claude-db 0.1.0 → 0.2.0
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/README.md +95 -243
- package/dist/capture/flush.d.ts +22 -0
- package/dist/capture/flush.js +45 -10
- package/dist/capture/identity.d.ts +6 -0
- package/dist/capture/identity.js +14 -0
- package/dist/capture/index.d.ts +5 -3
- package/dist/capture/index.js +4 -3
- package/dist/capture/manual.d.ts +21 -0
- package/dist/capture/manual.js +40 -0
- package/dist/capture/transcript.d.ts +12 -0
- package/dist/capture/transcript.js +69 -1
- package/dist/capture/turn-extractor.js +40 -8
- package/dist/cli/index.js +376 -30
- package/dist/cli/install.d.ts +15 -1
- package/dist/cli/install.js +127 -8
- package/dist/config/schema.d.ts +1 -0
- package/dist/config/schema.js +9 -1
- package/dist/context.js +24 -2
- package/dist/embed/builtin.js +3 -1
- package/dist/hooks/relevance.js +11 -4
- package/dist/hooks/session-end.js +2 -1
- package/dist/hooks/session-start.js +22 -3
- package/dist/mcp/server.js +56 -6
- package/dist/search/service.js +4 -1
- package/dist/store/adapter.d.ts +22 -6
- package/dist/store/adapter.js +10 -1
- package/dist/store/mongo/driver.d.ts +1 -0
- package/dist/store/mongo/index.d.ts +3 -2
- package/dist/store/mongo/index.js +57 -7
- package/dist/store/postgres/index.d.ts +22 -8
- package/dist/store/postgres/index.js +165 -21
- package/dist/store/sqlite/index.d.ts +3 -2
- package/dist/store/sqlite/index.js +112 -9
- package/dist/store/sqlite/schema.sql +4 -3
- package/dist/types.d.ts +28 -0
- package/dist/util/project.d.ts +9 -10
- package/dist/util/project.js +52 -13
- package/dist/util/vector.d.ts +11 -1
- package/dist/util/vector.js +14 -3
- package/package.json +2 -2
|
@@ -1,6 +1,7 @@
|
|
|
1
|
-
import { closeSync, fstatSync, openSync, readSync } from 'node:fs';
|
|
1
|
+
import { closeSync, fstatSync, openSync, readdirSync, readSync } from 'node:fs';
|
|
2
2
|
import { homedir } from 'node:os';
|
|
3
3
|
import { join } from 'node:path';
|
|
4
|
+
import { resolveProject } from '../util/project.js';
|
|
4
5
|
/**
|
|
5
6
|
* Claude Code already writes a complete record of every session to disk:
|
|
6
7
|
* prompts, replies, and tool calls, appended as JSONL.
|
|
@@ -158,3 +159,70 @@ export function transcriptPathFor(project, sessionId) {
|
|
|
158
159
|
const slug = project.replace(/[/.]/g, '-');
|
|
159
160
|
return join(homedir(), '.claude', 'projects', slug, `${sessionId}.jsonl`);
|
|
160
161
|
}
|
|
162
|
+
/**
|
|
163
|
+
* Every transcript belonging to a project, including sessions started from one
|
|
164
|
+
* of its subdirectories.
|
|
165
|
+
*
|
|
166
|
+
* Claude Code files transcripts under the directory the agent was launched in,
|
|
167
|
+
* while memory is keyed on the repository root, and those differ the moment
|
|
168
|
+
* anyone runs the agent from `repo/frontend`. Candidate directories are found
|
|
169
|
+
* by prefix and then confirmed by the `cwd` each transcript records: the slug
|
|
170
|
+
* maps both `/` and `.` to `-`, so it cannot tell `~/app` from `~/app-extra`,
|
|
171
|
+
* and without the second step a flush would ingest a neighbour's history.
