clementine-agent 1.0.97 → 1.0.98
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/agent/self-improve.d.ts +7 -0
- package/dist/agent/self-improve.js +114 -71
- package/dist/types.d.ts +2 -2
- package/package.json +1 -1
|
@@ -59,8 +59,15 @@ export declare class SelfImproveLoop {
|
|
|
59
59
|
private parseJsonResponse;
|
|
60
60
|
private withTimeout;
|
|
61
61
|
}
|
|
62
|
+
/** Validate that a proposed change has valid syntax for its target area. */
|
|
62
63
|
export declare function validateProposal(area: string, target: string, proposedChange: string): {
|
|
63
64
|
valid: boolean;
|
|
64
65
|
error?: string;
|
|
65
66
|
};
|
|
67
|
+
/**
|
|
68
|
+
* Resolve the on-disk path for a prompt-override `target` string.
|
|
69
|
+
* Accepted forms: 'global', 'agent:<slug>', 'job:<jobName>'. Returns null
|
|
70
|
+
* for malformed targets.
|
|
71
|
+
*/
|
|
72
|
+
export declare function promptOverridePathForTarget(target: string): string | null;
|
|
66
73
|
//# sourceMappingURL=self-improve.d.ts.map
|
|
@@ -11,9 +11,10 @@
|
|
|
11
11
|
import { randomBytes } from 'node:crypto';
|
|
12
12
|
import { appendFileSync, existsSync, mkdirSync, readFileSync, readdirSync, unlinkSync, writeFileSync, } from 'node:fs';
|
|
13
13
|
import matter from 'gray-matter';
|
|
14
|
+
import { load as yamlLoad } from 'js-yaml';
|
|
14
15
|
import path from 'node:path';
|
|
15
16
|
import pino from 'pino';
|
|
16
|
-
import { BASE_DIR, SELF_IMPROVE_DIR, SOUL_FILE, AGENTS_FILE, CRON_FILE, WORKFLOWS_DIR, VAULT_DIR, MEMORY_DB_PATH, AGENTS_DIR,
|
|
17
|
+
import { BASE_DIR, SELF_IMPROVE_DIR, SOUL_FILE, AGENTS_FILE, CRON_FILE, WORKFLOWS_DIR, VAULT_DIR, MEMORY_DB_PATH, AGENTS_DIR, CRON_REFLECTIONS_DIR, GOALS_DIR, } from '../config.js';
|
|
17
18
|
import { listAllGoals } from '../tools/shared.js';
|
|
18
19
|
const logger = pino({ name: 'clementine.self-improve' });
|
|
19
20
|
// ── Defaults ─────────────────────────────────────────────────────────
|
|
@@ -23,9 +24,13 @@ const DEFAULT_CONFIG = {
|
|
|
23
24
|
maxDurationMs: 3_600_000, // 1 hour
|
|
24
25
|
acceptThreshold: 0.7,
|
|
25
26
|
plateauLimit: 3,
|
|
26
|
-
// 'source' deprecated — self-improvement
|
|
27
|
-
//
|
|
28
|
-
|
|
27
|
+
// 'source' deprecated — self-improvement produces data, not engine TS edits.
|
|
28
|
+
// 'advisor-rule' writes YAML to ~/.clementine/advisor-rules/user/.
|
|
29
|
+
// 'prompt-override' writes markdown to ~/.clementine/prompt-overrides/.
|
|
30
|
+
areas: [
|
|
31
|
+
'soul', 'cron', 'workflow', 'memory', 'agent', 'communication', 'goal',
|
|
32
|
+
'advisor-rule', 'prompt-override',
|
|
33
|
+
],
|
|
29
34
|
autoApply: true,
|
|
30
35
|
sourceMode: 'skip',
|
|
31
36
|
};
|
|
@@ -132,20 +137,22 @@ function checkDrift(proposedContent) {
|
|
|
132
137
|
return { ok: similarity >= DRIFT_SIMILARITY_THRESHOLD, similarity };
|
|
133
138
|
}
|
|
134
139
|
/** Classify the risk level of a proposed change.
