The original title is "Session Discipline Kit: enforcing verification in AI coding agents"
Original: I built a kit to stop AI agents from lying about reading your code — then had to rebuild it for a tool that didn't exist yet
Short summary
The Session Discipline Kit is a set of hooks and skills that force AI coding agents to verify before they edit. A core hook blocks any file edit the agent hasn't actually opened and read in the current session, catching a common failure mode where agents confidently make wrong assumptions about code. Nine additional skills map onto real change stages — scoping, consumer enumeration, snapshot testing, diff review, and verified handoff generation — built from six generalized failure patterns observed in practice.
- •A hook blocks edits to files the agent hasn't actually read in the current session
- •Nine skills enforce verification at each stage of a real code change
- •Handoff documents are regenerated from git state, not conversation memory, to prevent unverified assumptions across sessions
Generated with AI, which can make mistakes.
Is this a good recommendation for you?



