Mechanism: Best-of-N — Run Multiple Approaches, Pick the Winner
Best-of-N is running N independent agent attempts on the same goal — separate branches, chats, or models — then choosing…
Mechanism: Branch-per-Experiment — Isolate Each Vibe
Branch-per-experiment means each agent attempt, slice, or hypothesis gets its own git branch — so you can compare, abandon, or…
Mechanism: Playground — Experiment Without Breaking Production
A playground is an isolated environment — branch, directory, sandbox repo, or local-only config — where agent experiments cannot damage…
Mechanism: Diff-as-Context — Steer Attention to What Changed
Diff-as-context is using the actual change set — `git diff`, PR patch, or rejected hunks — as the primary steering…
Mechanism: Session Handoff — Summary for the Next Chat
A session handoff is a short structured summary — goal, what changed, verify state, open questions — passed to the…
Mechanism: Rules & Skills as Memory — Persistent Conventions the Agent Re-Reads
Rules and skills as memory means storing stable project conventions in files the agent loads every session — so you…
Mechanism: Anchor Files — Always Include the Project’s Compass
Anchor files are the small set of documents or source files you almost always attach — README, types, architecture note,…
Mechanism: Progressive Disclosure — Add Context Only When Stuck
Progressive disclosure is starting an agent session with minimal context — goal, anchors, failing output — and adding files or…
Mechanism: Escalation — When to Switch Model, Mode, or Human
Escalation is a planned step-up when the current approach stalls — switch model, mode, human reviewer, narrower slice, or different…
Mechanism: Interface-First — Types and API Before Implementation
Interface-first means defining the contract — types, function signatures, route schemas, or module API — before the agent implements internals,…