this repo has no description
fix(capture): extract tasks from stream-of-consciousness text
The parser now handles unstructured brain dumps like:
"i have to send an email to hr. i'm finally sending..."
Extracts tasks from:
- Intent phrases: "need to X", "have to X", "gotta X", "should X"
- "Going to X", "gonna X"
- "Finally X" (procrastination indicator!)
Plus existing support for:
- Bullet points and numbered lists
- Action verbs at line starts
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
+53
-15
+53
-15
src/tools/capture.ts
+53
-15
src/tools/capture.ts
···
36
36
/**
37
37
* Parse brain dump text to extract tasks and ideas
38
38
*
39
-
* Uses simple heuristics for M2:
40
-
* - Lines starting with "- ", "* ", "TODO", numbers, or action verbs are tasks
41
-
* - Other lines are ideas
42
-
* - Strips common prefixes and cleans up text
39
+
* Handles both structured (bullet points) and unstructured (stream-of-consciousness) text.
40
+
* Extracts tasks based on:
41
+
* - Explicit list markers (-, *, numbers, TODO)
42
+
* - Intent phrases ("need to", "have to", "gotta", "should")
43
+
* - Action verbs at sentence starts
43
44
*/
44
45
function parseBrainDump(text: string): { tasks: string[]; ideas: string[] } {
45
-
const lines = text
46
-
.split('\n')
47
-
.map((line) => line.trim())
48
-
.filter((line) => line.length > 0);
49
-
50
46
const tasks: string[] = [];
51
47
const ideas: string[] = [];
52
48
53
-
// Common task prefixes to strip
49
+
// Common task prefixes to strip from list items
54
50
const taskPrefixes = [/^-\s+/, /^\*\s+/, /^\d+[).]\s+/, /^TODO:?\s*/i, /^TASK:?\s*/i];
55
51
56
52
// Action verbs that indicate tasks
57
53
const actionVerbs =
58
-
/^(add|create|implement|fix|update|write|build|test|review|refactor|delete|remove|setup|configure|install|deploy|investigate|research|learn|study|call|email|message|buy|order|schedule|plan|organize|clean|finish|complete|start|begin)\b/i;
54
+
/^(add|create|implement|fix|update|write|build|test|review|refactor|delete|remove|setup|configure|install|deploy|investigate|research|learn|study|call|email|message|buy|order|schedule|plan|organize|clean|finish|complete|start|begin|send|check|get|make|do|prepare|submit|apply|respond|reply|contact|reach|follow|set|book|cancel|return|pick|drop)\b/i;
55
+
56
+
// Intent phrases that signal tasks (captures "I need to X", "gotta X", etc.)
57
+
const intentPatterns = [
58
+
/\b(?:i\s+)?(?:need|have|got|gotta|should|must|want)\s+to\s+(\w+(?:\s+\w+){0,10}?)(?:\.|,|$|(?=\s+(?:and|but|i\s|i'm|also)))/gi,
59
+
/\b(?:i'm|i\s+am)\s+(?:gonna|going\s+to)\s+(\w+(?:\s+\w+){0,10}?)(?:\.|,|$|(?=\s+(?:and|but|i\s|i'm|also)))/gi,
60
+
/\bfinally\s+(\w+(?:\s+\w+){0,10}?)(?:\.|,|$|(?=\s+(?:and|but|i\s|i'm|also)))/gi,
61
+
];
62
+
63
+
// First, try to extract tasks from intent phrases in unstructured text
64
+
const extractedTasks = new Set<string>();
65
+
for (const pattern of intentPatterns) {
66
+
let match;
67
+
while ((match = pattern.exec(text)) !== null) {
68
+
const task = match[1]?.trim() ?? '';
69
+
if (task.length > 2 && task.length < 200) {
70
+
// Clean up the task - capitalize first letter
71
+
const cleanTask = task.charAt(0).toUpperCase() + task.slice(1);
72
+
extractedTasks.add(cleanTask);
73
+
}
74
+
}
75
+
}
76
+
77
+
// Add extracted tasks
78
+
for (const task of extractedTasks) {
79
+
tasks.push(task);
80
+
}
81
+
82
+
// Then process line by line for structured input
83
+
const lines = text
84
+
.split('\n')
85
+
.map((line) => line.trim())
86
+
.filter((line) => line.length > 0);
59
87
60
88
for (const line of lines) {
61
89
let content = line;
62
90
let isTask = false;
63
91
64
-
// Check for task prefixes
92
+
// Check for task prefixes (bullet points, numbers)
65
93
for (const prefix of taskPrefixes) {
66
94
if (prefix.test(content)) {
67
95
content = content.replace(prefix, '');
···
70
98
}
71
99
}
72
100
73
-
// If no prefix matched, check for action verbs
101
+
// If no prefix matched, check for action verbs at line start
74
102
if (!isTask && actionVerbs.test(content)) {
75
103
isTask = true;
76
104
}
···
80
108
81
109
if (content.length > 0) {
82
110
if (isTask) {
83
-
tasks.push(content);
84
-
} else {
111
+
// Avoid duplicates from intent extraction
112
+
if (!tasks.some((t) => t.toLowerCase() === content.toLowerCase())) {
113
+
tasks.push(content);
114
+
}
115
+
} else if (tasks.length === 0) {
116
+
// Only add as idea if we didn't extract any tasks from it
117
+
// (to avoid storing the whole brain dump as an idea when we extracted tasks)
85
118
ideas.push(content);
86
119
}
87
120
}
121
+
}
122
+
123
+
// If we extracted tasks but the whole text was one line, don't store it as an idea
124
+
if (tasks.length > 0 && ideas.length === 1 && ideas[0] === text.trim()) {
125
+
ideas.length = 0;
88
126
}
89
127
90
128
return { tasks, ideas };