|
|
172
|
+
*/
|
|
173
|
+
export function transcriptsFor(project) {
|
|
174
|
+
const root = join(homedir(), '.claude', 'projects');
|
|
175
|
+
const slug = project.replace(/[/.]/g, '-');
|
|
176
|
+
let candidates;
|
|
177
|
+
try {
|
|
178
|
+
candidates = readdirSync(root).filter((name) => name === slug || name.startsWith(`${slug}-`));
|
|
179
|
+
}
|
|
180
|
+
catch {
|
|
181
|
+
return [];
|
|
182
|
+
}
|
|
183
|
+
const transcripts = [];
|
|
184
|
+
for (const dir of candidates) {
|
|
185
|
+
let files;
|
|
186
|
+
try {
|
|
187
|
+
files = readdirSync(join(root, dir)).filter((name) => name.endsWith('.jsonl'));
|
|
188
|
+
}
|
|
189
|
+
catch {
|
|
190
|
+
continue;
|
|
191
|
+
}
|
|
192
|
+
for (const file of files) {
|
|
193
|
+
const path = join(root, dir, file);
|
|
194
|
+
const cwd = transcriptCwd(path);
|
|
195
|
+
if (cwd === null || resolveProject(cwd) === project)
|
|
196
|
+
transcripts.push(path);
|
|
197
|
+
}
|
|
198
|
+
}
|
|
199
|
+
return transcripts.sort();
|
|
200
|
+
}
|
|
201
|
+
/** First cwd recorded in a transcript, without reading a 90MB file. */
|
|
202
|
+
function transcriptCwd(path) {
|
|
203
|
+
let fd;
|
|
204
|
+
try {
|
|
205
|
+
fd = openSync(path, 'r');
|
|
206
|
+
}
|
|
207
|
+
catch {
|
|
208
|
+
return null;
|
|
209
|
+
}
|
|
210
|
+
try {
|
|
211
|
+
const buffer = Buffer.allocUnsafe(65_536);
|
|
212
|
+
const read = readSync(fd, buffer, 0, buffer.length, 0);
|
|
213
|
+
for (const line of buffer.subarray(0, read).toString('utf8').split('\n')) {
|
|
214
|
+
try {
|
|
215
|
+
const cwd = JSON.parse(line).cwd;
|
|
216
|
+
if (typeof cwd === 'string' && cwd.length > 0)
|
|
217
|
+
return cwd;
|
|
218
|
+
}
|
|
219
|
+
catch {
|
|
220
|
+
// Entry without a cwd, or the torn last line of the window.
|
|
221
|
+
}
|
|
222
|
+
}
|
|
223
|
+
return null;
|
|
224
|
+
}
|
|
225
|
+
finally {
|
|
226
|
+
closeSync(fd);
|
|
227
|
+
}
|
|
228
|
+
}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import { classifyCommand } from './command.js';
|
|
2
|
-
import { observationId } from './identity.js';
|
|
2
|
+
import { currentAuthor, observationId } from './identity.js';
|
|
3
3
|
/**
|
|
4
4
|
* Builds observations from transcript turns.
|
|
5
5
|
*
|
|
@@ -40,15 +40,18 @@ function buildObservation(turn, sessionId, project, config) {
|
|
|
40
40
|
]
|
|
41
41
|
.filter((line) => line !== '')
|
|
42
42
|
.join('\n'), config.capture.maxBodyChars));
|
|
43
|
+
const author = currentAuthor();
|
|
43
44
|
return {
|
|
44
45
|
id: observationId(sessionId, turn.timestamp, turn.prompt),
|
|
45
46
|
sessionId,
|
|
46
47
|
project,
|
|
47
|
-
|
|
48
|
-
|
|
48
|
+
...(author ? { author } : {}),
|
|
49
|
+
kind: classifyTurn(turn),
|
|
50
|
+
// Redacted like the body: the title is injected above every future prompt.
|
|
51
|
+
title: redact(buildTitle(turn, files)),
|
|
49
52
|
body,
|
|
50
53
|
files,
|
|
51
|
-
tags:
|
|
54
|
+
tags: topLevelDirs(files, project),
|
|
52
55
|
createdAt: turn.timestamp,
|
|
53
56
|
};
|
|
54
57
|
}
|
|
@@ -88,8 +91,34 @@ function headline(sentence) {
|
|
|
88
91
|
const word = clause.lastIndexOf(' ');
|
|
89
92
|
return `${clause.slice(0, word > 30 ? word : TITLE_MAX)}...`;
|
|
90
93
|
}
|
|
91
|
-
|
|
94
|
+
/**
|
|
95
|
+
* The first path segment under the project root, which is the repository name
|
|
96
|
+
* when the project directory holds several repos, and the top-level source
|
|
97
|
+
* folder when it is one repo. Tags carry real weight in every backend's index,
|
|
98
|
+
* so this is what makes a pooled memory filterable by where the work happened.
