Skip to main content

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.

This stack governs complex AI sessions where the AI must operate across files, repositories, external tools, or multiple passes — and where losing track of state, overwriting real targets, or conflating proposed changes with applied ones are real risks. Without explicit behavioral contracts, AI in agentic or multi-step sessions tends to widen scope silently, treat its own outputs as ground truth, skip verification steps, and produce confident summaries that do not reflect what actually happened. This stack prevents those failure modes by establishing posture, execution tracking, and — when needed — actor separation and evidence capture.
1

GPT Agentic Posture Contract

Establishes output style, inference constraints, verification requirements, ledger-based execution tracking, and stop conditions for the AI session. This is the baseline contract: it controls how the AI behaves throughout the session — what it can assume, when it must verify, what counts as done, and when it must stop rather than proceed.View protocol →
2

Triad AI Orchestration Protocol v3 (conditional)

Add this step only if executor/reviewer separation is needed — that is, if the session involves a reviewing AI evaluating the work of an executing AI, and you need strict authority ordering between them. This protocol defines a three-actor model (OP + AI_EXEC + AI_REVIEW) where the reviewing actor has advisory-only status and cannot override the executing actor or approve its own work.View protocol →
3

Field Findings & Bugs Protocol v2 (conditional)

Add this step if findings must be captured as structured evidence during the session — for example, if the session involves a repository audit, code inspection, or any pass where observations need to be held separately from applied changes. Keeps evidence, inference, and uncertainty separated throughout the session.View protocol →
Canvas compatibility notice: The GPT Agentic Posture Contract depends on Canvas as a persistent execution ledger — a separate surface where the AI tracks what has been proposed, verified, and applied. Since OpenAI’s May 28, 2026 GPT-5.5 update, Canvas is no longer available in GPT-5.5 Instant or GPT-5.5 Thinking. Without Canvas, the contract remains useful as posture guidance — it still shapes inference constraints, verification behavior, and stop conditions — but it should not be treated as a full agentic execution protocol. A limited workaround may exist in some ChatGPT sessions by opening Canvas before invoking the contract, but this is fragile, UI-dependent, and not contractual.

​
When to use this stack

  • Multi-step file editing sessions where the AI must track which changes have been proposed versus applied
  • Repository analysis and patching sessions where scope creep and silent assumption are real risks
  • Any session where you need the AI to maintain a running ledger of its own execution state, not just produce output
  • Review loops involving a separate reviewing actor who must not be able to approve their own work or override the executing actor
  • Sessions that combine file operations, tool calls, and evidence capture across multiple passes

Build docs developers (and LLMs) love

Get started for freeTalk to us