prompt-language-shell 0.6.8 → 0.7.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/index.js +2 -1
- package/dist/services/anthropic.js +20 -1
- package/dist/services/components.js +11 -0
- package/dist/services/configuration.js +23 -5
- package/dist/services/logger.js +64 -0
- package/dist/services/messages.js +2 -2
- package/dist/services/refinement.js +4 -0
- package/dist/services/skills.js +1 -2
- package/dist/skills/answer.md +10 -9
- package/dist/skills/config.md +16 -9
- package/dist/skills/execute.md +91 -53
- package/dist/skills/introspect.md +63 -47
- package/dist/skills/plan.md +419 -337
- package/dist/skills/validate.md +26 -12
- package/dist/types/types.js +1 -0
- package/dist/ui/Command.js +4 -0
- package/dist/ui/Component.js +3 -0
- package/dist/ui/Config.js +3 -2
- package/dist/ui/Debug.js +8 -0
- package/dist/ui/Introspect.js +3 -2
- package/dist/ui/Main.js +19 -5
- package/dist/ui/Plan.js +3 -2
- package/dist/ui/Validate.js +2 -2
- package/dist/ui/Workflow.js +2 -1
- package/package.json +1 -1
package/dist/index.js
CHANGED
|
@@ -4,6 +4,7 @@ import { existsSync, readFileSync } from 'fs';
|
|
|
4
4
|
import { dirname, join } from 'path';
|
|
5
5
|
import { fileURLToPath } from 'url';
|
|
6
6
|
import { render } from 'ink';
|
|
7
|
+
import { DebugLevel } from './services/configuration.js';
|
|
7
8
|
import { Main } from './ui/Main.js';
|
|
8
9
|
const __filename = fileURLToPath(import.meta.url);
|
|
9
10
|
const __dirname = dirname(__filename);
|
|
@@ -20,7 +21,7 @@ const app = {
|
|
|
20
21
|
version: packageJson.version,
|
|
21
22
|
description: packageJson.description,
|
|
22
23
|
isDev,
|
|
23
|
-
|
|
24
|
+
debug: DebugLevel.None,
|
|
24
25
|
};
|
|
25
26
|
// Get command from command-line arguments
|
|
26
27
|
const args = process.argv.slice(2);
|
|
@@ -1,5 +1,6 @@
|
|
|
1
1
|
import Anthropic from '@anthropic-ai/sdk';
|
|
2
2
|
import { getAvailableConfigStructure, getConfiguredKeys, } from './configuration.js';
|
|
3
|
+
import { logPrompt, logResponse } from './logger.js';
|
|
3
4
|
import { formatSkillsForPrompt, loadSkillsWithValidation } from './skills.js';
|
|
4
5
|
import { toolRegistry } from './registry.js';
|
|
5
6
|
/**
|
|
@@ -70,7 +71,7 @@ export class AnthropicService {
|
|
|
70
71
|
if (toolName === 'config') {
|
|
71
72
|
const configStructure = getAvailableConfigStructure();
|
|
72
73
|
const configuredKeys = getConfiguredKeys();
|
|
73
|
-
const configSection = '\n
|
|
74
|
+
const configSection = '\n## Available Configuration\n\n' +
|
|
74
75
|
'Config structure (key: description):\n' +
|
|
75
76
|
JSON.stringify(configStructure, null, 2) +
|
|
76
77
|
'\n\nConfigured keys (keys that exist in config file):\n' +
|
|
@@ -85,7 +86,15 @@ export class AnthropicService {
|
|
|
85
86
|
name: 'web_search',
|
|
86
87
|
});
|
|
87
88
|
}
|
|
89
|
+
// Collect debug components
|
|
90
|
+
const debug = [];
|
|
91
|
+
// Log prompt at Verbose level
|
|
92
|
+
const promptDebug = logPrompt(toolName, command, systemPrompt);
|
|
93
|
+
if (promptDebug) {
|
|
94
|
+
debug.push(promptDebug);
|
|
95
|
+
}
|
|
88
96
|
// Call API with tool
|
|
97
|
+
const startTime = Date.now();
|
|
89
98
|
const response = await this.client.messages.create({
|
|
90
99
|
model: this.model,
|
|
91
100
|
max_tokens: 1024,
|
|
@@ -99,6 +108,12 @@ export class AnthropicService {
|
|
|
99
108
|
},
|
|
100
109
|
],
|
|
101
110
|
});
|
|
111
|
+
const duration = Date.now() - startTime;
|
|
112
|
+
// Log response at Verbose level
|
|
113
|
+
const responseDebug = logResponse(toolName, response, duration);