|
|
99
|
+
*/
|
|
100
|
+
function topLevelDirs(files, project) {
|
|
101
|
+
const prefix = project.endsWith('/') ? project : `${project}/`;
|
|
102
|
+
const tags = new Set();
|
|
103
|
+
for (const file of files) {
|
|
104
|
+
if (!file.startsWith(prefix))
|
|
105
|
+
continue;
|
|
106
|
+
const parts = file.slice(prefix.length).split('/');
|
|
107
|
+
// Two or more parts means the first is a directory, not a root-level file.
|
|
108
|
+
if (parts.length > 1 && parts[0])
|
|
109
|
+
tags.add(parts[0]);
|
|
110
|
+
}
|
|
111
|
+
return [...tags].slice(0, 3);
|
|
112
|
+
}
|
|
113
|
+
function classifyTurn(turn) {
|
|
92
114
|
const text = `${turn.prompt} ${turn.reasoning}`.toLowerCase();
|
|
115
|
+
// Read from the prompt alone, and only where the phrasing is a standing
|
|
116
|
+
// instruction. "the build always fails" is a bugfix, not a house rule, so
|
|
117
|
+
// always/never have to be followed by something imperative to count.
|
|
118
|
+
if (/\b(from now on|going forward|remember to|prefer)\b/.test(turn.prompt.toLowerCase()) ||
|
|
119
|
+
/\b(always|never)\s+(use|run|do|add|write|call|put|name|commit|push|import|install)\b/.test(turn.prompt.toLowerCase())) {
|
|
120
|
+
return 'preference';
|
|
121
|
+
}
|
|
93
122
|
if (/\b(instead of|rather than|chose|decided|because|why we|trade-?off)\b/.test(text)) {
|
|
94
123
|
return 'decision';
|
|
95
124
|
}
|
|
@@ -98,9 +127,7 @@ function classifyTurn(turn, hasCommands) {
|
|
|
98
127
|
}
|
|
99
128
|
if (/\b(fix(ed)?|bug|broken|regression|error|crash)\b/.test(text))
|
|
100
129
|
return 'bugfix';
|
|
101
|
-
|
|
102
|
-
return 'pattern';
|
|
103
|
-
return hasCommands ? 'context' : 'context';
|
|
130
|
+
return turn.files.length > 0 ? 'pattern' : 'context';
|
|
104
131
|
}
|
|
105
132
|
function firstSentence(text) {
|
|
106
133
|
const trimmed = text.trim();
|
|
@@ -123,11 +150,16 @@ function isExcluded(file, patterns) {
|
|
|
123
150
|
return patterns.some((pattern) => file.includes(pattern));
|
|
124
151
|
}
|
|
125
152
|
const PRIVATE_BLOCK = /<private>[\s\S]*?<\/private>/gi;
|
|
153
|
+
const PRIVATE_KEY = /-----BEGIN [A-Z ]*PRIVATE KEY-----[\s\S]*?-----END [A-Z ]*PRIVATE KEY-----/g;
|
|
126
154
|
function redact(text) {
|
|
127
155
|
return text
|
|
128
156
|
.replace(PRIVATE_BLOCK, '[redacted]')
|
|
157
|
+
.replace(PRIVATE_KEY, '[redacted-private-key]')
|
|
129
158
|
.replace(/\b(sk-[A-Za-z0-9_-]{16,})\b/g, '[redacted-key]')
|
|
130
159
|
.replace(/\b(gh[pousr]_[A-Za-z0-9]{16,})\b/g, '[redacted-token]')
|
|
160
|
+
.replace(/\bAKIA[0-9A-Z]{16}\b/g, '[redacted-key]')
|
|
161
|
+
.replace(/\bxox[baprs]-[A-Za-z0-9-]{10,}/g, '[redacted-token]')
|
|
162
|
+
.replace(/\beyJ[A-Za-z0-9_-]{8,}\.[A-Za-z0-9_-]{8,}\.[A-Za-z0-9_-]{8,}/g, '[redacted-jwt]')
|
|
131
163
|
.replace(/(?<=(password|secret|token|api[_-]?key)"?\s*[:=]\s*)"[^"]+"/gi, '"[redacted]"');
|
|
132
164
|
}
|
|
133
165
|
function truncate(text, max) {
|