{"id":50065245,"url":"https://github.com/ray-x/copilot-agent.nvim","last_synced_at":"2026-07-07T12:01:54.770Z","repository":{"id":353897936,"uuid":"1221339736","full_name":"ray-x/copilot-agent.nvim","owner":"ray-x","description":"An autonomous orchestrator for the Copilot Go SDK in Neovim, delivering a native, agentic coding experience.","archived":false,"fork":false,"pushed_at":"2026-06-03T05:58:16.000Z","size":3375,"stargazers_count":14,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"master","last_synced_at":"2026-06-27T16:41:41.371Z","etag":null,"topics":["agent","agentic-coding","agentic-workflow","ai-agent","ai-coding","copilot","copilot-cli","copilot-sdk","github-copilot","neovim","nvim"],"latest_commit_sha":null,"homepage":"","language":"Lua","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"other","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/ray-x.png","metadata":{"files":{"readme":"README.md","changelog":"CHANGELOG.md","contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":"NOTICE","maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-04-26T04:17:10.000Z","updated_at":"2026-06-03T05:57:38.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/ray-x/copilot-agent.nvim","commit_stats":null,"previous_names":["ray-x/copilot-agent.nvim"],"tags_count":2,"template":false,"template_full_name":null,"purl":"pkg:github/ray-x/copilot-agent.nvim","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ray-x%2Fcopilot-agent.nvim","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ray-x%2Fcopilot-agent.nvim/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ray-x%2Fcopilot-agent.nvim/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ray-x%2Fcopilot-agent.nvim/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/ray-x","download_url":"https://codeload.github.com/ray-x/copilot-agent.nvim/tar.gz/refs/heads/master","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ray-x%2Fcopilot-agent.nvim/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":35226918,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-07-07T02:00:07.222Z","response_time":90,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["agent","agentic-coding","agentic-workflow","ai-agent","ai-coding","copilot","copilot-cli","copilot-sdk","github-copilot","neovim","nvim"],"created_at":"2026-05-21T23:00:21.360Z","updated_at":"2026-07-07T12:01:54.752Z","avatar_url":"https://github.com/ray-x.png","language":"Lua","funding_links":[],"categories":["AI"],"sub_categories":["Diagnostics"],"readme":"# copilot-agent.nvim\n\nGitHub Copilot's agentic runtime, natively in Neovim. A lightweight Go bridge to the [official SDK](https://github.com/github/copilot-sdk) with native tool execution, four chat modes, session-aware permissions, persistent sessions, repository-local agent/skill discovery, and LSP code actions.\n\n## Demo video\n\nOne minute video to create a terminal version of flappy bird:\n\u003cvideo src=\"https://gist.github.com/user-attachments/assets/5091ddcb-84f8-4c1a-b8a7-ea4bef6393f4\" controls width=\"70%\"\u003e\u003c/video\u003e\n\n## Key Capabilities\n\n| Feature | Technical Highlight |\n| :---------------------------- | :-------------------------------------------------------------------------- |\n| **Autonomous SDK Loop** | Native file I/O, terminal execution, and web search via official Go SDK. |\n| **State Rewind** | Undo/Restore codebase and session to any previous conversation point. |\n| **Session Continuity** | Resume sessions across Neovim, VS Code, and CLI with shared metadata. |\n| **LSP-Native** | AI-driven 'Fix', 'Test', and 'Explain' via standard `code_action` triggers. |\n| **Autopilot Modes** | Four levels of oversight, from per-step approval to full autonomy. |\n| **Project Intelligence** | Automatic discovery of `.github/agents` and `.github/skills` configs. |\n| **Integrate to nvim plugins** | review changes in Diffview; commit with fugitive and more |\n| **Live Observability** | Async statusline and streaming UI for sub-agent tracking and token usage. |\n\n---\n\n### Agent Tool Loop\n\nThe agentic loop is what makes this plugin different from simple chat wrappers. The assistant doesn't just answer; it acts: reading files, fetching web pages, running commands, writing code, and iterating until the task is done.\n\n## Architecture\n\n```mermaid\nflowchart TD\n subgraph Neovim[\"Neovim (Lua plugin)\"]\n UI[\"Chat buffer / input window\"]\n LSPClient[\"vim.lsp (code actions)\"]\n Cmds[\"CopilotAgent* commands\"]\n end\n\n subgraph GoService[\"Go HTTP service :8088\"]\n HTTP[\"HTTP router\\n/sessions /models /healthz\"]\n SSE[\"SSE fan-out\\nsession.event host.*\"]\n Sessions[\"Session manager\\npermissions · tools · models\"]\n end\n\n subgraph LSPProcess[\"LSP-only process\"]\n LSPServer[\"LSP server (stdio)\\ntextDocument/codeAction\\nworkspace/executeCommand\"]\n end\n\n subgraph SDK[\"GitHub Copilot SDK\"]\n CopilotSDK[\"copilot-sdk/go\\nCopilot CLI / API\"]\n Tools[\"Built-in tools\\nread_file · write_file · terminal\\nweb_search · ask_user · …\"]\n end\n\n UI --\u003e|\"curl POST /sessions/{id}/messages\"| HTTP\n UI --\u003e|\"curl -N SSE stream\"| SSE\n Cmds --\u003e|\"curl POST/DELETE\"| HTTP\n LSPClient \u003c--\u003e|\"JSON-RPC stdio\"| LSPServer\n LSPServer --\u003e|\"HTTP to existing service\"| HTTP\n HTTP --\u003e Sessions\n SSE --\u003e Sessions\n Sessions \u003c--\u003e|\"copilot.Session\"| CopilotSDK\n CopilotSDK --\u003e Tools\n```\n\nThe HTTP bridge and the LSP helper are separate processes. The HTTP service owns session state; `vim.lsp.start()` launches an LSP-only helper that talks to the already-running HTTP service. The Lua plugin communicates via `curl` shell-outs for all HTTP and SSE traffic.\n\n---\n\n## Comparison with Alternatives\n\n### vs CopilotChat.nvim\n\n[CopilotChat.nvim](https://github.com/CopilotC-Nvim/CopilotChat.nvim) calls the Copilot (or other) LLM REST APIs directly from Lua. It supports multiple providers but has no agent runtime of its own — tool execution and the agentic loop are implemented in Lua above the client.