|
|
114
|
+
if (responseDebug) {
|
|
115
|
+
debug.push(responseDebug);
|
|
116
|
+
}
|
|
102
117
|
// Check for truncation
|
|
103
118
|
if (response.stop_reason === 'max_tokens') {
|
|
104
119
|
throw new Error('Response was truncated due to length. Please simplify your request or break it into smaller parts.');
|
|
@@ -113,6 +128,7 @@ export class AnthropicService {
|
|
|
113
128
|
message: '',
|
|
114
129
|
tasks: [],
|
|
115
130
|
answer: cleanAnswerText(textContent.text),
|
|
131
|
+
debug,
|
|
116
132
|
};
|
|
117
133
|
}
|
|
118
134
|
}
|
|
@@ -143,6 +159,7 @@ export class AnthropicService {
|
|
|
143
159
|
message: input.message,
|
|
144
160
|
tasks: [],
|
|
145
161
|
commands: input.commands,
|
|
162
|
+
debug,
|
|
146
163
|
};
|
|
147
164
|
}
|
|
148
165
|
// Handle answer tool response
|
|
@@ -157,6 +174,7 @@ export class AnthropicService {
|
|
|
157
174
|
message: '',
|
|
158
175
|
tasks: [],
|
|
159
176
|
answer: cleanAnswerText(input.answer),
|
|
177
|
+
debug,
|
|
160
178
|
};
|
|
161
179
|
}
|
|
162
180
|
// Handle plan and introspect tool responses
|
|
@@ -175,6 +193,7 @@ export class AnthropicService {
|
|
|
175
193
|
return {
|
|
176
194
|
message: input.message,
|
|
177
195
|
tasks: input.tasks,
|
|
196
|
+
debug,
|
|
178
197
|
};
|
|
179
198
|
}
|
|
180
199
|
}
|
|
@@ -252,6 +252,17 @@ export function createMessage(text) {
|
|
|
252
252
|
},
|
|
253
253
|
};
|
|
254
254
|
}
|
|
255
|
+
export function createDebugDefinition(title, content, color) {
|
|
256
|
+
return {
|
|
257
|
+
id: randomUUID(),
|
|
258
|
+
name: ComponentName.Debug,
|
|
259
|
+
props: {
|
|
260
|
+
title,
|
|
261
|
+
content,
|
|
262
|
+
color,
|
|
263
|
+
},
|
|
264
|
+
};
|
|
265
|
+
}
|
|
255
266
|
export function createRefinement(text, onAborted) {
|
|
256
267
|
return {
|
|
257
268
|
id: randomUUID(),
|
|
@@ -9,6 +9,13 @@ export var AnthropicModel;
|
|
|
9
9
|
AnthropicModel["Opus"] = "claude-opus-4-1";
|
|
10
10
|
})(AnthropicModel || (AnthropicModel = {}));
|
|
11
11
|
export const SUPPORTED_MODELS = Object.values(AnthropicModel);
|
|
12
|
+
export var DebugLevel;
|
|
13
|
+
(function (DebugLevel) {
|
|
14
|
+
DebugLevel["None"] = "none";
|
|
15
|
+
DebugLevel["Info"] = "info";
|
|
16
|
+
DebugLevel["Verbose"] = "verbose";
|
|
17
|
+
})(DebugLevel || (DebugLevel = {}));
|
|
18
|
+
export const SUPPORTED_DEBUG_LEVELS = Object.values(DebugLevel);
|
|
12
19
|
export var ConfigDefinitionType;
|
|
13
20
|
(function (ConfigDefinitionType) {
|
|
14
21
|
ConfigDefinitionType["RegExp"] = "regexp";
|
|
@@ -62,8 +69,17 @@ function validateConfig(parsed) {
|
|
|
62
69
|
if (config.settings && typeof config.settings === 'object') {
|
|
63
70
|
const settings = config.settings;
|
|
64
71
|
validatedConfig.settings = {};
|
|
65
|
-
if ('debug' in settings
|
|
66
|
-
|
|
72
|
+
if ('debug' in settings) {
|
|
73
|
+
// Handle migration from boolean to enum
|
|
74
|
+
if (typeof settings.debug === 'boolean') {
|
|
75
|
+
validatedConfig.settings.debug = settings.debug
|
|
76
|
+
? DebugLevel.Info
|
|
77
|
+
: DebugLevel.None;
|
|
78
|
+
}
|
|
79
|
+
else if (typeof settings.debug === 'string' &&
|
|
80
|
+
SUPPORTED_DEBUG_LEVELS.includes(settings.debug)) {
|
|
81
|
+
validatedConfig.settings.debug = settings.debug;
|
|
82
|
+
}
|
|
67
83
|
}
|
|
68
84
|
}
|
|
69
85
|
return validatedConfig;
|
|
@@ -139,10 +155,10 @@ export function saveDebugSetting(debug) {
|
|
|
139
155
|
export function loadDebugSetting() {
|
|
140
156
|
try {
|
|
141
157
|
const config = loadConfig();
|
|
142
|
-
return config.settings?.debug ??
