Skip to content

Commit 38a37d3

Browse files
SimonSchubertSimonSchubert
committed
Add commands
1 parent 50b9922 commit 38a37d3

2 files changed

Lines changed: 267 additions & 0 deletions

File tree

‎assets/commands/index.txt‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -5490,6 +5490,7 @@ oq.md
54905490
orbiton.md
54915491
orc-tools.md
54925492
orca-c.md
5493+
ordewell.md
54935494
ori.md
54945495
oryx.md
54955496
osage.md

‎assets/commands/ordewell.md‎

Lines changed: 266 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,266 @@
1+
# TAGLINE
2+
3+
Plan, run, and verify coding-agent tasks from a goal
4+
5+
# TLDR
6+
7+
**Open the terminal UI** (chat on the left, plan on the right)
8+
9+
```ordewell```
10+
11+
**Plan a goal** using a coding agent you already have installed (no API key)
12+
13+
```AI_PROVIDER=claude-code ordewell plan --goal "[Add rate limiting to the public API]"```
14+
15+
**Execute** the last generated plan
16+
17+
```ordewell run```
18+
19+
**Inspect** recent sessions, or one session in detail
20+
21+
```ordewell status```
22+
23+
```ordewell status --session-id [session-id]```
24+
25+
**Pick the planner** (Claude Code, Codex, OpenCode, or an API provider)
26+
27+
```ordewell planner [claude-code]```
28+
29+
**Reassign a task's runner** before anything runs
30+
31+
```ordewell task-runner [2] [opencode]```
32+
33+
**Start the local API server** in the background
34+
35+
```ordewell web --daemon```
36+
37+
**Print the installed version**
38+
39+
```ordewell --version```
40+
41+
# SYNOPSIS
42+
43+
**ordewell** [_command_] [_options_]
44+
45+
**ordewell** **plan** **--goal** _text_ [_--runner_ _id_] [_--workspace_ _path_] [_--no-chat_]
46+
47+
**ordewell** **run** [_--session-id_ _id_]
48+
49+
# PARAMETERS
50+
51+
**--help**, **-h**
52+
> Print usage. Every slash command in the TUI is also a CLI subcommand of the same name.
53+
54+
**--version**, **-v**, **version**
55+
> Print the installed version as a bare string.
56+
57+
**--workspace** _path_
58+
> Workspace directory (default: current directory). The directory must already look like a project (a `.git`, `.ordewell`, or similar marker), or the TUI will ask to initialize it.
59+
60+
**--port** _N_
61+
> Daemon port. Default is **ORDEWELL_PORT**, else **3742**. The CLI, TUI, and VS Code extension are clients of this local HTTP + WebSocket server.
62+
63+
**--session-id** _id_
64+
> Session to act on (default: last planned session in this workspace).
65+
66+
**--goal** _text_
67+
> Required for **plan**. Plain-language description of the work.
68+
69+
**--runner** _id_
70+
> Runner to consider when planning (repeatable). Built-in ids: **claude-code**, **codex**, **opencode**. Default is every enabled runner.
71+
72+
**--no-chat**
73+
> One-shot plan; skip the planner dialogue.
74+
75+
**--json**
76+
> JSON output for **status** and **sessions**.
77+
78+
**--daemon**
79+
> For **web**: run the API server in the background instead of the foreground.
80+
81+
**--server**
82+
> For **stop**: stop the background server daemon rather than a running plan.
83+
84+
# COMMANDS
85+
86+
**tui**
87+
> Full-screen terminal UI (the default when stdin is a TTY and no command is given). Requires **tmux**.
88+
89+
**plan**
90+
> Research the workspace read-only and produce a typed plan of tasks. Nothing executes until **run**.
91+
92+
**run**
93+
> Execute the last generated plan (or **--session-id**). Independent tasks run in parallel.
94+
95+
**approve**
96+
> Sign off a plan paused for review and continue it.
97+
98+
**stop**
99+
> Stop execution of the last session, or **--server** to stop the daemon.
100+
101+
**status**
102+
> List recent sessions, or show one session in full with **--session-id**.
103+
104+
**sessions** **list**|**load**|**delete**
105+
> Manage named sessions stored under `.ordewell/sessions/`.
106+
107+
**add-task** **--title** _text_
108+
> Add a task to the current plan (**--type** ai|user, **--depends-on** _id_, **--prompt** _text_).