\n\n\u003cdetails\u003e\n\u003csummary\u003e\u003cstrong\u003e📂 Feature-by-feature comparison with CopilotChat.nvim\u003c/strong\u003e\u003c/summary\u003e\n\n| Feature | **copilot-agent.nvim** | CopilotChat.nvim |\n| ------------------------- | --------------------------------------------------------------------- | ------------------------- |\n| Backend | Official Copilot SDK (Go) | Direct LLM REST API (Lua) |\n| Agent / tool-use mode | ✅ full agentic (file edits, terminal, web search, …) | ❌ chat only |\n| Chat modes | ask · plan · **agent** · autopilot | ask only |\n| Permission management | ✅ interactive / approve-reads / approve-all / autopilot / reject-all | ❌ |\n| Config discovery | ✅ SDK-native (`.github/copilot-instructions.md`, etc.) | ❌ manual |\n| Custom agents / skills | ✅ SDK-native, same as VS Code | ❌ |\n| File \u0026 folder attachments | ✅ (buffer, selection, file, folder, image, clipboard paste) | ✅ (buffer context❓ ) |\n| Session persistence | ✅ per working directory | ❌ |\n| Model switching (live) | ✅ mid-session switching | ❓ |\n| LSP code actions | ✅ (explain / fix / add tests / add docs) | ❌ |\n| MCP support | ✅ | ❓ |\n| Multi-provider | ❌ (Copilot only, or Bring your own key) | ✅ (provider_resolver) |\n| Dependencies | codepilot-cli + go server | Pure Lua |\n\nThe model picker shows each model's billing multiplier when the SDK provides one.\n\n\u003c/details\u003e\n\n**When to choose CopilotChat.nvim**: zero-binary Lua setup, just want Copilot chat with buffer context, happy with a Lua-managed tool loop.\n\n**When to choose copilot-agent.nvim**: you want the Copilot SDK owning the agent loop with native tools, permission control, and session persistence.\n\n---\n\n### vs ACP plugins (codecompanion.nvim, avante.nvim in ACP mode)\n\n[**ACP (Agent Client Protocol)**](https://agentclientprotocol.com) lets a Neovim plugin act as a client to any external CLI agent. The plugin sends prompts and streams back results; the CLI agent owns the tool execution and agentic loop. Both codecompanion.nvim and avante.nvim support ACP.\n\n`copilot-agent.nvim` is narrower in scope but deeper in Copilot integration: the Go service embeds the Copilot SDK directly, so it gets SDK-native features (config discovery, custom agents, skill directories, sub-agent streaming) that no ACP bridge can expose.\n\n\u003cdetails\u003e\n\u003csummary\u003e\u003cstrong\u003e📂 Feature-by-feature comparison with ACP-based plugins\u003c/strong\u003e\u003c/summary\u003e\n\n| Feature | **copilot-agent.nvim** | codecompanion.nvim | avante.nvim |\n| ----------------------------- | ------------------------------------------------ | ----------------------------------------- | -------------------------------------- |\n| Agent backend | Copilot SDK (Go, embedded) | ACP CLI agents or direct LLM adapters | ACP CLI agents or direct LLM adapters |\n| ACP support | ❌ (no plan) | ✅ (Claude Code, Codex, Copilot CLI, …) | ✅ (Zen Mode) |\n| Multi-provider / BYO API key | ❌ (Copilot only) | ✅ (Anthropic, OpenAI, Gemini, Ollama, …) | ✅ (Claude, OpenAI, Gemini, Ollama, …) |\n| Tool-call execution | SDK built-ins (file, terminal, web, ask_user …) | Lua tools + ACP agent tools | Rust tools + ACP agent tools |\n| Sub-agent / streaming events | ✅ SDK-native | ❌ | ❌ |\n| MCP, agents \u0026 skill discovery | ✅ | ❌ | ❌ |\n| Permission management | ✅ 4 modes, 5 permissions (e.g. plan+allow_read) | ❌ | ❌ |\n| Session Persistence | ✅ Deep session \u0026 checkpoint resume | ❌ | ❌ |\n| LSP code actions | ✅ (explain / fix / add tests / add docs) | ✅ (via prompt library) | ❌ |\n\n\u003c/details\u003e\n\n**When to choose copilot-agent.nvim**: you're committed to GitHub Copilot and want the deepest possible SDK integration: native tools, permission management, session persistence, sub-agent events, and LSP code actions, without routing through an intermediate CLI.\n\n---\n\n## Prerequisites\n\n- Go 1.24+ (only for building from source; pre-built binary skips this)\n- `curl` on `PATH`\n- GitHub Copilot CLI runtime (`@github/copilot/index.js`) or access via `-cli-url`\n- Neovim 0.12+ (with native autocomplete and LSP setup)\n\n**Optional:**\n\n- [`delta`](https://github.com/dandavison/delta) — rich diff viewer for permission \"Show diff\" (auto side-by-side for wide windows; falls back to builtin if not installed)\n- `pngpaste` / `wl-paste` / `xclip` — clipboard image paste\n- snacks.picker / telescope / fzf-lua / mini.pick — fuzzy file picker\n\nRun `:checkhealth copilot_agent` after installation to verify all requirements.\n\n\u003e [!NOTE]\n\u003e Seeing `service.command is empty — auto_start will not work` in `:checkhealth` can be a false warning when `service.command = nil` (runtime auto-detection). The plugin can still auto-detect `\u003cplugin_root\u003e/bin/copilot-agent` or fall back to `go run .` at runtime.\n\n---\n\n## Installation\n\n### Step 1 — Download the pre-built binary (recommended)\n\nRun this command inside Neovim after installing the plugin:\n\n```\n:CopilotAgentInstall\n```\n\nThis downloads the correct binary for your platform from the\n[latest GitHub release](https://github.com/ray-x/copilot-agent.nvim/releases/tag/latest)\nand saves it to `\u003cplugin_root\u003e/bin/copilot-agent` (or `copilot-agent.exe` on Windows).\nNo Go toolchain required.\n\nSupported platforms:\n\n| Platform | Binary |\n| ------------------- | --------------------------------- |\n| Linux x86_64 | `copilot-agent-linux-amd64` |\n| Linux aarch64 | `copilot-agent-linux-arm64` |\n| Windows x86_64 | `copilot-agent-windows-amd64.exe` |\n| Windows aarch64 | `copilot-agent-windows-arm64.exe` |\n| macOS Apple Silicon | `copilot-agent-darwin-arm64` |\n\nYou can also download manually from the\n[releases page](https://github.com/ray-x/copilot-agent.nvim/releases/tag/latest)\nand place it anywhere; then set `service.command = { \"/path/to/copilot-agent\" }`.\n\nOn Windows, the plugin automatically switches its local control channel from a Unix socket to a loopback TCP listener, so you do not need MSYS2 just to make the service start.