|
|
158
|
+
return config.settings?.debug ?? DebugLevel.None;
|
|
143
159
|
}
|
|
144
160
|
catch {
|
|
145
|
-
return
|
|
161
|
+
return DebugLevel.None;
|
|
146
162
|
}
|
|
147
163
|
}
|
|
148
164
|
/**
|
|
@@ -193,8 +209,10 @@ const coreConfigSchema = {
|
|
|
193
209
|
description: 'Anthropic model',
|
|
194
210
|
},
|
|
195
211
|
'settings.debug': {
|
|
196
|
-
type: ConfigDefinitionType.
|
|
212
|
+
type: ConfigDefinitionType.Enum,
|
|
197
213
|
required: false,
|
|
214
|
+
values: SUPPORTED_DEBUG_LEVELS,
|
|
215
|
+
default: DebugLevel.None,
|
|
198
216
|
description: 'Debug mode',
|
|
199
217
|
},
|
|
200
218
|
};
|
|
@@ -0,0 +1,64 @@
|
|
|
1
|
+
import { createDebugDefinition } from './components.js';
|
|
2
|
+
import { DebugLevel, loadDebugSetting } from './configuration.js';
|
|
3
|
+
import { Palette } from './colors.js';
|
|
4
|
+
/**
|
|
5
|
+
* Debug logger for the application
|
|
6
|
+
* Logs information based on the current debug level setting
|
|
7
|
+
*/
|
|
8
|
+
let currentDebugLevel = DebugLevel.None;
|
|
9
|
+
/**
|
|
10
|
+
* Initialize the logger with the current debug level from config
|
|
11
|
+
*/
|
|
12
|
+
export function initializeLogger() {
|
|
13
|
+
currentDebugLevel = loadDebugSetting();
|
|
14
|
+
}
|
|
15
|
+
/**
|
|
16
|
+
* Set the debug level (used for testing or runtime changes)
|
|
17
|
+
*/
|
|
18
|
+
export function setDebugLevel(debug) {
|
|
19
|
+
currentDebugLevel = debug;
|
|
20
|
+
}
|
|
21
|
+
/**
|
|
22
|
+
* Get the current debug level
|
|
23
|
+
*/
|
|
24
|
+
export function getDebugLevel() {
|
|
25
|
+
return currentDebugLevel;
|
|
26
|
+
}
|
|
27
|
+
/**
|
|
28
|
+
* Create debug component for system prompts sent to the LLM
|
|
29
|
+
* Only creates at Verbose level
|
|
30
|
+
*/
|
|
31
|
+
export function logPrompt(toolName, command, instructions) {
|
|
32
|
+
if (currentDebugLevel !== DebugLevel.Verbose) {
|
|
33
|
+
return null;
|
|
34
|
+
}
|
|
35
|
+
const content = [
|
|
36
|
+
'',
|
|
37
|
+
`Tool: ${toolName}`,
|
|
38
|
+
`Command: ${command}`,
|
|
39
|
+
'',
|
|
40
|
+
instructions,
|
|
41
|
+
].join('\n');
|
|
42
|
+
// Calculate stats for the instructions
|
|
43
|
+
const lines = instructions.split('\n').length;
|
|
44
|
+
const bytes = Buffer.byteLength(instructions, 'utf-8');
|
|
45
|
+
const title = `SYSTEM PROMPT (${String(lines)} lines, ${String(bytes)} bytes)`;
|
|
46
|
+
return createDebugDefinition(title, content, Palette.Gray);
|
|
47
|
+
}
|
|
48
|
+
/**
|
|
49
|
+
* Create debug component for LLM responses received
|
|
50
|
+
* Only creates at Verbose level
|
|
51
|
+
*/
|
|
52
|
+
export function logResponse(toolName, response, durationMs) {
|
|
53
|
+
if (currentDebugLevel !== DebugLevel.Verbose) {
|
|
54
|
+
return null;
|
|
55
|
+
}
|
|
56
|
+
const content = [
|
|
57
|
+
'',
|
|
58
|
+
`Tool: ${toolName}`,
|
|
59
|
+
'',
|
|
60
|
+
JSON.stringify(response, null, 2),
|
|
61
|
+
].join('\n');
|
|
62
|
+
const title = `LLM RESPONSE (${String(durationMs)} ms)`;
|
|
63
|
+
return createDebugDefinition(title, content, Palette.AshGray);
|
|
64
|
+
}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { loadDebugSetting } from './configuration.js';
|
|
1
|
+
import { DebugLevel, loadDebugSetting } from './configuration.js';
|
|
2
2
|
export { formatDuration } from './utils.js';
|
|
3
3
|
/**
|
|
4
4
|
* Returns a natural language confirmation message for plan execution.