109+
110+
**remove-task** _id_
111+
> Remove a task. _id_ is an order number or a task ID.
112+
113+
**complete** _id_
114+
> Mark a task complete so dependents can run (alias: **mark-complete**).
115+
116+
**uncomplete** _id_
117+
> Mark a completed task not done.
118+
119+
**skip** _id_
120+
> Skip a task (marks it complete so dependents can run).
121+
122+
**force-start** _id_
123+
> Start a task now, ignoring dependencies.
124+
125+
**run-task** _id_
126+
> Run only one task.
127+
128+
**retry** _id_
129+
> Re-run a failed task.
130+
131+
**cancel** _id_
132+
> Kill a running task.
133+
134+
**terminal** _id_
135+
> Attach a real terminal to a task's runner (a tmux window).
136+
137+
**task-runner** _id_ [_runner_]
138+
> Set a task's executor. Changing the runner re-derives its model, thinking effort, and mode. Omit the value to list options.
139+
140+
**task-model** _id_ [_model_]
141+
> Set a task's executor model.
142+
143+
**task-effort** _id_ [_level_]
144+
> Set a task's thinking effort (`default` to clear).
145+
146+
**task-mode** _id_ [_mode_]
147+
> Set a task's runner mode.
148+
149+
**task-deps** _id_ [_a,b_|**none**]
150+
> Set which earlier tasks a task waits for.
151+
152+
**planner** [_provider_]
153+
> Choose who plans: an API provider, or a coding-agent CLI (**claude-code**, **codex**, **opencode**) on a subscription you already hold.
154+
155+
**model** [**set** _id_]
156+
> Show or set the planner model. Applies without a restart.
157+
158+
**planner-effort** [_level_]
159+
> Thinking effort for a coding-agent planner.
160+
161+
**key** [**set** _provider_ _key_]
162+
> Show which providers have a key, or store one in `.env` (never echoed back).
163+
164+
**runners** [_id_ **on**|**off**]
165+
> Enable or disable runners.
166+
167+
**allowlist** **set**|**clear**|**show**
168+
> Limit which models a runner may use.
169+
170+
**auto** [**on**|**off**]
171+
> Autonomous permission mode for new sessions.
172+
173+
**refresh**
174+
> Re-discover runners and model catalogs.
175+
176+
**models**
177+
> List every provider's catalog (works without a server).
178+
179+
**tdd** [**on**|**off**]
180+
> Toggle Test-Driven Development mode, which augments tasks with red-green-refactor instructions.
181+
182+
**verify** [**on**|**off**]
183+
> Toggle verification mode, which appends a final evidence-based task that runs the full test suite.
184+
185+
**web**
186+
> Start the local API server on **127.0.0.1:3742** (JSON + WebSocket, not a browser dashboard). Every other command starts it on demand.
187+
188+
**setup**
189+
> Interactive first-run setup wizard.
190+
191+
**plugins** **list**|**install**|**remove**|**create**
192+
> Manage runner plugins. **create** scaffolds a `manifest.json`; **install** accepts `github:user/repo` (also GitLab, Bitbucket, Codeberg).
193+
194+
# DESCRIPTION
195+
196+
**ordewell** turns one goal into an ordered plan of coding-agent tasks, then executes and verifies them. Each task carries its own runner, model, thinking effort, and mode. You can rewrite any of those — add or remove tasks, rewire dependencies — without losing completed work or asking the planner to regenerate the whole plan.
197+
198+
Planning is a conversation, not a hidden internal state. The planner researches the workspace **read-only**: reads run in parallel, anything reaching outside the workspace asks once, and commands that would write are refused. Its final message *is* the plan. Claude Code, Codex, or OpenCode can be the planner on the subscription you already hold; alternatively, any of about twenty-five API providers (OpenRouter, Anthropic, OpenAI, Gemini, xAI, Groq, DeepSeek, and others) or an OpenAI-compatible local server can plan. Mutation always stays with the runners.
199+
200+
A task completes only when its unique **completion marker** appears in the runner's output. Exit code is kept as diagnostic evidence. The model is never the tie-breaker.
201+
202+