\n\n### Step 2 — Plugin setup with lazy.nvim\n\nFor most users, this minimal setup is enough:\n\n```lua\n{\n \"ray-x/copilot-agent.nvim\",\n build = \":CopilotAgentInstall\",\n opts = function()\n require(\"copilot_agent\").setup({\n -- lsp = { enabled = true }, -- auto-starts the helper LSP from setup(); call start_lsp() only for explicit/delayed startup\n })\n end,\n}\n```\n\n`service.auto_start` defaults to `true`. For backward compatibility, a top-level `auto_start = ...` in `setup()` is still accepted and mapped to `service.auto_start` unless you explicitly set `service.auto_start`.\n\n\u003cdetails\u003e\n\u003csummary\u003e\u003cstrong\u003e📂 Full setup example with common options\u003c/strong\u003e\u003c/summary\u003e\n\n```lua\n{\n \"ray-x/copilot-agent.nvim\",\n build = \":CopilotAgentInstall\",\n config = function()\n require(\"copilot_agent\").setup({\n -- When auto_start=true the plugin connects to the shared Go service if\n -- one already exists, otherwise it starts exactly one and reads its\n -- port from stderr automatically. No manual base_url needed.\n base_url = \"http://127.0.0.1:8088\", -- only for externally-started services\n client_name = \"nvim-copilot\",\n default_provider = \"copilot\", -- or \"claude\"; Claude/LiteLLM uses ANTHROPIC_* from Neovim (or service.env) on session create/resume\n permission_mode = \"approve-all\", -- \"interactive\" | \"approve-all\" | \"autopilot\" | \"reject-all\"\n auto_create_session = true,\n lsp = {\n enabled = true, -- auto-start the helper LSP from setup()\n },\n session = {\n working_directory = function() return vim.fn.getcwd() end,\n model = nil, -- nil = Copilot picks a default\n agent = nil, -- nil = \"default\"; or \"coding\", \"gpt-4.1\", a custom agent name\n streaming = true,\n enable_config_discovery = true, -- respects .github/copilot-instructions.md etc.\n replay_permission_history = false, -- false (default) skips permission replay on resume for faster session loads\n history_turn_limit = 256, -- only replay the most recent N turns when opening a large session\n history_activity_turn_limit = 64, -- keep detailed activity only for the most recent N turns (\u003c= history_turn_limit)\n history_preview_chars = 120, -- truncation length for summarized historical activity/tool outputs\n auto_resume = \"prompt\", -- \"prompt\" (default) | \"auto\" — when multiple sessions exist\n },\n service = {\n auto_start = true,\n -- command = nil means auto: uses \u003cplugin_root\u003e/bin/copilot-agent if present,\n -- otherwise falls back to { \"go\", \"run\", \".\" } (requires Go toolchain).\n command = nil,\n cwd = nil, -- defaults to \u003cplugin_root\u003e/server\n detach = true, -- default: reuse one detached background service across Neovim instances\n port_range = nil, -- e.g. \"18000-19000\" for fixed range\n log = {\n enabled = false, -- set true to enable Go service file logging\n path = nil, -- nil = stdpath(\"state\") .. \"/copilot-agent-service.log\"\n },\n startup_timeout_ms = 15000,\n startup_poll_interval_ms = 250,\n },\n chat = {\n title = \"Copilot Chat\",\n system_notify_timeout = 3000, -- ms before auto-clearing transient notices\n render_markdown = true, -- set false to disable render-markdown.nvim (faster on long playbacks)\n protect_markdown_buffer = true, -- upstream Neovim Treesitter workaround for the prompt buffer; set false to disable\n diff_cmd = { 'delta' }, -- external diff viewer; false = builtin float\n diff_review = true, -- offer vimdiff after agent modifies a git-tracked file; clean buffers auto-reload, conflicting modified buffers prompt before reload\n activity_view = 'hover', -- 'hover' (default) opens a read-only preview via K (or CursorHold when enabled); 'diff' opens editable file diffs on \u003cCR\u003e; 'raw' keeps the patch-text viewer\n activity_diff_tool = 'native', -- 'native', 'diffview', 'fugitive', or a custom Vim command name\n -- Hover \u0026 preview controls:\n -- activity_hover_key: string (default: 'K') - key to toggle the read-only hover preview when activity_view='hover' while keeping focus in chat.\n -- activity_hover_focus_key: string (default: 'gK') - key to move focus into the current hover preview (opens it first if needed).\n -- activity_hover_cursor_hold: boolean (default: false) - when true, show hover on CursorHold/CursorHoldI instead of the hover key.\n -- activity_hover_timeout_ms: number (default: 2500) - auto-close timeout for hover preview in milliseconds (\u003c=0 disables auto-close).\n },\n prompt = {\n style = \"cold\", -- \"cold\" (default) = red-violet/violet/blue, \"warm\" = red/yellow/green\n },\n compose = {\n width = 0.4, -- left split width; fraction of chat width, or absolute columns\n min_width = 40,\n max_width = 100,\n promote_keymap = \"\u003cleader\u003ecc\", -- set false to disable the prompt-buffer promotion mapping\n },\n statusline = {\n enabled = false, -- default: keep plugin-owned chat/input local statuslines disabled\n components = { -- default: all true\n mode = true,\n permission = true,\n busy = true,\n session = true,\n model = true,\n tool = true,\n intent = true,\n context = true,\n config = true,\n attachments = true,\n help = true,\n },\n },\n notify = true, -- set false to silence all [copilot-agent] vim.notify calls\n file_log_level = \"WARN\", -- TRACE | DEBUG | INFO | WARN | ERROR; TRACE logs raw host/session payloads, DEBUG logs plugin actions and HTTP details to stdpath(\"log\") .. \"/copilot_agent.log\"\n file_log_batch = {\n enabled = true, -- queue file-log writes and flush in batches\n flush_interval_ms = 2000, -- flush pending log lines at least every 2 seconds\n max_entries = 20, -- flush immediately when queue reaches this size\n },\n })\n -- Start the combined HTTP + LSP service.\n -- lsp.enabled = true already starts the helper automatically from setup().\n -- Call explicitly here only if you want immediate or delayed startup control:\n require(\"copilot_agent\").start_lsp()\n end,\n}\n```\n\n`file_log_batch` controls buffered file logging behavior. Set `enabled = false` to restore immediate per-line writes.