
digitalnomadjoe/messages
Developer toolsこのページは AI により自動翻訳されています。技術的な詳細は原文をご確認ください。 原文を見る
messages b/w claude and gpt
概要
New here — GPT, Codex, Claude, or any other coding agent? Start with , then . The standard ChatGPT GitHub connector returns (tested twice). It can inspect Telephone but cannot operate it. Browser operation requires the private — actions/TELEPHONE_GPT_SETUP.md has the token scope, OpenAPI schema and exact UI steps. Paste this to bootstrap a session (once the Action is set up): Use browser-mode Telephone from https://github.com/digitalnomadjoe/messages. Read AGENTS.md and TELEPHONE.md, then inspect projects/brittle/state/browser_status.json. If browser_telephone_ready is false, give me the exact commands from required_local_actions to run locally. Do not request workstation access. After status becomes ready, operate Telephone through browser requests only. If browser_telephone_ready is false the agent will hand you the exact local commands to run — it will never ask for SSH, a password, a key or terminal access, and it cannot run them itself.
README
以下は原文 README の抜粋です。原文を見る。
BRITTLE messages
Using Telephone from a new AI session
New here — GPT, Codex, Claude, or any other coding agent? Start with
AGENTS.md, then TELEPHONE.md.
| file | read it before |
|---|---|
AGENTS.md |
doing anything at all in this repository |
TELEPHONE.md |
operating Telephone (bounded autonomous runs) |
skills/telephone/SKILL.md |
starting, inspecting or stopping a run |
skills/brittle-messages/SKILL.md |
acting as a lane agent on one ticket |
⚠ The GitHub connector is read-only
The standard ChatGPT GitHub connector returns HTTP 403 on writes (tested
twice). It can inspect Telephone but cannot operate it. Browser operation
requires the private Telephone custom GPT Action —
actions/TELEPHONE_GPT_SETUP.md has the
token scope, OpenAPI schema and exact UI steps.
Paste this to bootstrap a browser-mode session (once the Action is set up):
Use browser-mode Telephone from
https://github.com/digitalnomadjoe/messages. ReadAGENTS.mdandTELEPHONE.md, then inspectprojects/brittle/state/browser_status.json. Ifbrowser_telephone_readyis false, give me the exact commands fromrequired_local_actionsto run locally. Do not request workstation access. After status becomes ready, operate Telephone through browser requests only.
If browser_telephone_ready is false the agent will hand you the exact local
commands to run — it will never ask for SSH, a password, a key or terminal
access, and it cannot run them itself.
Paste this to bootstrap a workstation session:
Use the Telephone skill from this repository:
https://github.com/digitalnomadjoe/messagesReadTELEPHONE.mdfirst, inspect Telephone status, and do not begin or resume a run until you have verified the current lane, run state, cycle limit, stopping criterion, spending guard, and open tickets or escalations.
Then instruct it in plain language:
Run Telephone on locomotion for 10 loops.
Run Telephone on locomotion until both touchdown speeds are below -100, maximum 12 loops.
Show Telephone status without changing anything.
Stop the active Telephone run after the current claimed ticket finishes.
Repository read access is enough to understand and review Telephone.
Operating it additionally needs repository write access plus access to the
configured workstation running messagesctl. A session that cannot execute
those commands must say so and hand back the exact command — never claim a run
was started.
An autonomous, durable message bus between the BRITTLE lane agents (Claude) and an OpenAI reviewer, backed by this GitHub repository.
Before: Joe moved reports and tickets between ChatGPT and Claude by hand — downloading, copying, pasting, forwarding. After: an agent publishes a report,
推奨ツール
別のキーワードを試すか、フィルタを外してください。
インストール
npx skillfish add digitalnomadjoe/messages