Bare `ordewell` on a TTY opens the TUI. Piped or scripted invocations without a command print help instead, because the TUI needs a real terminal. `tab` swaps the chat and plan panes; `/help` lists the rest. A separate VS Code extension (`ordewell.ordewell`) shares the same core and does not need the npm CLI.
203+
204+
# CONFIGURATION
205+
206+
Keys typed into **ordewell key set** or `/key` are masked on screen and written to `.env`. Each settings command pushes to the running server *before* writing the file, so a refused connection cannot leave a setting the daemon never saw.
207+
208+
**AI_PROVIDER**
209+
> Force the planner backend. `claude-code`, `codex`, and `opencode` plan with that CLI and need no API key. Otherwise auto-detected from whichever `*_API_KEY` is set.
210+
211+
**OPENROUTER_API_KEY**, **ANTHROPIC_API_KEY**, **GEMINI_API_KEY**, …
212+
> One provider key if you want an API planner rather than a coding agent. Run **ordewell key** for the full list of variable names. Anything else that speaks the OpenAI API works via **OPENAI_COMPATIBLE_BASE_URL**.
213+
214+
**ORCHESTRATOR_MODEL**
215+
> Planner model. Default: `deepseek/deepseek-v4-flash`. With a coding-agent planner it must be one of that agent's own model ids.
216+
217+
**ORDEWELL_PLANNER_EFFORT**
218+
> Thinking effort for a coding-agent planner (`low`, `high`, `adaptive`, …). Ignored by vendor planners, whose effort is baked into the model id.
219+
220+
**ORDEWELL_MAX_PARALLEL**
221+
> Max concurrent AI task sessions (1–5, default 3). The dependency graph is always respected.
222+
223+
**ORDEWELL_PORT**
224+
> Daemon port CLI commands target (default 3742).
225+
226+
**ORDEWELL_AUTONOMOUS_MODE**
227+
> Approval posture for new sessions (see **ordewell auto**).
228+
229+
**ORDEWELL_RESEARCH_ENABLED**
230+
> `true` (default) or `false`.
231+
232+
**ORDEWELL_TUI_MOUSE**
233+
> Set `false` (or run `/mouse off`) to give the terminal its own drag-to-select back. Capturing the mouse is what disables it.
234+
235+
**OPENROUTER_BASE_URL**
236+
> Default: `https://openrouter.ai/api/v1`.
237+
238+
Sessions auto-save to `.ordewell/sessions/` inside the workspace. The last-session pointer is `<workspace>/.ordewell/last-session.json`, not a machine-global file. VS Code mirrors these settings under `ordewell.*`.
239+
240+
# CAVEATS
241+
242+
Requires **Node.js 20** or newer. The TUI needs **tmux** on every platform — it is what backs each task's live terminal. On Windows, run the TUI under WSL; the CLI, API server, and VS Code extension run natively.
243+
244+
At least one coding agent (Claude Code, Codex, or OpenCode) must be installed to execute tasks. Planning without an API key uses that same agent in a read-only harness.
245+
246+
The local daemon binds **127.0.0.1**. A workspace without a project marker is refused rather than treated as a confinement boundary. Plugin installs run third-party manifests; only install plugins you trust.
247+
248+
Versions before **0.4.9** had disclosed issues around command classification, credential redaction, an unauthenticated daemon attack chain, and plugin-install code execution. Upgrade, and rotate any secrets that may have been read into a pre-0.4.9 session file.
249+
250+
# HISTORY
251+
252+
Ordewell is written by **Alessandro Costanzo Ciano** and released under the **Apache License 2.0** (the name and logos are not covered by that licence). First public release was **0.4.0** on **31 July 2026**. The npm package is `@ordewell/cli` (also published as the unscoped name `ordewell`).
253+
254+
# SEE ALSO
255+
256+
[claude](/man/claude)(1), [codex](/man/codex)(1), [opencode](/man/opencode)(1), [aider](/man/aider)(1), [tmux](/man/tmux)(1), [npm](/man/npm)(1)
257+
258+
# RESOURCES
259+
260+
```[Source code](https://github.com/ordewell/ordewell)```
261+
262+
```[Homepage](https://ordewell.ai)```
263+
264+
```[Documentation](https://ordewell.ai/docs.html)```
265+
266+
<!-- verified: 2026-09-15 -->

0 commit comments

Comments
 (0)