\n\n`service.log.enabled` defaults to `false`. When enabled, `service.log.path` overrides the log file location; when `path` is `nil`, the default is `stdpath(\"state\") .. \"/copilot-agent-service.log\"`.\n\n\u003c/details\u003e\n\n\u003cdetails\u003e\n\u003csummary\u003e\u003cstrong\u003e📂 Advanced service command and isolation examples\u003c/strong\u003e\u003c/summary\u003e\n\nIf you want to point at a binary in a custom location:\n\n```lua\n-- Dynamic port (recommended for multiple nvim instances)\nservice = { auto_start = true, command = { \"/path/to/copilot-agent\" } }\n\n-- Fixed port\nservice = { auto_start = true, command = { \"/path/to/copilot-agent\", \"--addr\", \"127.0.0.1:8088\" } }\n\n-- Port range (first free port in 18000–19000)\nservice = { auto_start = true, command = { \"/path/to/copilot-agent\" }, port_range = \"18000-19000\" }\n```\n\nWhen the plugin starts the service on Windows, it appends a localhost control address automatically (`--control-addr 127.0.0.1:0`) unless you already provided `--control-addr` or `--control-socket` yourself. Keep `service.command` as a list/table so those flags can be added.\n\n**Global service vs isolated instances**\n\nBy default, `auto_start = true` connects to the shared service if it is already running, otherwise it starts a single detached background service for your user account. In managed mode, the plugin first discovers the live HTTP address via the local control endpoint (`copilot-agent.sock` / `--control-addr`), then falls back to `stdpath(\"state\") .. \"/copilot-agent.addr\"` only when that saved address still belongs to a live shared service. Session resume is still matched by `session.working_directory`, but the service process and persisted session catalog are global by default. On quit, the last Neovim instance now requests detached-service shutdown in the background so exit does not wait on a slow control socket.\n\nThe helper LSP reuses that same shared service; it does not spawn a separate Copilot backend.\n\nIf you want **per-project isolation**, pin a project-specific address and disable detaching:\n\n```lua\nrequire(\"copilot_agent\").setup({\n base_url = \"http://127.0.0.1:18121\",\n session = {\n working_directory = function() return vim.fn.getcwd() end,\n },\n service = {\n auto_start = true,\n detach = false,\n port_range = \"18121-18121\",\n },\n})\n```\n\nReuse the same stable port when reopening the same project later. If you open multiple isolated projects at the same time, each one needs its own port.\n\n\u003c/details\u003e\n\n---\n\n## Running the Service Manually\n\nSee [server/README.md](server/README.md#running-the-service-manually) for manual startup, build flags, and service runtime details.\n\n### Startup troubleshooting\n\n- A `copilot-agent` binary alone is not enough: the GitHub Copilot CLI runtime must also be resolvable (`-cli-path`, environment, or `PATH`).\n- GUI Neovim launches (Finder/Spotlight/app launchers) often have a different `PATH` than terminal shells. If startup fails there but works in terminal Neovim, set explicit paths in your config (`service.command`, service env vars, or CLI path env) and re-run `:checkhealth copilot_agent`.\n\n---\n\n## Commands\n\nUse `:CopilotAgentDashboard` or `:CopilotAgentChat` to get started.\n\n\u003cdetails\u003e\n\u003csummary\u003e\u003cstrong\u003e📂 Full command reference\u003c/strong\u003e\u003c/summary\u003e\n\n| Command | Description |\n| ------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------ |\n| `:CopilotAgentInstall` | Download pre-built binary for the current platform |\n| `:CopilotAgentDashboard` | Open the Copilot Agent startup dashboard |\n| `:CopilotAgentChat [fullscreen]` | Open the chat buffer; `fullscreen` opens in a new tab |\n| `:CopilotAgentChatToggle` | Hide/show chat + input UI windows without reconnect/replay |\n| `:CopilotAgentChatFocus` | Focus or switch to an open chat buffer |\n| `:CopilotAgentAsk [prompt]` | Send a prompt; no argument opens the input buffer |\n| `:CopilotAgentMode [mode]` | Set the input mode (`ask`, `plan`, `agent`, `test`, `autopilot`) |\n| `:CopilotAgentCompose [tab]` | Open the compose scratch buffer; `tab` opens it in a new tab |\n| `:CopilotAgentPromoteToCompose` | Move current prompt-buffer text into compose |\n| `:CopilotAgentSendBuffer` | Send the active compose buffer |\n| `:CopilotAgentNewSession` | Disconnect current session and start a fresh one |\n| `:CopilotAgentSwitchSession` | Pick from all persisted sessions and switch |\n| `:CopilotAgentProvider [name]` | Switch provider; reuse a matching session or create a new one |\n| `:CopilotAgentDeleteSession` | Pick a session by summary + exact ID and delete it |\n| `:CopilotAgentModel [id]` | Pick or set a model; tab-completes from service model list |\n| `:CopilotAgentStart` | Start the Go service with the current config |\n| `:CopilotAgentServerStart` | Start the Copilot Agent server with the current config |\n| `:CopilotAgentStop` | Disconnect the active session |\n| `:CopilotAgentStop!` | Delete the active session; checkpoint cleanup waits 7 days |\n| `:CopilotAgentCancel` | Cancel the current agent turn |\n| `:CopilotAgentDiff` | Pick two checkpoints and open vimdiff for a changed file |\n| `:CopilotAgentFugitiveCommit [last]` | Generate a commit message and open fugitive commit; waits for final post-tool output; `last` reuses the latest assistant reply |\n| `:CopilotAgentStatus` | Show service URL, session id, stream status |\n| `:CopilotAgentLsp` | Start (or reuse) the LSP client for code actions |\n| `:CopilotAgentPasteImage` | Paste clipboard image as attachment |\n| `:CopilotAgentRetryInput` | Re-show the last dismissed ask_user prompt |\n\n\u003c/details\u003e\n\n---\n\n## Input Buffer\n\nOpen with `:CopilotAgentChat`, then press `i` or `\u003cEnter\u003e` in the chat buffer.