> For the complete documentation index, see [llms.txt](https://novusedge0.gitbook.io/docket-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://novusedge0.gitbook.io/docket-docs/technical-reference/integrations.md). # Manual agent setup Use this page to configure an agent manually or inspect the files that connect it to Docket. For guided setup, start with [Installation](/docket-docs/getting-started/installation.md). For daily use, read [Working with your agent](/docket-docs/everyday-use/agents.md). For an agent carrying out setup on the user's behalf, follow [Setup instructions for agents](/docket-docs/technical-reference/agent-setup.md). ## Configure an agent harness Each section below gives the hook and instruction files for one harness. Replace example paths with the absolute paths on your machine. Check that the selected Python interpreter is version 3.11 or later. Quote paths that contain spaces, and merge hook entries into existing configuration instead of replacing the file. `docket context --for gemini|copilot|cursor` wraps the ledger in the selected harness's hook envelope. The examples use this form, so a hook needs no shell pipe. ### Claude Code Add the marketplace and install Docket: ``` /plugin marketplace add NovusEdge/docket /plugin install docket@NovusEdge ``` Start a new session after installation. The plugin loads `hooks/hooks.json` and the Docket skill. ### OpenAI Codex CLI The repository contains `.codex-plugin/plugin.json`. Add the marketplace and install Docket: ```sh codex plugin marketplace add NovusEdge/docket codex plugin add docket@NovusEdge ``` Start a new Codex task after installation. Codex loads the bundled skill and session hook. Set `DOCKET_AUTHOR=codex` in the hook environment when you require this author name. Docket uses automatic detection when the variable is absent. ### OpenCode Create `~/.config/opencode/plugins/docket.ts`, or `.opencode/plugins/docket.ts` for one project. These locations follow the [OpenCode plugin guide](https://opencode.ai/docs/plugins/). The installer writes `plugins/docket.ts` for you. Its [plugin loader](https://github.com/anomalyco/opencode/blob/dev/packages/opencode/src/config/plugin.ts) scans files directly inside `plugins/`, so a file one level down is never discovered. Installers before 0.11.0 wrote `plugins/docket/index.ts`, which OpenCode never loaded. A new install or an update removes that file. Keep one active copy to avoid duplicate context. ```js import { execFileSync } from "node:child_process" const PYTHON = "/usr/bin/python3" const DOCKET = "/path/to/docket/bin/docket" export const Docket = async () => { return { "experimental.chat.system.transform": async (input, output) => { try { const ledger = execFileSync(PYTHON, [DOCKET, "context"], { encoding: "utf8", env: { ...process.env, DOCKET_AUTHOR: "opencode" }, }) if (ledger.trim()) output.system.push(ledger) } catch { return } }, } } ``` A plugin is a named export. It is an async function that returns the hooks. The hook adds the ledger to the system prompt before each model call. This example uses the named-export plugin interface and the experimental `chat.system.transform` hook. Check the interface when upgrading OpenCode. Copy `skills/docket/SKILL.md` to `.opencode/skills/docket/SKILL.md`. OpenCode also finds skills in `.agents/skills/`. ### Gemini CLI Put this in `~/.gemini/settings.json`, or in `.gemini/settings.json` for one project: ```json { "hooks": { "SessionStart": [ { "hooks": [ { "type": "command", "name": "docket", "command": "/path/to/docket/bin/docket context --for gemini", "timeout": 5000 } ] } ] } } ``` The entry has no `matcher`, so it applies to each session-start event. Gemini's [hook reference](https://geminicli.com/docs/hooks/reference/) uses exact strings for lifecycle matchers and regular expressions for tool matchers. The timeout is in milliseconds. Copy `skills/docket/SKILL.md` to `.gemini/skills/docket/SKILL.md`. ### GitHub Copilot CLI Put this in `.github/hooks/sessionStart.json` for a repository, or `~/.copilot/hooks/sessionStart.json` for yourself: ```json { "version": 1, "hooks": { "sessionStart": [ { "type": "command", "bash": "DOCKET_AUTHOR=copilot /path/to/docket/bin/docket context --for copilot", "powershell": "$env:DOCKET_AUTHOR=\"copilot\"; & \"C:\\Path\\To\\python.exe\" \"C:\\path\\to\\docket\\bin\\docket\" context --for copilot", "timeoutSec": 5 } ] } } ``` Copilot parses the hook output as JSON. It adds `additionalContext` to the model context. Copilot runs the `bash` field on Linux and macOS. It runs the `powershell` field on Windows. PowerShell requires the call operator `&` before a quoted command path. For instructions, Copilot reads `.github/copilot-instructions.md`. Copy the body of `skills/docket/SKILL.md` into it. ### Cursor Use two files to install Docket for Cursor. Hook, in `.cursor/hooks.json` for the project or `~/.cursor/hooks.json` for yourself: ```json { "version": 1, "hooks": { "sessionStart": [ { "command": "/path/to/docket/bin/docket context --for cursor" } ] } } ``` Cursor calls its output field `additional_context`. Instructions go in `.cursor/rules/docket.mdc`. The file requires this frontmatter to make the rule apply automatically: ``` --- alwaysApply: true --- ``` ### A harness with no hooks Without a session-start hook, the harness cannot run a command for each session. 1. Write a current ledger snapshot to the instruction file: ```sh docket context > DOCKET_CONTEXT.md ``` 2. Configure the harness to load `DOCKET_CONTEXT.md`. 3. Regenerate the file after each decision. ## Instruction files Use the instruction or skill location described for each tool above. Those locations configure Docket; they are not a complete list of every instruction format an agent tool supports. ## Integration references * [OpenCode plugins](https://opencode.ai/docs/plugins/) * [OpenCode agent skills](https://opencode.ai/docs/skills) * [Gemini CLI hooks](https://geminicli.com/docs/hooks/reference/) * [GitHub Copilot CLI hooks](https://docs.github.com/en/copilot/reference/hooks-reference) * [Cursor hooks](https://cursor.com/docs/hooks) * [OpenAI Codex plugins](https://learn.chatgpt.com/docs/plugins) ## Verification scope The Gemini, Copilot, and Cursor hook formats were checked against the linked references on 13 September 2026. The Codex plugin commands were checked against the installed CLI help. The OpenCode example follows its documented plugin location and named-export interface. These checks do not establish that a live session in every tool loads the briefing. After manual setup, start a new session and check that the agent can read your project's Docket context. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://novusedge0.gitbook.io/docket-docs/technical-reference/integrations.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.