cm

Name: cm
Author: Dicklesworthstone

69stars13forks

How to install cm?

dicklesworthstone/agent_flywheel_clawdbot_skills_and_integrations/cm

$npx -y skills add Dicklesworthstone/agent_flywheel_clawdbot_skills_and_integrations --skill cm

Installs into the current project.

Summary

CASS Memory System - procedural memory for AI coding agents. Three-layer cognitive architecture with confidence decay, anti-pattern learning, cross-agent knowledge transfer, trauma guard safety system. Bun/TypeScript CLI.

Files · 1

View on GitHub

SKILL.md

1	# CM - CASS Memory System
2
3	Procedural memory for AI coding agents. Transforms scattered sessions into persistent, cross-agent memory. Uses a three-layer cognitive architecture that mirrors human expertise development.
4
5	## Why This Exists
6
7	AI coding agents accumulate valuable knowledge but it's:
8	- Trapped in sessions - Context lost when session ends
9	- Agent-specific - Claude doesn't know what Cursor learned
10	- Unstructured - Raw logs aren't actionable guidance
11	- Subject to collapse - Naive summarization loses critical details
12
13	You've solved auth bugs three times this month across different agents. Each time you started from scratch.
14
15	CM solves this with cross-agent learning: a pattern discovered in Cursor is immediately available to Claude Code.
16
17	---
18
19	## Three-Layer Cognitive Architecture
20
21	```
22	┌─────────────────────────────────────────────────────────────────────┐
23	│ EPISODIC MEMORY (cass) │
24	│ Raw session logs from all agents — the "ground truth" │
25	│ Claude Code │ Codex │ Cursor │ Aider │ PI │ Gemini │ ChatGPT │ ...│
26	└───────────────────────────┬─────────────────────────────────────────┘
27	│ cass search
28	▼
29	┌─────────────────────────────────────────────────────────────────────┐
30	│ WORKING MEMORY (Diary) │
31	│ Structured session summaries: accomplishments, decisions, etc. │
32	└───────────────────────────┬─────────────────────────────────────────┘
33	│ reflect + curate (automated)
34	▼
35	┌─────────────────────────────────────────────────────────────────────┐
36	│ PROCEDURAL MEMORY (Playbook) │
37	│ Distilled rules with confidence tracking and decay │
38	└─────────────────────────────────────────────────────────────────────┘
39	```
40
41	Every agent's sessions feed the shared memory. A pattern discovered in Cursor automatically helps Claude Code on the next session.
42
43	---
44
45	## The One Command You Need
46
47	```bash
48	cm context "<your task>" --json
49	```
50
51	Run this before starting any non-trivial task. Returns:
52	- relevantBullets - Rules from playbook scored by task relevance
53	- antiPatterns - Things that have caused problems
54	- historySnippets - Past sessions (yours and other agents')
55	- suggestedCassQueries - Deeper investigation searches
56
57	### Filtering History by Source
58
59	`historySnippets[].origin.kind` is `"local"` or `"remote"`. Remote hits include `origin.host`:
60
61	```json
62	{
63	"historySnippets": [
64	{
65	"source_path": "~/.claude/sessions/session-001.jsonl",
66	"origin": { "kind": "local" }
67	},
68	{
69	"source_path": "/home/user/.codex/sessions/session.jsonl",
70	"origin": { "kind": "remote", "host": "workstation" }
71	}
72	]
73	}
74	```
75
76	---
77
78	## Confidence Decay System
79
80	Rules aren't immortal. Confidence decays without revalidation:
81
82	\| Mechanism \| Effect \|
83	\|-----------\|--------\|
84	\| 90-day half-life \| Confidence halves every 90 days without feedback \|
85	\| 4x harmful multiplier \| One mistake counts 4× as much as one success \|
86	\| Maturity progression \| `candidate` → `established` → `proven` \|
87
88	### Score Decay Visualization
89
90	```
91	Initial score: 10.0 (10 helpful marks today)
92
93	After 90 days (half-life): 5.0
94	After 180 days: 2.5
95	After 270 days: 1.25
96	After 365 days: 0.78
97	```
98
99	### Effective Score Formula
100
101	```typescript
102	effectiveScore = decayedHelpful - (4 × decayedHarmful)
103
104	// Where decay factor = 0.5 ^ (daysSinceFeedback / 90)
105	```
106
107	### Maturity State Machine
108
109	```
110	┌──────────┐ ┌─────────────┐ ┌────────┐
111	│ candidate│──────▶│ established │───▶│ proven │
112	└──────────┘ └─────────────┘ └────────┘
113	│ │ │
114	│ │ (harmful >25%) │
115	│ ▼ │
116	│ ┌─────────────┐ │
117	└────────────▶│ deprecated │◀─────────┘
118	└─────────────┘
119	```
120
121	Transition Rules:
122
123	\| Transition \| Criteria \|
124	\|------------\|----------\|
125	\| `candidate` → `established` \| 3+ helpful, harmful ratio <25% \|
126	\| `established` → `proven` \| 10+ helpful, harmful ratio <10% \|
127	\| `any` → `deprecated` \| Harmful ratio >25% OR explicit deprecation \|
128
129	---
130
131	## Anti-Pattern Learning
132
133	Bad rules don't just get deleted. They become warnings:
134
135	```
136	"Cache auth tokens for performance"
137	↓ (3 harmful marks)
138	"PITFALL: Don't cache auth tokens without expiry validation"
139	```
140
141	When a rule is marked harmful multiple times (>50% harmful ratio with 3+ marks), it's automatically inverted into an anti-pattern.
142
143	---
144
145	## ACE Pipeline (How Rules Are Created)