|
|
@@ -96,7 +96,7 @@ export const FeedbackMessages = {
|
|
|
96
96
|
*/
|
|
97
97
|
export function formatErrorMessage(error) {
|
|
98
98
|
const rawMessage = error instanceof Error ? error.message : 'Unknown error occurred';
|
|
99
|
-
if (loadDebugSetting()) {
|
|
99
|
+
if (loadDebugSetting() !== DebugLevel.None) {
|
|
100
100
|
return rawMessage;
|
|
101
101
|
}
|
|
102
102
|
// Try to extract message from Anthropic API error format
|
|
@@ -24,6 +24,10 @@ export async function handleRefinement(selectedTasks, service, originalCommand,
|
|
|
24
24
|
const refinedResult = await service.processWithTool(refinedCommand, 'plan');
|
|
25
25
|
// Complete the Refinement component
|
|
26
26
|
handlers.completeActive();
|
|
27
|
+
// Add debug components to timeline if present
|
|
28
|
+
if (refinedResult.debug && refinedResult.debug.length > 0) {
|
|
29
|
+
handlers.addToTimeline(...refinedResult.debug);
|
|
30
|
+
}
|
|
27
31
|
// Route refined tasks to appropriate components
|
|
28
32
|
routeTasksWithConfirm(refinedResult.tasks, refinedResult.message, service, originalCommand, handlers, false // No DEFINE tasks in refined result
|
|
29
33
|
);
|
package/dist/services/skills.js
CHANGED
|
@@ -131,7 +131,6 @@ export function formatSkillsForPrompt(skills) {
|
|
|
131
131
|
return '';
|
|
132
132
|
}
|
|
133
133
|
const header = `
|
|
134
|
-
|
|
135
134
|
## Available Skills
|
|
136
135
|
|
|
137
136
|
The following skills define domain-specific workflows. When the user's
|
|
@@ -147,7 +146,7 @@ brackets for additional information. Use commas instead. For example:
|
|
|
147
146
|
- WRONG: "Build project Alpha (the legacy version)"
|
|
148
147
|
|
|
149
148
|
`;
|
|
150
|
-
const skillsContent = skills.join('\n\n');
|
|
149
|
+
const skillsContent = skills.map((s) => s.trim()).join('\n\n');
|
|
151
150
|
return header + skillsContent;
|
|
152
151
|
}
|
|
153
152
|
/**
|
package/dist/skills/answer.md
CHANGED
|
@@ -5,9 +5,9 @@ command-line concierge. Your role is to **answer questions** and provide
|
|
|
5
5
|
up-to-date information when a task with type "answer" has been planned and
|
|
6
6
|
confirmed.
|
|
7
7
|
|
|
8
|
-
**IMPORTANT**: Use web search to find current, accurate information. This
|
|
9
|
-
is designed for quick answers from the terminal without needing to open a
|
|
10
|
-
browser. Always search for the latest data rather than relying solely on
|
|
8
|
+
**IMPORTANT**: Use web search to find current, accurate information. This
|
|
9
|
+
tool is designed for quick answers from the terminal without needing to open a
|
|
10
|
+
web browser. Always search for the latest data rather than relying solely on
|
|
11
11
|
training data.
|
|
12
12
|
|
|
13
13
|
## Execution Flow
|
|
@@ -60,11 +60,11 @@ TypeScript code compiles to JavaScript and runs anywhere JavaScript runs.
|
|
|
60
60
|
|
|
61
61
|
Bad answer (too verbose):
|
|
62
62
|
```
|
|
63
|
-
TypeScript is a strongly typed programming language that builds on
|
|
64
|
-
giving you better tooling at any scale. TypeScript adds
|
|
65
|
-
JavaScript to support a tighter integration with your
|
|
66
|
-
early in development by checking types. TypeScript
|
|
67
|
-
which runs anywhere.
|
|
63
|
+
TypeScript is a strongly typed programming language that builds on
|
|
64
|
+
JavaScript, giving you better tooling at any scale. TypeScript adds
|
|
65
|
+
additional syntax to JavaScript to support a tighter integration with your
|
|
66
|
+
editor. It catches errors early in development by checking types. TypeScript
|
|
67
|
+
code converts to JavaScript which runs anywhere.
|
|
68
68
|
```
|
|
69
69
|
|
|
70
70
|
### Example 2: Technical explanation
|
|
@@ -101,7 +101,8 @@ They enable cleaner, more reusable component logic.