|
|
135
|
-
* - low: agent prompts, individual cron job prompts
|
|
140
|
+
* - low: agent prompts, individual cron job prompts, advisor rules, prompt overrides
|
|
136
141
|
* - medium: SOUL.md, AGENTS.md, MEMORY.md — needs owner approval
|
|
137
|
-
* - high: source code — stays blocked
|
|
142
|
+
* - high: source code — stays blocked (deprecated path; kept for back-compat)
|
|
138
143
|
*/
|
|
139
144
|
function classifyRisk(area) {
|
|
140
145
|
switch (area) {
|
|
141
|
-
case 'agent': return 'low';
|
|
142
|
-
case 'cron': return 'low';
|
|
143
|
-
case 'workflow': return 'low';
|
|
146
|
+
case 'agent': return 'low';
|
|
147
|
+
case 'cron': return 'low';
|
|
148
|
+
case 'workflow': return 'low';
|
|
149
|
+
case 'advisor-rule': return 'low'; // YAML files, hot-reloaded, easily deleted
|
|
150
|
+
case 'prompt-override': return 'low'; // Markdown files, hot-reloaded, easily deleted
|
|
144
151
|
case 'soul': return 'medium'; // Core personality — needs approval
|
|
145
152
|
case 'communication': return 'medium'; // Global operating instructions
|
|
146
153
|
case 'memory': return 'medium'; // Memory config
|
|
147
|
-
case 'source': return 'high'; // Code changes — always blocked in auto mode
|
|
148
154
|
case 'goal': return 'medium'; // New goals need owner review before activating
|
|
155
|
+
case 'source': return 'high'; // Deprecated — quarantined in Phase 1
|
|
149
156
|
default: return 'high';
|
|
150
157
|
}
|
|
151
158
|
}
|
|
@@ -836,7 +843,15 @@ export class SelfImproveLoop {
|
|
|
836
843
|
`Area notes:\n` +
|
|
837
844
|
`- For "goal": target = "{owner}/{goal-slug}" (e.g. "clementine/improve-reply-rates" or "ross-the-sdr/book-demos"). ` +
|
|
838
845
|
`Propose when you observe a pattern in completed tasks or cron runs that suggests a missing or stale goal. ` +
|
|
839
|
-
`The proposedChange must be a JSON goal object with at minimum: title, description, priority, reviewFrequency.\n
|
|
846
|
+
`The proposedChange must be a JSON goal object with at minimum: title, description, priority, reviewFrequency.\n` +
|
|
847
|
+
`- For "advisor-rule": target = ruleId in kebab-case (e.g. "skip-turn-bump-on-unleashed"). ` +
|
|
848
|
+
`Use when the fix is a behavioral rule that affects ALL jobs matching some scope, not just one cron job. ` +
|
|
849
|
+
`Examples: "for unleashed jobs, never bump maxTurns" or "for ross-the-sdr, double timeout on max_turns". ` +
|
|
850
|
+
`The proposedChange must be a full advisor rule YAML body with: schemaVersion: 1, id (must match target), description, priority (use 100+ to override builtins), appliesTo, when[], then[]. ` +
|
|
851
|
+
`User rules at priority 100+ override engine builtins of the same id.\n` +
|
|
852
|
+
`- For "prompt-override": target = "global", "agent:<slug>", or "job:<jobName>" (e.g. "job:market-leader-followup"). ` +
