AI Protocol Kit is a collection of practical, field-tested behavioral contracts for AI-assisted work. These are not magic prompts, clever hacks, or system messages designed to unlock hidden model behavior. They are working contracts — explicit instructions you give to an AI before it starts work, so it knows how to approach a task, what to check, what not to assume, when to stop, and what kind of output is acceptable. The kit covers the full range of practical AI workflows: rough ideas, repository reviews, README authorship, GitHub publication, public pages, bug findings, structured outputs, and complex sessions where the AI needs stronger behavioral rails before it acts.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/xxyoudeadpunkxx/ai-protocol-kit/llms.txt
Use this file to discover all available pages before exploring further.
Protocols vs. Prompts
A prompt says what you want. A protocol says how the AI should work to get there. When you paste a prompt, you are telling the AI the destination. When you load a protocol, you are giving it a working contract: a defined operating mode, a set of gates it must pass before proceeding, explicit rules about what it cannot assume, and a stop condition it must respect if something is missing. Protocols separate shaping from drafting, evidence from inference, and observed state from claimed state. They are not decoration — they are the behavioral structure that replaces improvisation.Pick by Failure Mode, Not by Name
The most important principle in the kit is the selection method: do not pick a protocol by name first — pick it by the failure mode you want to prevent. Every protocol in this kit exists because a specific, recurring failure showed up in real AI-assisted work. Some tasks collapse too quickly into the obvious answer before constraints are surfaced. Some AI sessions invent repository state instead of inspecting it. Some outputs are written for the AI’s idea of a reader, not an actual one. Some bug reports mix evidence with inference until neither can be trusted. Find the failure you are trying to prevent, then find the protocol that addresses it. The six failure lanes below are the starting points.The Six Failure Lanes
The 17 protocols in this kit are organized into six lanes. Each lane describes a class of work where AI-assisted sessions reliably break down without explicit behavioral constraints. Shape the Work — The task is ambiguous, premature, overloaded, or underspecified. AI collapses fuzzy input into a confident answer before the real problem is even visible. The protocols in this lane force shaping before synthesis: they extract constraints, identify real gates, and prevent the AI from switching into execution mode before the structure is stable. Brief the Artifact — AI needs a clear brief before writing, building, or structuring output. Without one, it defaults to a generic format that fits no particular reader, platform, or purpose. The protocols here close the brief before anything is drafted — reader, purpose, tone, constraints, and language are all locked first. Review Findings and Evidence — Inspection work, bug reports, and field observations collapse when evidence and inference get mixed together. The protocols in this lane keep findings structured as artifacts, separate observable facts from hypotheses, and preserve uncertainty instead of resolving it through confident-sounding prose. Reasoning, Constraints & System Reading — Some tasks are not just unclear — they are structurally easy to misread, over-simplify, or resolve with a fake compromise. The protocols here force expansion before answer, read the gap between declared intent and observable behavior, and handle constraint interactions without letting one constraint quietly dominate the others. GitHub / Public Repo & Page Publishing — GitHub work has its own failure modes: publishing before review, incorrect discovery files, wrong README framing, badges that track nothing real, and Pages paths that break silently. The six protocols in this lane cover the full publication surface from repository preparation to public page metadata. Govern Agentic Sessions — When AI operates across files, tools, repo state, and multiple passes, it needs stronger behavioral rails than a single prompt can provide. The protocols in this lane govern the session itself: execution ledgers, review loops, executor/reviewer separation, and explicit stop conditions that prevent the AI from approving its own work.Protocols do not guarantee correctness. They reduce specific, known failure modes — rushing to the obvious answer, writing for the wrong reader, inventing unsupported claims, treating chat summaries as repository truth, publishing before review, or letting AI tools approve their own work. They are not a replacement for judgment.Higher-priority instructions always override any protocol in this kit. Safety, legality, privacy, data integrity, and tool limits are above every behavioral contract here.
Where to Go Next
How to Use
Four steps for selecting, loading, and running a protocol in any AI session.
Protocol Map
All 17 protocols organized by lane and failure mode.
Minimal Stacks
Pre-built combinations for common workflows like GitHub publication and agentic sessions.
Browse Protocols
Start with Idea Shaping or jump directly to the protocol you need.