Auto-generated by trend-scout.py β review and edit as needed.
π What problem it solves
Token-efficient C++ context graph + memory for AI coding assistants. 1110/1110 tests, 40 MCP tools, Apache-2.0.
π
Timeline
| Field |
Value |
| Created |
2026-05-06 |
| Last pushed |
2026-05-29 |
| Stars |
2 |
| Forks |
1 |
| Open issues |
1 |
| License |
Apache-2.0 |
| Language |
Python |
| Topics |
ai-agents, ai-coding, anthropic, claude-code, cli, cline, context-engineering, cpp17, cursor, developer-tools, knowledge-graph, llm-tools, local-first, mcp, mcp-server, prompt-engineering, semantic-search, sqlite, token-optimization |
β
Strengths
- Well-tagged: ai-agents, ai-coding, anthropic, claude-code, cli, cline
- Primary language: Python
- Actively maintained (pushed within 30 days)
β οΈ Weaknesses / Risks
- Low star count (2) β limited community validation
π‘ What this repo can learn
- Claude Code session patterns: this repo's Claude Code integration approach could improve
claude-adapter.py's JSONL parsing β e.g., handling new session event types or extracting richer metadata from Claude Code tool-use blocks
- Graph-based knowledge linking: a relation graph over session entries could let
briefing.py surface related decisions and mistakes by topic proximity β e.g., linking a mistake:auth record to pattern:jwt across separate session files without requiring identical keywords
- Git hook / workflow patterns: hook design from this repo could strengthen the
hooks/ enforcement chain (e.g., auto-briefing, commit guards, learn reminders)
- Editor integration (cursor):
watch-sessions.py could be extended to detect and parse cursor session formats natively, broadening the range of AI sessions indexed into knowledge.db
- MCP tool-server surface: the MCP server interface here could expose
query-session.py and briefing.py as directly callable MCP tools β e.g., a Copilot agent could invoke the briefing lookup in-process rather than shelling out to a subprocess, reducing latency in hook-driven workflows
README excerpt
<p align="center">
<img src="assets/logo.svg" alt="Icemage" width="160"/>
</p>
# Icemage <sub><sup>(`icmg`)</sup></sub>
[](https://github.com/ncmonx/icm-graph/releases)
[](https://github.com/ncmonx/icm-graph/releases)
[](https://github.com/ncmonx/icm-graph/commits/main)
[](#)
[](#)
[](LICENSE)
[](https://securityscorecards.dev/viewer/?uri=github.com/ncmonx/icm-graph)
[](https://www.bestpractices.dev/projects/12818)
[](https://github.com/sponsors/ncmonx)
[](https://ko-fi.com/ncmonx)
> **Stop burning tokens. Stop losing context. Ship faster.**
A small helper app that makes AI coding assistants β Claude Code, Cursor, and friends β **70 β 98 % cheaper**
*(truncated)*
Scouted on 2026-05-29 Β· View on GitHub
π Trend Scout: ncmonx/icm-graph
π What problem it solves
Token-efficient C++ context graph + memory for AI coding assistants. 1110/1110 tests, 40 MCP tools, Apache-2.0.
π Timeline
β Strengths
π‘ What this repo can learn
claude-adapter.py's JSONL parsing β e.g., handling new session event types or extracting richer metadata from Claude Code tool-use blocksbriefing.pysurface related decisions and mistakes by topic proximity β e.g., linking amistake:authrecord topattern:jwtacross separate session files without requiring identical keywordshooks/enforcement chain (e.g., auto-briefing, commit guards, learn reminders)watch-sessions.pycould be extended to detect and parse cursor session formats natively, broadening the range of AI sessions indexed intoknowledge.dbquery-session.pyandbriefing.pyas directly callable MCP tools β e.g., a Copilot agent could invoke the briefing lookup in-process rather than shelling out to a subprocess, reducing latency in hook-driven workflowsREADME excerpt
Scouted on 2026-05-29 Β· View on GitHub