|
|
|
101
101
|
|
|
102
102
|
## Guidelines
|
|
103
103
|
|
|
104
|
-
1. **Be direct**: Answer the question immediately, don't introduce your
|
|
104
|
+
1. **Be direct**: Answer the question immediately, don't introduce your
|
|
105
|
+
answer
|
|
105
106
|
2. **Be accurate**: Provide correct, factual information
|
|
106
107
|
3. **Be concise**: Respect the 4-line, 80-character constraints strictly
|
|
107
108
|
4. **Be helpful**: Focus on what the user needs to know
|
package/dist/skills/config.md
CHANGED
|
@@ -1,33 +1,40 @@
|
|
|
1
1
|
## Overview
|
|
2
2
|
|
|
3
|
-
You are the CONFIG tool for "pls" (please), a professional command-line
|
|
4
|
-
Your role is to determine which configuration settings the user
|
|
5
|
-
based on their query.
|
|
3
|
+
You are the CONFIG tool for "pls" (please), a professional command-line
|
|
4
|
+
concierge. Your role is to determine which configuration settings the user
|
|
5
|
+
wants to configure based on their query.
|
|
6
6
|
|
|
7
7
|
## Input
|
|
8
8
|
|
|
9
9
|
You will receive:
|
|
10
|
-
- `configStructure`: Object mapping config keys to descriptions (e.g.,
|
|
11
|
-
|
|
10
|
+
- `configStructure`: Object mapping config keys to descriptions (e.g.,
|
|
11
|
+
{"anthropic.key": "Anthropic API key", "settings.debug": "Debug mode
|
|
12
|
+
(optional)"})
|
|
13
|
+
- `configuredKeys`: Array of keys that exist in the user's config file
|
|
14
|
+
(e.g., ["anthropic.key", "anthropic.model", "settings.debug"])
|
|
12
15
|
- `query`: User's request (e.g., "app", "mode", "anthropic", or empty)
|
|
13
16
|
|
|
14
17
|
## Task
|
|
15
18
|
|
|
16
|
-
Determine which config keys the user wants to configure and return them
|
|
19
|
+
Determine which config keys the user wants to configure and return them
|
|
20
|
+
as tasks.
|
|
17
21
|
|
|
18
22
|
## Mapping Rules
|
|
19
23
|
|
|
20
24
|
### Query: "app" or empty/unclear
|
|
21
25
|
- Return all **required** config keys (those needed for the app to work)
|
|
22
|
-
- Also include any keys marked as "(optional)" that appear in
|
|
23
|
-
|
|
26
|
+
- Also include any keys marked as "(optional)" that appear in
|
|
27
|
+
`configuredKeys` (optional settings that exist in user's config file)
|
|
28
|
+
- Also include any keys marked as "(discovered)" (they exist in user's
|
|
29
|
+
config file but aren't in schema)
|
|
24
30
|
- Required keys: `anthropic.key`, `anthropic.model`
|
|
25
31
|
|
|
26
32
|
### Query: "mode"
|
|
27
33
|
- Return only: `settings.debug`
|
|
28
34
|
|
|
29
35
|
### Query: "anthropic"
|
|
30
|
-
- Return all keys starting with `anthropic.` (usually `anthropic.key` and
|
|
36
|
+
- Return all keys starting with `anthropic.` (usually `anthropic.key` and
|
|
37
|
+
`anthropic.model`)
|
|
31
38
|
|
|
32
39
|
### Other queries
|
|
33
40
|
- Match the query against config key names and descriptions
|
package/dist/skills/execute.md
CHANGED
|
@@ -1,8 +1,8 @@
|
|
|
1
1
|
## Overview
|
|
2
2
|
|
|
3
|
-
You are the execution component of "pls" (please), a professional
|
|
4
|
-
concierge. Your role is to **execute shell commands** and
|
|
5
|
-
with type "execute" have been planned and confirmed.
|
|
3
|
+
You are the execution component of "pls" (please), a professional
|
|
4
|
+
command-line concierge. Your role is to **execute shell commands** and
|
|
5
|
+
operations when tasks with type "execute" have been planned and confirmed.
|
|
6
6
|
|
|
7
7
|
## Execution Flow
|
|
8
8
|
|
|
@@ -11,23 +11,23 @@ This tool is invoked AFTER:
|
|
|
11
11
|
2. User reviewed and confirmed the plan
|
|
12
12
|
3. The execute tasks are now being executed
|
|
13
13
|
|
|
14
|
-
Your task is to translate the planned actions into specific shell commands
|
|
15
|
-
can be run in the terminal.
|
|
14
|
+
Your task is to translate the planned actions into specific shell commands
|
|
15
|
+
that can be run in the terminal.