|
|
853
|
+
`Use when a job/agent needs more standing guidance — markdown that gets prepended to its prompt. ` +
|
|
854
|
+
`The proposedChange is the markdown body (optionally with gray-matter frontmatter for priority/position).\n\n` +
|
|
840
855
|
`Return your answer as a JSON object matching the schema: { "results": [ ... ] }. Up to 3 items. If absolutely nothing actionable today, return { "results": [] }.`;
|
|
841
856
|
const analysisResult = await this.assistant.runPlanStep('si-analyze', analysisPrompt, {
|
|
842
857
|
tier: 2,
|
|
@@ -915,10 +930,6 @@ export class SelfImproveLoop {
|
|
|
915
930
|
const agentFile = path.join(AGENTS_DIR, target, 'agent.md');
|
|
916
931
|
return existsSync(agentFile) ? readFileSync(agentFile, 'utf-8') : '';
|
|
917
932
|
}
|
|
918
|
-
case 'source': {
|
|
919
|
-
const srcFile = path.join(PKG_DIR, 'src', target);
|
|
920
|
-
return existsSync(srcFile) ? readFileSync(srcFile, 'utf-8') : '';
|
|
921
|
-
}
|
|
922
933
|
case 'communication':
|
|
923
934
|
return existsSync(AGENTS_FILE) ? readFileSync(AGENTS_FILE, 'utf-8') : '';
|
|
924
935
|
case 'memory': {
|
|
@@ -946,6 +957,23 @@ export class SelfImproveLoop {
|
|
|
946
957
|
return '(no goals yet for this owner)';
|
|
947
958
|
return goals.map((g) => `[${g.status ?? 'unknown'}] ${g.title}: ${(g.description ?? '').slice(0, 120)}`).join('\n');
|
|
948
959
|
}
|
|
960
|
+
case 'advisor-rule': {
|
|
961
|
+
// target = ruleId (kebab-case). Show user override file if present, else builtin.
|
|
962
|
+
const userPath = path.join(BASE_DIR, 'advisor-rules', 'user', `${target}.yaml`);
|
|
963
|
+
if (existsSync(userPath))
|
|
964
|
+
return readFileSync(userPath, 'utf-8');
|
|
965
|
+
const builtinPath = path.join(BASE_DIR, 'advisor-rules', 'builtin', `${target}.yaml`);
|
|
966
|
+
if (existsSync(builtinPath))
|
|
967
|
+
return readFileSync(builtinPath, 'utf-8');
|
|
968
|
+
return '(no existing advisor rule for this id — proposing a new one)';
|
|
969
|
+
}
|
|
970
|
+
case 'prompt-override': {
|
|
971
|
+
// target = 'global' | 'agent:<slug>' | 'job:<jobName>'
|
|
972
|
+
const filePath = promptOverridePathForTarget(target);
|
|
973
|
+
if (filePath && existsSync(filePath))
|
|
974
|
+
return readFileSync(filePath, 'utf-8');
|
|
975
|
+
return '(no existing prompt override for this scope — proposing a new one)';
|
|
976
|
+
}
|
|
949
977
|
default:
|
|
950
978
|
return '';
|
|
951
979
|
}
|
|
@@ -1011,38 +1039,10 @@ export class SelfImproveLoop {
|
|
|
1011
1039
|
if (!targetPath) {
|
|
1012
1040
|
return `Cannot resolve target path for area=${pending.area}, target=${pending.target}`;
|
|
1013
1041
|
}
|
|
1014
|
-
//
|
|
1042
|
+
// 'source' area is deprecated (Phase 1 quarantine). Reject up-front so a
|
|
1043
|
+
// misbehaving proposal cannot reach the safeSourceEdit primitive.