\n\n### Keybindings\n\n| Key | Action |\n| --------------------------------- | ------------------------------------------------------------------------------------------------------- |\n| `\u003cCR\u003e` / `\u003cC-s\u003e` | Send message |\n| `q` / `\u003cEsc\u003e` | Close input (normal mode) |\n| `\u003cC-t\u003e` | Cycle chat mode: **💬 ask → 📋 plan → 🤖 agent → 🚀 autopilot** |\n| `\u003cM-m\u003e` | Open model picker |\n| `\u003cM-P\u003e` | Switch provider |\n| `\u003cM-a\u003e` | Cycle permission mode: **🔐 interactive → 📂 approve-reads → ✅ approve-all → 🤖 autopilot** |\n| `\u003cC-a\u003e` | Attach resource — opens picker menu (see below) |\n| `\u003cM-v\u003e` | Paste image from clipboard as attachment |\n| `\u003cC-x\u003e` | Toggle session tools (enable/disable individual tools) |\n| `\u003cTab\u003e` | Trigger completion (`@file` or `/command`) |\n| `\u003cC-e\u003e` | Dismiss the active completion popup |\n| `@\u003cpath\u003e` / `@\"path with spaces\"` | Attach a file by path or open buffer (autocomplete from working directory and named open buffers) |\n| `/\u003ccmd\u003e` | Run a built-in slash command (autocomplete with `\u003cTab\u003e`) |\n| `\u003cC-w\u003e` (insert) | Delete previous word in the prompt while keeping the mode prefix (`ask❯❯❯`/`plan❯❯❯`/`agent❯❯❯`) intact |\n| `\u003cC-u\u003e` (insert) | Delete from prompt start to cursor while keeping the mode prefix (`ask❯❯❯`/`plan❯❯❯`/`agent❯❯❯`) intact |\n| `\u003cC-p\u003e` / `\u003cM-p\u003e` | Previous prompt from history |\n| `\u003cC-n\u003e` / `\u003cM-n\u003e` | Next prompt from history |\n| `\u003cC-c\u003e` (output) | Cancel current turn |\n| `zA` (output) | Toggle collapsed `Activity:` transcript blocks |\n| `\u003cCR\u003e` (output) | On Activity lines, open the editable diff split; otherwise open input |\n| `K` (output) | Toggle the read-only Activity hover preview while keeping focus in chat |\n| `gK` (output) | Move focus into the current Activity hover preview (opens it first if needed) |\n| `\u003cC-w\u003ej` (output) | Move into Activity hover preview when available; otherwise fallback to normal window-down (`\u003cC-w\u003ej`) |\n| `CursorHold` / `CursorHoldI` | When `activity_hover_cursor_hold=true`, show the concise read-only hover preview on Activity lines |\n| `[[` / `]]` (output) | Jump to previous/next conversation (`Prompt:` block) |\n| `[a` / `]a` (output) | Jump to previous/next `Response:` or `Activity:` block |\n| `gT` (normal) | Open TODO float for the current turn |\n| `g?` (normal) | Show help float with keybindings, session commands, and recovery tips |\n\n### Slash Commands\n\nThe input buffer supports built-in slash commands handled by the plugin before the text is sent as a normal Copilot prompt. Type `/` and press `\u003cTab\u003e` to browse and complete the available commands. Press `\u003cC-e\u003e` while the popup is open to dismiss it without accepting a candidate. Some of the commands are still experimental.\n\n`/agent` completion is optimized for inline prompting: selecting an agent suggestion inserts the agent name without the `/agent` prefix, so `/agent Git Commit Agent` completion becomes `Git Commit Agent`.\n\n\u003cdetails\u003e\n\u003csummary\u003e\u003cstrong\u003e📂 Supported slash commands\u003c/strong\u003e\u003c/summary\u003e\n\n| Command | Arguments | What it does |\n| ---------------------- | ---------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |\n| `/add-dir` | `[path]` | Add a directory to the session's allowed-directory list; prompts if omitted |\n| `/agent` | `[name\\|default\\|clear]` | Pick a discovered custom agent, or reset back to the default agent |\n| `/allow-all` | — | Switch permission mode to `approve-all` |\n| `/ask` | `[question]` or `@visual-selection [question]` | Ask a side question in a temporary read-only session; `@visual-selection` attaches the current selection to the prompt |\n| `/clear` | — | Clear the current conversation and start a fresh session |\n| `/compact` | — | Compact session history and refresh context usage |\n| `/context` | — | Show current context-window token usage plus the latest quota and usage snapshot |\n| `/cwd` | `[path]` | Show or change the working directory used for future session actions |\n| `/diff` | `[vNNN\\|from to\\|from..to] [--difftool [name]]` | Show a checkpoint diff summary (default shows latest checkpoint changes), or open a visual diff via `--difftool` (`CodeDiff`, `Fugitive`, or native vim diff when no name is provided) |\n| `/env` | — | Show the current environment, service, session, model, and mode snapshot |\n| `/fleet` | `[prompt]` | Start fleet mode for the active session |\n| `/init` | `[args]` | Run the project initialization helper |\n| `/instructions` | `[name]` | Open a discovered instructions file from the repository |\n| `/list-dir` | — | List allowed directories for the current session |\n| `/list-dirs` | — | Alias for `/list-dir` |\n| `/list-tools` | — | List remembered tool approvals for the current session; if empty, explain that the backend does not expose a full available-tools inventory |\n| `/lsp` | `[create\\|status\\|show\\|test\\|help]` | Bootstrap or inspect project LSP config for Copilot CLI |\n| `/mcp` | `[add\\|show\\|edit\\|delete\\|disable\\|enable\\|reload] ...` | Manage MCP servers in `.mcp.json`, `.vscode/mcp.json`, and `~/.copilot/mcp-config.json` (global): add/show/edit/delete entries, toggle disabled state, and reconnect the active session to reload discovery |\n| `/model` | `[id]` | Pick or switch the active model |\n| `/new` | — | Start a fresh session |\n| `/plan` | `[draft text]` | Switch the input buffer to plan mode and optionally prefill the prompt |\n| `/rename` | `[name]` | Rename the active session |\n| `/research` | `[topic]` | Send a research-oriented prompt that can use GitHub and web sources |\n| `/reset-allowed-tools` | — | Clear remembered tool approvals for the current session |\n| `/resume` | `[session-id]` | Resume or switch to another saved session; prompts if omitted |\n| `/review` | `[focus]` | Ask Copilot to review the current changes, optionally with extra focus text |\n| `/rewind` | `[checkpoint]` | Rewind the session to a checkpoint such as `v003`, then queue the target git hash and reverted checkpoint summaries into the next Copilot prompt |\n| `/search` | `[text]` | Search the current transcript and jump to a matching entry |\n| `/session` | `[info\\|checkpoints\\|files\\|plan\\|rename\\|cleanup\\|prune\\|delete] ...