Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/r-design-j/tri-party-framework/agents-mdgit clone --depth 1 https://github.com/r-design-j/tri-party-frameworkWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/r-design-j/tri-party-framework/agents-md)<a href="https://agentmods.dev/instructions/r-design-j/tri-party-framework/agents-md"><img src="https://agentmods.dev/badge/instructions/r-design-j/tri-party-framework/agents-md.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.02002 | $0.02002 |
| Opus 5 | $0.01001 | $0.01001 |
| Sonnet 5 | $0.00400 | $0.00400 |
| Haiku 4.5 | $0.00200 | $0.00200 |
Grade A, and why
tri-party-framework AGENTS.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 91 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Global Codex Working Agreements
Output Review Standard
- Before sending any user-visible Codex window output, final response, document content, Figma/design change summary, or generated artifact for review, perform at least 3 self-review passes.
- The 3 passes must check:
- Alignment with the user's latest request and wording.
- Completeness and correctness of the actual result, not just the intended result.
- Visible quality issues such as overlap, overflow, missing content, wrong labels, wrong hierarchy, or unclear placement when the work involves UI, Figma, documents, slides, or screenshots.
- Do not say the work is done until the review passes. If a check fails, fix it first and review again.
- When the user asks to "固化" a working agreement, prefer adding it here so future Codex sessions inherit it.
Daily Work Standard Extraction
- At the end of each working day, summarize that day's completed work items and decisions.
- From the daily work summary, extract reusable, abstract, and implementable standards rather than only recording events.
- Present the extracted standards as a clear list, with each item written so it can be applied to future tri-party framework optimization.
- Treat these standards as real operational data and source material for continuously improving the existing tri-party framework system.
- When summarizing, explicitly look for optimization opportunities, missing capabilities, reusable patterns, and feature ideas that could enrich the framework.
- The Codex, Claude, and Gemini model parties should cross-check each other's reasoning where relevant, identify reusable lessons, and use the results as input for ongoing self-iteration.
Tri-party Framework Definition
- "Tri-party framework" and "three-party protocol" mean collaboration among exactly these three model parties: Codex, Claude, and Gemini.
- Default capability roles:
- Codex: primary owner for real project code, repository edits, implementation, tests, and in-worktree execution.
- Claude: primary owner for complex reasoning, long-chain agent planning, architecture tradeoff analysis, and autonomous multi-step strategy.
- Gemini: primary owner for multimodal work involving PDFs, video, audio, images, Google Search/Maps/URL context, and Google ecosystem synthesis.
- Current model bindings live in
docs/framework/model-binding.yaml. - The executable protocol and source-check template live in
docs/framework/tri-party-protocol.md. - Prefer the executable scripts
scripts/triparty-preflight.shandscripts/triparty-review.shfor source checks and repeatable tri-party reviews before relying on manual CLI calls. - Codex sub-agents, local explorer agents, worker agents, or other Codex-internal delegations do not count as Claude or Gemini.
- Before claiming that a result is a tri-party conclusion, verify and state the source of each party's input:
- Codex: local Codex reasoning or execution result.
- Claude: direct Claude CLI/tool/API result, connector result, or user-provided Claude transcript.
- Gemini: direct Gemini CLI/tool/API result, connector result, or user-provided Gemini transcript.
- Source availability checks must include both exposed tools/connectors and local shell CLI availability, including
type -a claudeandtype -a geminiwhen shell access is available. - If Claude or Gemini is not actually available in the current environment, do not simulate or substitute them with Codex sub-agents. Label the result as "Codex-only provisional" or "Codex plus Codex sub-agents" instead.
- Any summary that mentions Codex, Claude, and Gemini must include a source-status line showing whether each party was directly called, supplied by the user, or unavailable.
- When one or more parties are unavailable, proceed only with clearly marked partial analysis and list the missing party inputs needed for a true tri-party review.
- If a previous response incorrectly described Codex sub-agent output as tri-party output, correct the record immediately and add the failure mode to the daily standard extraction candidates.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago First seen · 91 lines · 2,002 tokens per session scan A 63e9a113fd63
tri-party-framework AGENTS.md is an instructions file published in the GitHub repository r-design-j/tri-party-framework (4 stars, last pushed 2mo ago), licensed MIT. It adds 2,002 tokens to every session, about $0.0100 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
ai-assistant-instructions AGENTS.md
AGENTS.md instructions for dryvist/ai-assistant-instructions, covering ai agents configuration, coding behavior, tool choice, starting any change and git workflow.
ai-assistant-instructions CLAUDE.md
Claude Code instructions for dryvist/ai-assistant-instructions: Canonical agent instructions live in AGENTS.md and agentsmd/rules/. Nix-managed machines auto-load both globally via nix-ai, so this file deliberately does not re-import them — doing so would load every byte twice in each session. Non-Nix contributors…
ai-assistant-instructions copilot-instructions.md
Copilot instructions for dryvist/ai-assistant-instructions: ../agentsmd/docs/codereviewinstructions.md.
ai-assistant-instructions GEMINI.md
Gemini CLI instructions for dryvist/ai-assistant-instructions, a project described as: A comprehensive, vendor-agnostic framework for consistent AI-assisted development workflows - standardized instructions and commands that work seamlessly across Claude, Gemini, Copilot, and local AI models. See also…
ClawRouter CLAUDE.md
Claude Code instructions for BlockRunAI/ClawRouter, covering clawrouter, commands, project structure, key dependencies and conventions.
kungfu copilot-instructions.md
Copilot instructions for kungfu-systems/kungfu: Read and follow AGENTS.md before proposing or running changes. It is the shared agent router for this repository, including the canonical Shifu development and build entrypoint.