People who liked this also viewed

6 picks

Type

cm

69stars13forks

View on GitHub

How to install cm?

dicklesworthstone/agent_flywheel_clawdbot_skills_and_integrations/cm

$npx -y skills add Dicklesworthstone/agent_flywheel_clawdbot_skills_and_integrations --skill cm

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# CM - CASS Memory System
2
3	Procedural memory for AI coding agents. Transforms scattered sessions into persistent, cross-agent memory. Uses a three-layer cognitive architecture that mirrors human expertise development.
4
5	## Why This Exists
6
7	AI coding agents accumulate valuable knowledge but it's:
8	- Trapped in sessions - Context lost when session ends
9	- Agent-specific - Claude doesn't know what Cursor learned
10	- Unstructured - Raw logs aren't actionable guidance
11	- Subject to collapse - Naive summarization loses critical details
12
13	You've solved auth bugs three times this month across different agents. Each time you started from scratch.
14
15	CM solves this with cross-agent learning: a pattern discovered in Cursor is immediately available to Claude Code.
16
17	---
18
19	## Three-Layer Cognitive Architecture
20
21	```
22	┌─────────────────────────────────────────────────────────────────────┐
23	│ EPISODIC MEMORY (cass) │
24	│ Raw session logs from all agents — the "ground truth" │
25	│ Claude Code │ Codex │ Cursor │ Aider │ PI │ Gemini │ ChatGPT │ ...│
26	└───────────────────────────┬─────────────────────────────────────────┘
27	│ cass search
28	▼
29	┌─────────────────────────────────────────────────────────────────────┐
30	│ WORKING MEMORY (Diary) │
31	│ Structured session summaries: accomplishments, decisions, etc. │
32	└───────────────────────────┬─────────────────────────────────────────┘
33	│ reflect + curate (automated)
34	▼
35	┌─────────────────────────────────────────────────────────────────────┐
36	│ PROCEDURAL MEMORY (Playbook) │
37	│ Distilled rules with confidence tracking and decay │
38	└─────────────────────────────────────────────────────────────────────┘
39	```
40
41	Every agent's sessions feed the shared memory. A pattern discovered in Cursor automatically helps Claude Code on the next session.
42
43	---
44
45	## The One Command You Need
46
47	```bash
48	cm context "<your task>" --json
49	```
50
51	Run this before starting any non-trivial task. Returns:
52	- relevantBullets - Rules from playbook scored by task relevance
53	- antiPatterns - Things that have caused problems
54	- historySnippets - Past sessions (yours and other agents')
55	- suggestedCassQueries - Deeper investigation searches
56
57	### Filtering History by Source
58
59	`historySnippets[].origin.kind` is `"local"` or `"remote"`. Remote hits include `origin.host`:
60
61	```json
62	{
63	"historySnippets": [
64	{
65	"source_path": "~/.claude/sessions/session-001.jsonl",
66	"origin": { "kind": "local" }
67	},
68	{
69	"source_path": "/home/user/.codex/sessions/session.jsonl",
70	"origin": { "kind": "remote", "host": "workstation" }
71	}
72	]
73	}
74	```
75
76	---
77
78	## Confidence Decay System
79
80	Rules aren't immortal. Confidence decays without revalidation:
81
82	\| Mechanism \| Effect \|
83	\|-----------\|--------\|
84	\| 90-day half-life \| Confidence halves every 90 days without feedback \|
85	\| 4x harmful multiplier \| One mistake counts 4× as much as one success \|
86	\| Maturity progression \| `candidate` → `established` → `proven` \|
87
88	### Score Decay Visualization
89
90	```
91	Initial score: 10.0 (10 helpful marks today)
92
93	After 90 days (half-life): 5.0
94	After 180 days: 2.5
95	After 270 days: 1.25
96	After 365 days: 0.78
97	```
98
99	### Effective Score Formula
100
101	```typescript
102	effectiveScore = decayedHelpful - (4 × decayedHarmful)
103
104	// Where decay factor = 0.5 ^ (daysSinceFeedback / 90)
105	```
106
107	### Maturity State Machine
108
109	```
110	┌──────────┐ ┌─────────────┐ ┌────────┐
111	│ candidate│──────▶│ established │───▶│ proven │
112	└──────────┘ └─────────────┘ └────────┘
113	│ │ │
114	│ │ (harmful >25%) │
115	│ ▼ │
116	│ ┌─────────────┐ │
117	└────────────▶│ deprecated │◀─────────┘
118	└─────────────┘
119	```
120
121	Transition Rules:
122
123	\| Transition \| Criteria \|
124	\|------------\|----------\|
125	\| `candidate` → `established` \| 3+ helpful, harmful ratio <25% \|
126	\| `established` → `proven` \| 10+ helpful, harmful ratio <10% \|
127	\| `any` → `deprecated` \| Harmful ratio >25% OR explicit deprecation \|
128
129	---
130
131	## Anti-Pattern Learning
132
133	Bad rules don't just get deleted. They become warnings:
134
135	```
136	"Cache auth tokens for performance"
137	↓ (3 harmful marks)
138	"PITFALL: Don't cache auth tokens without expiry validation"
139	```
140
141	When a rule is marked harmful multiple times (>50% harmful ratio with 3+ marks), it's automatically inverted into an anti-pattern.
142
143	---
144
145	## ACE Pipeline (How Rules Are Created)

People who liked this also viewed

6 picks

Type