@claudetools/tools 0.7.5 → 0.7.6
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.
|
@@ -191,6 +191,23 @@ export function buildOrchestratorPrompt(params) {
|
|
|
191
191
|
|
|
192
192
|
<!-- Layer 4: Domain Knowledge -->
|
|
193
193
|
<domain_knowledge>
|
|
194
|
+
<memory_best_practices priority="CRITICAL">
|
|
195
|
+
When storing facts with memory_store_fact(), follow these rules:
|
|
196
|
+
|
|
197
|
+
✅ STORE (learned contextual knowledge):
|
|
198
|
+
- Project discoveries: "portal-v2 HAD_ANTI_PATTERN TanStack Query unused despite being in dependencies"
|
|
199
|
+
- User preferences: "Owen PREFERS_WORKFLOW Comprehensive analysis → Review → Approve with 'go' → Parallel dispatch"
|
|
200
|
+
- Architectural decisions WHY: "auth-service REJECTED JWT because stateless tokens don't support session invalidation"
|
|
201
|
+
- Solutions that worked: "Hardcoded colors SOLVED_BY Creating src/lib/colors.ts with design tokens"
|
|
202
|
+
|
|
203
|
+
❌ DON'T STORE (generic documentation):
|
|
204
|
+
- Tool behavior: "task_plan creates tasks with ready status" (already in API docs)
|
|
205
|
+
- Universal knowledge: "React uses JSX" (not learned, universal)
|
|
206
|
+
- Temporary state: "Current task is in_progress" (session-specific)
|
|
207
|
+
|
|
208
|
+
Rule of thumb: "Would I need to rediscover this in 3 months?" → YES = store, NO = don't store
|
|
209
|
+
</memory_best_practices>
|
|
210
|
+
|
|
194
211
|
<available_workers>
|
|
195
212
|
${availableWorkers.length > 0
|
|
196
213
|
? availableWorkers.map(w => `- ${w}`).join('\n ')
|
|
@@ -139,6 +139,25 @@ function buildStandardsSection() {
|
|
|
139
139
|
- Include file paths with line numbers in references
|
|
140
140
|
</formatting>
|
|
141
141
|
|
|
142
|
+
<memory_usage priority="CRITICAL">
|
|
143
|
+
When storing facts with memory_store_fact(), follow these rules:
|
|
144
|
+
|
|
145
|
+
✅ STORE (learned contextual knowledge):
|
|
146
|
+
- Project discoveries: Anti-patterns found, bugs fixed, patterns discovered
|
|
147
|
+
- User preferences: Workflows, communication styles learned through interaction
|
|
148
|
+
- Architectural decisions WHY: Reasoning behind choices, rejected alternatives
|
|
149
|
+
- Solutions that worked: Specific implementations with context
|
|
150
|
+
|
|
151
|
+
❌ DON'T STORE (generic documentation):
|
|
152
|
+
- Tool behavior: How the tool system works (already in API docs)
|
|
153
|
+
- Universal knowledge: Programming concepts, language features
|
|
154
|
+
- Temporary state: Current task status, session-specific data
|
|
155
|
+
|
|
156
|
+
Rule of thumb: "Would I need to rediscover this in 3 months?"
|
|
157
|
+
→ YES = store it (learned pattern, specific to this context)
|
|
158
|
+
→ NO = don't store it (universal knowledge, temporary state)
|
|
159
|
+
</memory_usage>
|
|
160
|
+
|
|
142
161
|
<documentation_files priority="MANDATORY">
|
|
143
162
|
NEVER create .md files in random locations. Follow these rules:
|
|
144
163
|
|
package/dist/tools.js
CHANGED
|
@@ -89,7 +89,27 @@ export function registerToolDefinitions(server) {
|
|
|
89
89
|
},
|
|
90
90
|
{
|
|
91
91
|
name: 'memory_store_fact',
|
|
92
|
-
description:
|
|
92
|
+
description: `Store a specific fact as a relationship between two entities. Use this when you learn something concrete about entities.
|
|
93
|
+
|
|
94
|
+
CRITICAL - WHAT TO STORE:
|
|
95
|
+
✅ LEARNED patterns discovered through work (not generic knowledge)
|
|
96
|
+
✅ PROJECT-SPECIFIC decisions and their rationale (the WHY, not just the WHAT)
|
|
97
|
+
✅ USER PREFERENCES learned through repeated interactions
|
|
98
|
+
✅ SOLUTIONS that worked for specific problems
|
|
99
|
+
✅ Ask yourself: "Would I need to rediscover this in 3 months?" → If YES, store it
|
|
100
|
+
|
|
101
|
+
CRITICAL - WHAT NOT TO STORE:
|
|
102
|
+
❌ Generic tool documentation (that's already in API docs)
|
|
103
|
+
❌ Universal programming knowledge (language features, best practices)
|
|
104
|
+
❌ Temporary state or session-specific data
|
|
105
|
+
❌ System behavior that's documented in the codebase
|
|
106
|
+
❌ Ask yourself: "Is this specific to THIS context or universal?" → If universal, DON'T store it
|
|
107
|
+
|
|
108
|
+
EXAMPLES:
|
|
109
|
+
✅ GOOD: "portal-v2 HAD_SECURITY_ISSUE Mux webhook signature verification returned true unconditionally - Fixed with HMAC-SHA256"
|
|
110
|
+
✅ GOOD: "Owen PREFERS_WORKFLOW Comprehensive analysis before task creation → Review → Approve with 'go' → Parallel dispatch"
|
|
111
|
+
❌ BAD: "task_plan CREATES_TASKS_WITH ready status not pending" (generic system behavior)
|
|
112
|
+
❌ BAD: "React USES JSX for templating" (universal knowledge)`,
|
|
93
113
|
inputSchema: {
|
|
94
114
|
type: 'object',
|
|
95
115
|
properties: {
|
|
@@ -107,7 +127,7 @@ export function registerToolDefinitions(server) {
|
|
|
107
127
|
},
|
|
108
128
|
context: {
|
|
109
129
|
type: 'string',
|
|
110
|
-
description: 'Description or context of the fact',
|
|
130
|
+
description: 'Description or context of the fact - focus on WHY and project-specific details',
|
|
111
131
|
},
|
|
112
132
|
project_id: {
|
|
113
133
|
type: 'string',
|