blob: 65327c33ffcf8ed9a11a8bf096ca07de2948b2ea (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
|
---
title: Agent skill
description: A concise guide for coding agents running inside Herdr.
---
This page teaches coding agents how to use Herdr while they are running inside a Herdr pane.
Use Herdr to organize parallel work instead of stacking everything in one terminal.
## Core model
Herdr has four levels:
| Level | Meaning |
| --- | --- |
| Workspace | A project or task context. |
| Tab | A view inside a workspace. |
| Pane | A terminal area inside a tab. |
| Agent | A detected coding agent running in a pane. |
Use workspace and tab actions for layout. Use pane actions for commands and output.
## Read the current layout
```bash
herdr pane list
```
Use the pane IDs returned by Herdr for later commands.
## Create a helper pane
Split a pane and run a command:
```bash
herdr pane split 1-1 --direction right
herdr pane run 1-2 "just test"
```
Prefer `pane run` over sending raw keys when you want to submit a full command line.
## Read output
```bash
herdr pane read 1-2 --source recent --lines 80
```
Use `recent-unwrapped` when reading logs or matching long lines.
```bash
herdr pane read 1-2 --source recent-unwrapped --lines 120
```
## Wait for agents
Wait for another agent to finish or become blocked:
```bash
herdr wait agent-status 1-3 --status done
herdr wait agent-status 1-3 --status blocked
```
Use agent waits for recognized coding agents. Use normal output matching for ordinary commands, servers, and tests.
## Preserve focus
Do not steal visible focus unless the user asked for it. Create panes, submit commands, and read output without changing the user’s active view when possible.
## Use clear pane aliases
When your tool supports aliases, use names like `server`, `tests`, `reviewer`, or `docs`. This makes later pane operations easier to audit.
## Good patterns
Run long work in a separate pane:
```bash
herdr pane split 1-1 --direction right
herdr pane run 1-2 "just check"
```
Watch a server until it is ready:
```bash
herdr pane read 1-2 --source recent-unwrapped --lines 80
```
Ask a reviewer agent to inspect a change:
```bash
herdr pane split 1-1 --direction right
herdr pane run 1-2 "pi --model <review-model>"
```
## Avoid these patterns
Do not pile every command into the current pane.
Do not use raw key injection for command submission when `pane run` is available.
Do not wait for `agent-status` on a normal process that is not a coding agent.
Do not change implementation direction without telling the user.
|