|
|
1015
1044
|
if (pending.area === 'source') {
|
|
1016
|
-
|
|
1017
|
-
const result = await safeSourceEdit(PKG_DIR, [
|
|
1018
|
-
{ relativePath: `src/${pending.target}`, content: pending.proposedChange },
|
|
1019
|
-
], { experimentId, reason: `self-improve: ${pending.hypothesis.slice(0, 60)}`, description: pending.hypothesis });
|
|
1020
|
-
if (!result.success) {
|
|
1021
|
-
return `Source edit failed: ${result.error}${result.preflightErrors ? '\n' + result.preflightErrors.join('\n') : ''}`;
|
|
1022
|
-
}
|
|
1023
|
-
// Update experiment log — mark as approved
|
|
1024
|
-
this.updateExperimentStatus(experimentId, 'approved');
|
|
1025
|
-
try {
|
|
1026
|
-
unlinkSync(pendingFile);
|
|
1027
|
-
}
|
|
1028
|
-
catch { /* ignore */ }
|
|
1029
|
-
const state = this.loadState();
|
|
1030
|
-
state.pendingApprovals = Math.max(0, state.pendingApprovals - 1);
|
|
1031
|
-
this.saveState(state);
|
|
1032
|
-
// Schedule impact measurement for 24h later
|
|
1033
|
-
try {
|
|
1034
|
-
appendFileSync(IMPACT_CHECKS_FILE, JSON.stringify({
|
|
1035
|
-
experimentId,
|
|
1036
|
-
area: pending.area,
|
|
1037
|
-
target: pending.target,
|
|
1038
|
-
appliedAt: new Date().toISOString(),
|
|
1039
|
-
checkAfterMs: 24 * 60 * 60 * 1000,
|
|
1040
|
-
}) + '\n');
|
|
1041
|
-
}
|
|
1042
|
-
catch (err) {
|
|
1043
|
-
logger.warn({ err }, 'Failed to schedule impact check');
|
|
1044
|
-
}
|
|
1045
|
-
return `Applied source change to ${pending.target} — restart triggered.`;
|
|
1045
|
+
return 'source area is deprecated — propose advisor-rule or prompt-override instead';
|
|
1046
1046
|
}
|
|
1047
1047
|
// Goal area: parse JSON, inject required fields, ensure parent dir exists
|
|
1048
1048
|
if (pending.area === 'goal') {
|
|
@@ -1095,6 +1095,7 @@ export class SelfImproveLoop {
|
|
|
1095
1095
|
}
|
|
1096
1096
|
}
|
|
1097
1097
|
// Write the change (non-source areas)
|
|
1098
|
+
mkdirSync(path.dirname(targetPath), { recursive: true });
|
|
1098
1099
|
writeFileSync(targetPath, pending.proposedChange);
|
|
1099
1100
|
// Record version for rollback lineage
|
|
1100
1101
|
this.recordVersion(experimentId, pending.area, pending.target, pending.hypothesis, pending.before);
|
|
@@ -1680,9 +1681,6 @@ export class SelfImproveLoop {
|
|
|
1680
1681
|
case 'agent': {
|
|
1681
1682
|
return path.join(AGENTS_DIR, target, 'agent.md');
|
|
1682
1683
|
}
|
|
1683
|
-
case 'source': {
|
|
1684
|
-
return path.join(PKG_DIR, 'src', target);
|
|
1685
|
-
}
|
|
1686
1684
|
case 'communication':
|
|
1687
1685
|
return AGENTS_FILE;
|
|
1688
1686
|
case 'memory':
|
|
@@ -1696,6 +1694,12 @@ export class SelfImproveLoop {
|
|
|
1696
1694
|
return path.join(GOALS_DIR, `${goalSlug}.json`);
|
|
1697
1695
|
return path.join(AGENTS_DIR, owner, 'goals', `${goalSlug}.json`);
|
|
1698
1696
|
}
|
|
1697
|
+
case 'advisor-rule':
|
|
1698
|
+
if (!/^[a-z0-9-]+$/.test(target))
|
|
1699
|
+
return null;
|
|
1700
|
+
return path.join(BASE_DIR, 'advisor-rules', 'user', `${target}.yaml`);
|
|
1701
|
+
case 'prompt-override':
|
|
1702
|
+
return promptOverridePathForTarget(target);
|
|
1699
1703
|
default:
|
|
1700
1704
|
return null;
|
|
1701
1705
|
}
|
|
@@ -1733,22 +1737,6 @@ export class SelfImproveLoop {
|
|
|
1733
1737
|
}
|
|
1734
1738
|
// ── Utility ──────────────────────────────────────────────────────────
|
|
1735
1739
|
