, builds, linters) before any status reporting.

• Implements a mandatory identification-execution-verification-reporting workflow for every development task.

• Identifies red flags such as speculative language ('should', 'probably') and premature satisfaction expressions.

• Prevents reliance on agent introspection or external summaries instead of actual CLI tool outputs.

• Standardizes evidence-based reporting to ensure that PRs, commits, and task transitions are technically sound.

• The agent must always identify the specific command that proves the claim before acting.

• Requires the agent to read full tool output rather than relying on truncated or inferred logs.

• Treats the absence of an explicit verification command as a failure to meet the skill protocol.

• Applicable during every phase of software development: implementation, refactoring, debugging, and pre-deployment.

• Explicitly forbids moving to a new task, committing, or pushing code without first passing the verification gate.