|
|
16
16
|
|
|
17
17
|
## Input
|
|
18
18
|
|
|
19
19
|
You will receive:
|
|
20
20
|
- An array of tasks with their actions and parameters
|
|
21
|
-
- Each task describes what needs to be done (e.g., "Create a new file
|
|
22
|
-
test.txt", "List files in the current directory")
|
|
21
|
+
- Each task describes what needs to be done (e.g., "Create a new file
|
|
22
|
+
called test.txt", "List files in the current directory")
|
|
23
23
|
- Tasks may include params with specific values (paths, filenames, etc.)
|
|
24
|
-
- Tasks from user-defined skills include params.skill (skill name) and
|
|
25
|
-
values that were substituted into the action
|
|
24
|
+
- Tasks from user-defined skills include params.skill (skill name) and
|
|
25
|
+
parameter values that were substituted into the action
|
|
26
26
|
|
|
27
27
|
## Skill-Based Command Generation
|
|
28
28
|
|
|
29
|
-
**CRITICAL**: When tasks originate from a user-defined skill, you MUST use
|
|
30
|
-
skill's **Execution** section to generate commands, NOT invent your own.
|
|
29
|
+
**CRITICAL**: When tasks originate from a user-defined skill, you MUST use
|
|
30
|
+
the skill's **Execution** section to generate commands, NOT invent your own.
|
|
31
31
|
|
|
32
32
|
### Understanding Skill Structure
|
|
33
33
|
|
|
@@ -35,16 +35,18 @@ User-defined skills have two key sections:
|
|
|
35
35
|
- **Steps**: Describes WHAT to do (shown to user as task actions)
|
|
36
36
|
- **Execution**: Describes HOW to do it (actual shell commands)
|
|
37
37
|
|
|
38
|
-
Each line in Steps corresponds to a line in Execution at the same
|
|
38
|
+
Each line in Steps corresponds to a line in Execution at the same
|
|
39
|
+
position.
|
|
39
40
|
|
|
40
41
|
### How to Generate Commands from Skills
|
|
41
42
|
|
|
42
43
|
1. **Identify skill tasks**: Check if tasks have params.skill
|
|
43
|
-
2. **Find the skill**: Look up the skill in "Available Skills" section
|
|
44
|
+
2. **Find the skill**: Look up the skill in "Available Skills" section
|
|
45
|
+
below
|
|
44
46
|
3. **Match tasks to Execution**: Each task action came from a Steps line;
|
|
45
47
|
use the corresponding Execution line for the command
|
|
46
|
-
4. **Substitute parameters**: Replace {PARAM} placeholders with actual
|
|
47
|
-
from task params
|
|
48
|
+
4. **Substitute parameters**: Replace {PARAM} placeholders with actual
|
|
49
|
+
values from task params
|
|
48
50
|
|
|
49
51
|
### Example Skill
|
|
50
52
|
|
|
@@ -59,8 +61,8 @@ Process Data
|
|
|
59
61
|
|
|
60
62
|
### Execution
|
|
61
63
|
- curl -O https://data.example.com/{SOURCE}.csv
|
|
62
|
-
- python3 transform.py --input {SOURCE}.csv --output
|
|
63
|
-
- csvtool col 1-3
|
|
64
|
+
- python3 transform.py --input {SOURCE}.csv --output data.csv
|
|
65
|
+
- csvtool col 1-3 data.csv > output.{FORMAT}
|
|
64
66
|
```
|
|
65
67
|
|
|
66
68
|
### Matching Process
|
|
@@ -77,9 +79,9 @@ Given tasks from this skill:
|
|
|
77
79
|
Do NOT invent different commands - use exactly what the skill specifies,
|
|
78
80
|
with parameter placeholders replaced by actual values.
|
|
79
81
|
|
|
80
|
-
**CRITICAL**: Take the exact command from the ### Execution section. Do
|
|
81
|
-
modify, improve, or rewrite the command in any way. The user wrote
|
|
82
|
-
commands specifically for their environment and workflow.
|
|
82
|
+
**CRITICAL**: Take the exact command from the ### Execution section. Do
|
|
83
|
+
not modify, improve, or rewrite the command in any way. The user wrote
|
|
84
|
+
these commands specifically for their environment and workflow.