/** Validate that a proposed change has valid syntax for its target area. */
|
|
1736
|
-
/** Files that must never be modified by self-improvement (catastrophic blast radius or self-referential). */
|
|
1737
|
-
const SOURCE_BLOCKLIST = new Set([
|
|
1738
|
-
'config.ts',
|
|
1739
|
-
'types.ts',
|
|
1740
|
-
'gateway/router.ts',
|
|
1741
|
-
'gateway/lanes.ts',
|
|
1742
|
-
'gateway/heartbeat-scheduler.ts',
|
|
1743
|
-
'gateway/cron-scheduler.ts',
|
|
1744
|
-
'gateway/security-scanner.ts',
|
|
1745
|
-
'agent/self-improve.ts',
|
|
1746
|
-
'agent/safe-restart.ts',
|
|
1747
|
-
'agent/source-mods.ts',
|
|
1748
|
-
'cli/index.ts',
|
|
1749
|
-
'cli/dashboard.ts',
|
|
1750
|
-
'security/scanner.ts',
|
|
1751
|
-
]);
|
|
1752
1740
|
export function validateProposal(area, target, proposedChange) {
|
|
1753
1741
|
if (!proposedChange.trim()) {
|
|
1754
1742
|
return { valid: false, error: 'Proposed change is empty' };
|
|
@@ -1790,16 +1778,71 @@ export function validateProposal(area, target, proposedChange) {
|
|
|
1790
1778
|
}
|
|
1791
1779
|
}
|
|
1792
1780
|
if (area === 'source') {
|
|
1793
|
-
//
|
|
1794
|
-
|
|
1795
|
-
|
|
1781
|
+
// Deprecated — Phase 1 quarantined source self-edit. Reject up front so
|
|
1782
|
+
// a misbehaving LLM proposal doesn't even get cached.
|
|
1783
|
+
return { valid: false, error: 'source area is deprecated; propose advisor-rule or prompt-override instead' };
|
|
1784
|
+
}
|
|
1785
|
+
if (area === 'advisor-rule') {
|
|
1786
|
+
// Must parse as YAML and have schemaVersion: 1, id matching target, when[], then[].
|
|
1787
|
+
if (!/^[a-z0-9-]+$/.test(target)) {
|
|
1788
|
+
return { valid: false, error: `advisor-rule target must be kebab-case (got "${target}")` };
|
|
1789
|
+
}
|
|
1790
|
+
let parsed;
|
|
1791
|
+
try {
|
|
1792
|
+
parsed = yamlLoad(proposedChange);
|
|
1793
|
+
}
|
|
1794
|
+
catch (err) {
|
|
1795
|
+
return { valid: false, error: `advisor-rule YAML parse error: ${err instanceof Error ? err.message : String(err)}` };
|
|
1796
|
+
}
|
|
1797
|
+
if (!parsed || typeof parsed !== 'object') {
|
|
1798
|
+
return { valid: false, error: 'advisor-rule body did not parse as a YAML object' };
|
|
1799
|
+
}
|
|
1800
|
+
const r = parsed;
|
|
1801
|
+
if (r.schemaVersion !== 1)
|
|
1802
|
+
return { valid: false, error: 'advisor-rule must declare schemaVersion: 1' };
|
|
1803
|
+
if (typeof r.id !== 'string' || r.id !== target) {
|
|
1804
|
+
return { valid: false, error: `advisor-rule id must match target ("${target}")` };
|
|
1805
|
+
}
|
|
1806
|
+
if (!Array.isArray(r.when) || !Array.isArray(r.then)) {
|
|
1807
|
+
return { valid: false, error: 'advisor-rule must have when[] and then[] arrays' };
|
|
1808
|
+
}
|
|
1809
|
+
}
|
|
1810
|
+
if (area === 'prompt-override') {
|
|
1811
|
+
// target format: 'global' | 'agent:<slug>' | 'job:<jobName>'
|
|
1812
|
+
const path = promptOverridePathForTarget(target);
|
|
1813
|
+
if (!path) {
|
|
1814
|
+
return { valid: false, error: `prompt-override target must be 'global', 'agent:<slug>', or 'job:<jobName>' (got "${target}")` };
|
|
1815
|
+
}
|
|
1816
|
+
if (proposedChange.length > 20_000) {
|
|
1817
|
+
return { valid: false, error: 'prompt-override content exceeds 20KB sanity bound' };
|
|
1796
1818
|
}
|
|
1797
|
-
// Size sanity: reject wholesale rewrites (proposed content > 2x original would be caught by caller).
