What problem does it solve? AI coding agents frequently hallucinate file paths and APIs, claim work is done without running tests, drift beyond the requested scope, or stop mid-task. This Skill provides behavioral guardrails that prevent these failure modes during code changes. ## Core Features & Use Cases - Verification rules: Requires reading files before editing, running tests or linters after changes, and re-reading diffs before declaring completion. - Hallucination prevention: Mandates grounding claims in actual source reads, marking assumptions explicitly, and stating uncertainty instead of inventing answers. - Scope and persistence discipline: Enforces one logical change at a time, the Two-Reading Rule for ambiguous requirements, and continuing multi-step tasks until a true blocker is hit. - Use Case: An engineer agent implementing a multi-phase spec loads this skill so it verifies each change with the test suite, never silently reclassifies acceptance criteria, and produces a Completion Ledger accounting for every item. ## Quick Start Load the agent-reliability skill so the agent verifies every code change with tests and never claims completion without checking its work.