|
|
83
85
|
|
|
84
86
|
## Response Format
|
|
85
87
|
|
|
@@ -90,9 +92,11 @@ Return a structured response with commands to execute:
|
|
|
90
92
|
- **commands**: Array of command objects to execute sequentially
|
|
91
93
|
|
|
92
94
|
**Command object structure:**
|
|
93
|
-
- **description**: Brief description of what this command does (max 64
|
|
95
|
+
- **description**: Brief description of what this command does (max 64
|
|
96
|
+
chars)
|
|
94
97
|
- **command**: The exact shell command to run
|
|
95
|
-
- **workdir**: Optional working directory for the command (defaults to
|
|
98
|
+
- **workdir**: Optional working directory for the command (defaults to
|
|
99
|
+
current)
|
|
96
100
|
- **timeout**: Optional timeout in milliseconds (defaults to 30000)
|
|
97
101
|
- **critical**: Whether failure should stop execution (defaults to true)
|
|
98
102
|
|
|
@@ -101,24 +105,31 @@ Return a structured response with commands to execute:
|
|
|
101
105
|
When generating commands:
|
|
102
106
|
|
|
103
107
|
1. **Be precise**: Generate exact, runnable shell commands
|
|
104
|
-
2. **Be safe**: Never generate destructive commands without explicit user
|
|
105
|
-
|
|
108
|
+
2. **Be safe**: Never generate destructive commands without explicit user
|
|
109
|
+
intent
|
|
110
|
+
3. **Use parameters**: Extract values from task params and incorporate
|
|
111
|
+
them
|
|
106
112
|
4. **Handle paths**: Use proper quoting for paths with spaces
|
|
107
113
|
5. **Be portable**: Prefer POSIX-compatible commands when possible
|
|
108
114
|
|
|
109
115
|
**Safety rules:**
|
|
110
116
|
- NEVER run `rm -rf /` or any command that could delete system files
|
|
111
|
-
- NEVER run commands that modify system configuration without explicit
|
|
117
|
+
- NEVER run commands that modify system configuration without explicit
|
|
118
|
+
request
|
|
112
119
|
- NEVER expose sensitive information in command output
|
|
113
|
-
- Always use safe defaults (e.g., prefer `rm -i` over `rm -f` for
|
|
120
|
+
- Always use safe defaults (e.g., prefer `rm -i` over `rm -f` for
|
|
121
|
+
deletions)
|
|
114
122
|
- For file deletions, prefer moving to trash over permanent deletion
|
|
115
123
|
|
|
116
124
|
## Examples
|
|
117
125
|
|
|
118
126
|
### Example 1: Simple file creation
|
|
119
127
|
|
|
120
|
-
Task: {
|
|
121
|
-
|
|
128
|
+
Task: {
|
|
129
|
+
action: "Create a new file called test.txt",
|
|
130
|
+
type: "execute",
|
|
131
|
+
params: { filename: "test.txt" }
|
|
132
|
+
}
|
|
122
133
|
|
|
123
134
|
Response:
|
|
124
135
|
```
|
|
@@ -130,7 +141,10 @@ commands:
|
|
|
130
141
|
|
|
131
142
|
### Example 2: Directory listing
|
|
132
143
|
|
|
133
|
-
Task: {
|
|
144
|
+
Task: {
|
|
145
|
+
action: "Show files in the current directory",
|
|
146
|
+
type: "execute"
|
|
147
|
+
}
|
|
134
148
|
|
|
135
149
|
Response:
|
|
136
150
|
```
|
|
@@ -143,8 +157,11 @@ commands:
|
|
|
143
157
|
### Example 3: Multiple sequential commands
|
|
144
158
|
|
|
145
159
|
Tasks:
|
|
146
|
-
- {
|
|
147
|
-
|
|
160
|
+
- {
|
|
161
|
+
action: "Create project directory",
|
|
162
|
+
type: "execute",
|
|
163
|
+
params: { path: "my-project" }
|
|
164
|
+
}
|
|
148
165
|
- { action: "Initialize git repository", type: "execute" }
|
|
149
166
|
- { action: "Create README file", type: "execute" }
|
|
150
167
|
|
|
@@ -164,7 +181,10 @@ commands:
|
|
|
164
181
|
|
|
165
182
|
### Example 4: Install dependencies
|
|
166
183
|
|
|
167
|
-
Task: {
|
|
184
|
+
Task: {
|
|
185
|
+
action: "Install dependencies",
|
|
186
|
+
type: "execute"
|
|
187
|
+
}
|
|
168
188
|
|
|
169
189
|
Response:
|
|
170
190
|
```
|
|
@@ -177,20 +197,30 @@ commands:
|
|
|
177
197
|
|
|
178
198
|
### Example 5: Skill-based execution
|
|
179
199
|
|
|
180
|
-
When executing from a skill like "Process Data", tasks include