` | Inspect and manage sessions; bare `/session` still opens the switch picker. `/session prune --older-than \u003cdays\u003e [--dry-run] [--include-named]` prunes old saved sessions, and `/session prune --keep-last \u003ccount\u003e [--session \u003cid\u003e] [--dry-run]` trims old checkpoint snapshots for a session |\n| `/share` | `[markdown\\|html] [path]` | Export the current transcript as Markdown or HTML |\n| `/skills` | `[name]` | Open a discovered skill from the repository |\n| `/tasks` | `[filter]` | Show background task and sub-agent activity |\n| `/undo` | — | Restore the latest checkpoint for the current session, then queue the restore git hash/context into the next Copilot prompt |\n| `/usage` | — | Show detailed session usage, quota snapshot, discovery counts, and context-window stats |\n\n\u003c/details\u003e\n\n## Compose Buffer\n\nOpen with `:CopilotAgentCompose` to open it as a left-side split next to chat output, or `:CopilotAgentCompose tab` to edit in a new tab.\n\nThe compose buffer is a separate markdown scratch buffer for long prompts. It keeps the same `@` attachment and slash-command completion as the normal input buffer. From the prompt buffer, press `\u003cleader\u003ecc` or run `:CopilotAgentPromoteToCompose` to promote the current prompt into compose and continue editing there. Customize or disable the mapping with `compose.promote_keymap`.\n\n| Key / Command | Action |\n| ------------------------------- | -------------------------------- |\n| `\u003cleader\u003ecc` (prompt buffer) | Move current prompt into compose |\n| `:CopilotAgentPromoteToCompose` | Move current prompt into compose |\n| `\u003cleader\u003ecs` / `\u003cC-s\u003e` | Send compose buffer |\n| `:w` / `:wq` | Send compose buffer |\n| `:CopilotAgentSendBuffer` | Send compose buffer |\n| `q` / `\u003cEsc\u003e` | Close compose buffer |\n\n### Attaching Files and Images\n\nPress `\u003cC-a\u003e` in the input buffer to open the resource picker:\n\n| Choice | What it does |\n| -------------------------- | -------------------------------------------------------------- |\n| Current buffer | Attaches the previously focused file buffer |\n| Visual selection | Attaches the current visual selection with file path + line range |\n| File | Opens fuzzy picker → select one or more files (multi-select) |\n| Folder | Opens fuzzy picker / `vim.ui.input` to pick a directory |\n| Instructions file | Same as File but marked as an instructions context file (📋) |\n| Image file | Opens fuzzy picker to select an image (png/jpg/gif/…) |\n| Paste image from clipboard | Saves clipboard image to a temp PNG and attaches it (🖼️) |\n\n`\u003cM-v\u003e` is a direct shortcut for **Paste image from clipboard** without opening the menu.\n`:CopilotAgentPasteImage` is the equivalent command.\n\nPending attachments appear in the input statusline as `📎 N`. Each attachment is also shown below the prompt text in the chat buffer once sent.\n\n#### Fuzzy Picker Integration\n\nThe File / Folder / Image / Instructions choices auto-detect and use the best available picker — no extra configuration needed:\n\n| Priority | Picker | Notes |\n| -------- | ----------------- | ------------------------------------------------------ |\n| 1 | **snacks.picker** | `snacks.picker.files` / `snacks.picker.directories` |\n| 2 | **telescope** | `find_files`; `telescope-file-browser` for directories |\n| 3 | **fzf-lua** | `fzf.files`; `fzf_exec fd --type d` for directories |\n| 4 | **mini.pick** | `mp.builtin.files` (files only) |\n| 5 | **vim.ui.input** | Fallback — type the path with completion |\n\nOverride the picker or force the native fallback via config:\n\n```lua\nchat = { file_picker = 'auto' } -- default: detect best available\nchat = { file_picker = 'telescope' } -- always use telescope\nchat = { file_picker = 'fzf-lua' } -- always use fzf-lua\nchat = { file_picker = 'native' } -- always use vim.ui.input\n```\n\n#### Clipboard Image Requirements\n\n| Platform | Tool required | Install |\n| ------------- | ----------------- | ------------------------------- |\n| macOS | `pngpaste` | `brew install pngpaste` |\n| Linux/Wayland | `wl-paste` | `sudo apt install wl-clipboard` |\n| Linux/X11 | `xclip` or `xsel` | `sudo apt install xclip` |\n\n### Chat Modes\n\nCycled with `\u003cC-t\u003e` in the chat/input buffer. The mode is shown in the statusline.\n\n| Mode | Icon | SDK session mode | Description |\n| ------------- | ---- | ---------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |\n| **ask** | 💬 | `interactive` | Single-turn Q\u0026A. The model answers once and stops. Tools are available but each call requires approval. Use for questions, explanations, and code reviews. |\n| **plan** | 📋 | `plan` | Like ask, but the model is guided to produce a structured implementation plan before taking any action. Good for scoping a task before committing to it. |\n| **agent** | 🤖 | `autopilot` | Multi-step agentic loop. The model calls tools (read files, run commands, write code) repeatedly and iterates until the task is done. Pauses to prompt you before writes/shell commands. Read access to the workspace is auto-approved. |\n| **autopilot** | 🚀 | `autopilot` | Same agentic loop as agent, but with `approve-all` permission — every tool call (reads _and_ writes) is silently approved. Fully autonomous. |\n\n**SDK note:** The Copilot SDK has three session modes — `interactive`, `plan`, and `autopilot`. VS Code's four-label model maps exactly onto these: ask → `interactive`, plan → `plan`, agent and autopilot both → `autopilot` (the difference is the permission mode, not the SDK mode).