|
|
1798
|
-
// Source proposals may be small patches or modules without import/export statements;
|
|
1799
|
-
// callers that apply source changes do the syntax-aware validation.
|
|
1800
1819
|
}
|
|
1801
1820
|
return { valid: true };
|
|
1802
1821
|
}
|
|
1822
|
+
/**
|
|
1823
|
+
* Resolve the on-disk path for a prompt-override `target` string.
|
|
1824
|
+
* Accepted forms: 'global', 'agent:<slug>', 'job:<jobName>'. Returns null
|
|
1825
|
+
* for malformed targets.
|
|
1826
|
+
*/
|
|
1827
|
+
export function promptOverridePathForTarget(target) {
|
|
1828
|
+
const root = path.join(BASE_DIR, 'prompt-overrides');
|
|
1829
|
+
if (target === 'global')
|
|
1830
|
+
return path.join(root, '_global.md');
|
|
1831
|
+
const idx = target.indexOf(':');
|
|
1832
|
+
if (idx <= 0)
|
|
1833
|
+
return null;
|
|
1834
|
+
const scope = target.slice(0, idx);
|
|
1835
|
+
const key = target.slice(idx + 1);
|
|
1836
|
+
if (!key)
|
|
1837
|
+
return null;
|
|
1838
|
+
if (/[\/\\\.]/.test(key))
|
|
1839
|
+
return null;
|
|
1840
|
+
if (scope === 'agent')
|
|
1841
|
+
return path.join(root, 'agents', `${key}.md`);
|
|
1842
|
+
if (scope === 'job')
|
|
1843
|
+
return path.join(root, 'jobs', `${key}.md`);
|
|
1844
|
+
return null;
|
|
1845
|
+
}
|
|
1803
1846
|
function ensureDirs() {
|
|
1804
1847
|
for (const dir of [SELF_IMPROVE_DIR, PENDING_DIR]) {
|
|
1805
1848
|
if (!existsSync(dir)) {
|
package/dist/types.d.ts
CHANGED
|
@@ -501,7 +501,7 @@ export interface SelfImproveExperiment {
|
|
|
501
501
|
startedAt: string;
|
|
502
502
|
finishedAt: string;
|
|
503
503
|
durationMs: number;
|
|
504
|
-
area: 'soul' | 'cron' | 'workflow' | 'memory' | 'agent' | 'source' | 'communication' | 'goal';
|
|
504
|
+
area: 'soul' | 'cron' | 'workflow' | 'memory' | 'agent' | 'source' | 'communication' | 'goal' | 'advisor-rule' | 'prompt-override';
|
|
505
505
|
target: string;
|
|
506
506
|
hypothesis: string;
|
|
507
507
|
proposedChange: string;
|
|
@@ -549,7 +549,7 @@ export interface SelfImproveConfig {
|
|
|
549
549
|
maxDurationMs: number;
|
|
550
550
|
acceptThreshold: number;
|
|
551
551
|
plateauLimit: number;
|
|
552
|
-
areas: ('soul' | 'cron' | 'workflow' | 'memory' | 'agent' | 'source' | 'communication' | 'goal')[];
|
|
552
|
+
areas: ('soul' | 'cron' | 'workflow' | 'memory' | 'agent' | 'source' | 'communication' | 'goal' | 'advisor-rule' | 'prompt-override')[];
|
|
553
553
|
/** Enable tiered auto-apply: low-risk changes apply without approval. Default: false. */
|
|
554
554
|
autoApply?: boolean;
|
|
555
555
|
/** Target a specific agent slug (for per-agent improvement cycles). */
|