|
|
200
|
+
When executing from a skill like "Process Data", tasks include
|
|
201
|
+
params.skill:
|
|
181
202
|
|
|
182
203
|
Tasks:
|
|
183
|
-
- {
|
|
184
|
-
|
|
185
|
-
|
|
186
|
-
params: { skill: "Process Data", source: "sales", format: "json" }
|
|
187
|
-
|
|
188
|
-
|
|
204
|
+
- {
|
|
205
|
+
action: "Load the sales dataset",
|
|
206
|
+
type: "execute",
|
|
207
|
+
params: { skill: "Process Data", source: "sales", format: "json" }
|
|
208
|
+
}
|
|
209
|
+
- {
|
|
210
|
+
action: "Transform the sales data",
|
|
211
|
+
type: "execute",
|
|
212
|
+
params: { skill: "Process Data", source: "sales", format: "json" }
|
|
213
|
+
}
|
|
214
|
+
- {
|
|
215
|
+
action: "Export the results to json",
|
|
216
|
+
type: "execute",
|
|
217
|
+
params: { skill: "Process Data", source: "sales", format: "json" }
|
|
218
|
+
}
|
|
189
219
|
|
|
190
220
|
The "Process Data" skill's Execution section specifies:
|
|
191
221
|
- Line 1: curl -O https://data.example.com/{SOURCE}.csv
|
|
192
|
-
- Line 2: python3 transform.py --input {SOURCE}.csv --output
|
|
193
|
-
- Line 3: csvtool col 1-3
|
|
222
|
+
- Line 2: python3 transform.py --input {SOURCE}.csv --output data.csv
|
|
223
|
+
- Line 3: csvtool col 1-3 data.csv > output.{FORMAT}
|
|
194
224
|
|
|
195
225
|
Response (using skill's Execution commands):
|
|
196
226
|
```
|
|
@@ -200,19 +230,23 @@ commands:
|
|
|
200
230
|
command: "curl -O https://data.example.com/sales.csv"
|
|
201
231
|
timeout: 60000
|
|
202
232
|
- description: "Transform the sales data"
|
|
203
|
-
command: "python3 transform.py --input sales.csv --output
|
|
233
|
+
command: "python3 transform.py --input sales.csv --output data.csv"
|
|
204
234
|
timeout: 120000
|
|
205
235
|
- description: "Export the results to json"
|
|
206
|
-
command: "csvtool col 1-3
|
|
236
|
+
command: "csvtool col 1-3 data.csv > output.json"
|
|
207
237
|
```
|
|
208
238
|
|
|
209
|
-
Note: Commands come directly from the skill's Execution section, with
|
|
210
|
-
replaced by "sales" and {FORMAT} replaced by "json" from task
|
|
239
|
+
Note: Commands come directly from the skill's Execution section, with
|
|
240
|
+
{SOURCE} replaced by "sales" and {FORMAT} replaced by "json" from task
|
|
241
|
+
params.
|
|
211
242
|
|
|
212
243
|
### Example 6: File operations with paths
|
|
213
244
|
|
|
214
|
-
Task: {
|
|
215
|
-
|
|
245
|
+
Task: {
|
|
246
|
+
action: "Copy config to backup",
|
|
247
|
+
type: "execute",
|
|
248
|
+
params: { source: "~/.config/app", destination: "~/.config/app.backup" }
|
|
249
|
+
}
|
|
216
250
|
|
|
217
251
|
Response:
|
|
218
252
|
```
|
|
@@ -224,7 +258,10 @@ commands:
|
|
|
224
258
|
|
|
225
259
|
### Example 7: Checking system information
|
|
226
260
|
|
|
227
|
-
Task: {
|
|
261
|
+
Task: {
|
|
262
|
+
action: "Check disk space",
|
|
263
|
+
type: "execute"
|
|
264
|
+
}
|
|
228
265
|
|
|
229
266
|
Response:
|
|
230
267
|
```
|
|
@@ -240,11 +277,12 @@ For complex multi-step operations:
|
|
|
240
277
|
|
|
241
278
|
1. **Sequential dependencies**: Mark early commands as critical so failure
|
|
242
279
|
stops the chain
|
|
243
|
-
2. **Long-running processes**: Set appropriate timeouts (build processes
|
|
244
|
-
need 10+ minutes)
|
|
245
|
-
3. **Working directories**: Use workdir to ensure commands run in the
|
|
246
|
-
location
|
|
247
|
-
4. **Error handling**: For non-critical cleanup steps, set critical:
|
|
280
|
+
2. **Long-running processes**: Set appropriate timeouts (build processes
|
|
281
|
+
may need 10+ minutes)
|
|
282
|
+
3. **Working directories**: Use workdir to ensure commands run in the
|
|
283
|
+
right location
|
|
284
|
+
4. **Error handling**: For non-critical cleanup steps, set critical:
|
|
285
|
+
false
|
|
248
286
|
|
|
249
287
|
## Common Mistakes to Avoid
|
|
250
288
|
|