\n\n### Performance Tips\n\n- **render-markdown.nvim** integrates automatically when installed. On very long responses it can cause visible lag. Disable with `chat = { render_markdown = false }` to use treesitter highlighting only (much faster).\n- **Streaming** is enabled by default (`session.streaming = true`). The chat buffer updates incrementally as tokens arrive.\n- **Permission history replay** is filtered by default (`session.replay_permission_history = false`) so large sessions resume faster. Set it to `true` if you want to review historical permission requests/completions in the restored chat transcript.\n- **Large-session replay window** defaults to the latest 256 turns (`session.history_turn_limit`) and keeps full activity detail only for the latest 64 turns (`session.history_activity_turn_limit`). Older replayed turns keep full assistant messages, while activity/tool outputs are summarized with previews capped by `session.history_preview_chars` (default 120).\n- **Usage telemetry** (`assistant.usage`) updates statusline/context snapshots silently instead of adding extra `Activity:` transcript blocks. Use `/context` or `/usage` for full usage details.\n- **Session resume**: with 1 matching session it resumes silently; with multiple sessions `auto_resume = \"prompt\"` (default) shows a picker titled with the current project name and path so you can choose which session to continue.\n\n### Permission Modes\n\nCycled with `\u003cM-a\u003e` in the input buffer, or set via config / `POST /sessions/{id}/permission-mode`.\nEach chat mode sets a sensible default permission automatically; `\u003cM-a\u003e` overrides it for the current mode.\n\n| Icon | Mode | Behaviour |\n| ---- | ----------------- | ------------------------------------------------------------------------------------------------------------------- |\n| 🔐 | **interactive** | Neovim prompts `Allow / Deny` for every tool call |\n| 📂 | **approve-reads** | Workspace read-file requests are auto-approved; writes and shell commands still prompt. Default for **agent** mode. |\n| ✅ | **approve-all** | All tool calls silently approved. Default for **autopilot** mode. |\n| 🤖 | **autopilot** | Approve all + auto-answer any `ask_user` questions (fully autonomous) |\n| 🚫 | **reject-all** | Reject all tool calls (safe read-only mode) |\n\n#### `ask_user` requests — when will Copilot ask you a question?\n\nThere are **two separate interruption points** in an agentic loop:\n\n1. **Tool-call approval** — before the SDK executes a tool (read file, run shell, write code, etc.).\n Controlled entirely by the permission mode above.\n\n2. **`ask_user` requests** — the _model itself_ decides to pause and ask you a clarifying question\n mid-task (e.g. \"Which branch should I target?\" or \"There are two test files — which one?\").\n This is independent of tool approval and happens at the model's discretion.\n\n| Permission mode | Tool calls | `ask_user` questions |\n| ----------------- | ------------------- | ------------------------------------------ |\n| **interactive** | prompt every call | shown via `vim.ui.input/select` |\n| **approve-reads** | prompt writes/shell | shown via `vim.ui.input/select` |\n| **approve-all** | silent | shown via `vim.ui.input/select` |\n| **autopilot** | silent | **auto-answered** (first choice, silently) |\n| **reject-all** | all rejected | shown via `vim.ui.input/select` |\n\n\u003e **Practical rule:** use **agent** mode (`approve-reads`) for day-to-day tasks — Copilot will still\n\u003e ask you clarifying questions but won't prompt for every file read. Switch to **autopilot** permission\n\u003e only when you are confident in the task scope and want zero interruptions.\n\n---\n\n## LSP Code Actions\n\nThe Go binary runs an LSP server on stdio. Setting `lsp.enabled = true` starts the helper automatically from `setup()`. `:CopilotAgentLsp` or `require(\"copilot_agent\").start_lsp()` is only needed when you want explicit or delayed startup, and the helper reuses the existing HTTP service instead of spawning a fresh backend session.\n\nThis helper server is separate from the project language servers used by Copilot CLI for code intelligence. Use `/lsp create` to bootstrap `.github/lsp.json` from active Neovim project LSP clients, then restart the Copilot service so config discovery reloads it.\n\nAvailable code actions (triggered on any selection via `vim.lsp.buf.code_action()`):\n\n- **Explain selection** — Ask Copilot to explain the selected code\n- **Fix selection** — Ask Copilot to suggest a fix\n- **Add tests for selection** — Generate unit tests\n- **Add docs for selection** — Generate documentation\n\nThe action builds a prompt from the selected text, file path, and line range, then POSTs it to the active HTTP session.\n\n---\n\n## Statusline\n\nExpose Copilot state in your statusline. Each function returns a short string.\n\nBy default, plugin-managed chat/input local statuslines are disabled (`statusline.enabled = false`), so you can keep your own statusline setup unchanged.\n\nIf you want plugin-managed local statuslines in Copilot windows, enable them explicitly:\n\n```lua\nrequire(\"copilot_agent\").setup({\n statusline = {\n enabled = true,\n },\n})\n```\n\nYou can choose which components are rendered with `statusline.components` (default: all enabled). Supported component keys are: `mode`, `permission`, `busy`, `session`, `model`, `tool`, `intent`, `context`, `config`, `attachments`, `help`.\n\n```lua\nrequire(\"copilot_agent\").setup({\n statusline = {\n enabled = true,\n components = {\n mode = true,\n busy = true,\n session = true,\n permission = false,\n model = false,\n tool = false,\n intent = false,\n context = false,\n config = false,\n attachments = false,\n help = false,\n },\n },\n})\n```\n\nWhen enabled, default examples:\n\n```text\ninput window\n🤖agent (loop·approve-all) ✅approve-all ✅ready default 󱃕 Instruction: 0 󱜙 Agent: 0 󱨚 Skill: 0  MCP: 0 (g? for help)\n\nchat window\n🤖agent (loop·approve-all) ✅approve-all ✅ready default 󱃕 Instruction: 0 󱜙 Agent: 0 󱨚 Skill: 0  MCP: 0 session: [#project name 20260501 0905]\n```\n\nAs the session becomes active, the statusline updates live with readiness (`⏳working`, `📝sync`, `🧩2 tasks`, `❓input`, `✅ready`), the current tool/intent, context tokens, quota remaining from `assistant.usage` events, pending attachments, and the active session label.\n\n### Use statusline API\n\n```lua\n-- lualine\nrequire(\"lualine\").setup {\n sections = {\n lualine_x = {\n require(\"copilot_agent\").statusline_mode, -- [ask] / [plan] / [agent] / [autopilot]\n require(\"copilot_agent\").statusline_model, -- claude-sonnet-4.6 / default\n require(\"copilot_agent\").statusline_busy, -- ✅ready / ⏳working / 📝sync / 🧩2 tasks / ❓input\n require(\"copilot_agent\").statusline_permission, -- 🔐interactive / ✅approve-all / 🤖autopilot\n require(\"copilot_agent\").statusline_attachments, -- 📎3 (when attachments pending)\n require(\"copilot_agent\").statusline_tool, -- 🔧 read_file (active tool)\n require(\"copilot_agent\").statusline_intent, -- current agent intent\n require(\"copilot_agent\").statusline_context, -- 12k/200k plus quota remaining when available\n }\n }\n}\n\n-- heirline / \u0026statusline\n-- %{v:lua.require'copilot_agent'.statusline()}\n```\n\n---\n\n## Session Persistence\n\nSession resume is scoped per project by `working_directory`: `pick_or_create_session` filters persisted sessions by working directory, so opening a different project starts a fresh session. The auto-started service itself is still global by default, which is why the startup picker now shows the current project name and path explicitly. Use `:CopilotAgentNewSession` to force a new one in the same directory, `:CopilotAgentSwitchSession` to pick from all persisted sessions across projects, or `:CopilotAgentDeleteSession` to remove one from a newest-first picker that includes the exact session ID and marks the active session with `●`.\n\nSessions are auto-named by the SDK after the first conversation turn. You can rename a session by typing `/rename My Session Name` in the input buffer.\n\nWhen a session is deleted, its checkpoint git worktree is soft-deleted instead of being removed immediately. This applies both to `:CopilotAgentDeleteSession` and `:CopilotAgentStop!`. The checkpoint metadata records the deletion time and the worktree is pruned automatically after 7 days on the next plugin startup or checkpoint/session lifecycle operation.\n\nWhen plugin-managed statuslines are enabled, the chat/output statusline shows the active session summary together with its short session ID, and transcript separators between turns render the completed-turn checkpoint label (`v001`, `v002`, ...) as a virtual rule so you can copy it for rewind/recovery workflows without opening the input buffer.\n\n**Session selection behaviour:**\n\n| Matching sessions for project | Behaviour |\n| -------------------------------------- | ----------------------------------------------------------------- |\n| 0 | Creates a new session automatically |\n| 1 | Resumes it silently — no prompt |\n| 2+ (`auto_resume = 'prompt'`, default) | Shows `vim.ui.select` picker; most recent session is listed first |\n| 2+ (`auto_resume = 'auto'`) | Silently resumes the most recent session |\n\nTo always skip the picker, set in your config:\n\n```lua\nsession = { auto_resume = 'auto' }\n```\n\n---\n\n## HTTP API Reference\n\nSee [server/README.md](server/README.md#http-api-reference) for the full endpoint list, session permission modes, and SSE event reference.\n\n---\n\n## Tutorial\n\n1. 📖 **[Build a To-Do App with Copilot Agent](doc/tutorial-flask.md)** — Full-stack to-do list app with a REST API and responsive UI.\n2. 🌤️ **[Build a Weather Dashboard with Copilot Agent](doc/tutorial-weather-dashboard.md)** — Glassmorphism weather app with wttr.in data, animated UI, and backend tests.\n3. 🧩 **[Build a Weather Dashboard with Preset Copilot Agents](doc/tutorial-custom-agent-weather.md)** — Copy a ready-made `.github/agents/` pack and use separate UI, Python, and QA specialists.\n4. 🐦 **[Build a Terminal Flappy Bird with Custom Agents and Skills](doc/tutorial-flappy-bird-go.md)** — Bootstrap a repo with script-copied prompts/agents/skills, implement a Go terminal game, and run `go vet`/`go test` before launching from Neovim terminal.\n\n---\n\n## Included repo-local custom agents\n\n- **Go Quality Engineer** — runs the repository's existing Go quality checks\n- **Selene Lua Quality Engineer** — runs Selene against the Lua sources\n- **Code Review Engineer** — reviews Lua and Go changes for correctness, code quality, performance, and security\n- **Git Commit Agent** — inspects git status, runs repo-appropriate pre-commit checks, and prepares commit messages from the staged diff\n- **Document Update Agent** — reviews user-facing docs against recent plugin changes and updates commands, keymaps, changelog notes, and gotchas\n\nCustomize the commit agent's default checks and feedback rules in `.github/commit-agent.md`.\n\n---\n\n## Development\n\n### Formatting\n\nThe project uses `make` targets for formatting. Run from the repo root:\n\n```bash\nmake fmt # format everything (Lua + Go)\nmake fmt-lua # Lua only — stylua lua/ plugin/ (config: stylua.toml)\nmake fmt-go # Go only — gofmt -w ./server\n```\n\nRequirements: [`stylua`](https://github.com/JohnnyMorganz/StyLua) and `gofmt` (bundled with Go).\n\nCI enforces both formatters on every push — PRs with unformatted code will fail the **Lint** workflow.\n\n## Testing\n\n```bash\n# Lua unit tests (no Neovim required)\nbusted --lpath='lua/?.lua;lua/?/init.lua' tests/unit/\n\n# Neovim integration tests (requires nvim on PATH)\nnvim --headless -u tests/minimal_init.lua \\\n -c \"PlenaryBustedFile tests/integration/setup_spec.lua\"\n```\n\nCI runs all of the above automatically on push and PR via GitHub Actions.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fray-x%2Fcopilot-agent.nvim","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fray-x%2Fcopilot-agent.nvim","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fray-x%2Fcopilot-agent.nvim/lists"}