Project Context Mapper (Memory Optimizer)
English
Orchestration & Integration
Connects and orchestrates with relevant domain skills like brainstorming, zero-to-prod-orchestrator, and project-context-mapper to ensure cohesive execution.
Description
LLMs struggle with repositories containing hundreds of files; they lose track of where things are, leading to hallucinations and excessive tool calls. This skill forces the agent to create, read, and maintain a CONTEXT_MAP.md—a highly compressed index of the repository's architecture, dependency graphs, and critical file locations. This acts as the agent's "photographic memory".
Trigger Conditions
Activate this skill when:
- The agent enters a massive, existing codebase for the first time.
- The user complains that the AI is "editing the wrong file" or "forgetting how the app is structured."
- Context token usage is dangerously high and needs optimization.
Core Concepts
1. The Context Map Format
The agent should execute a script or traverse the directory to generate a compressed map in CONTEXT_MAP.md (or within .agents/).
# CONTEXT MAP (Do not delete)
## Architecture
- Frontend: Next.js App Router (src/app)
- Backend: Hono RPC (src/server)
- DB: Drizzle (src/db/schema.ts)
## Dependency Graph (Critical Paths)
- `src/app/page.tsx` -> `src/components/ui/Button.tsx`
- `src/server/routes/users.ts` -> `src/db/queries/users.ts`
## State Locations
- Global Auth: `src/store/auth.ts` (Zustand)
2. Workflow
- On Start: The agent reads
CONTEXT_MAP.mdinstead of blindly listing directories. - On Modification: When the agent creates a new major component, route, or service, it MUST append that location to the context map.
Best Practices
- Keep it Ultra-Compressed: Do not store full code in the map. Store ONLY file paths, responsibilities (1 sentence), and connections.
- BM25 Search First: Before writing to the context map, consider using
grep_searchor semantic search for targeted lookups. The map is for architectural understanding, not exact variable finding.
Integration with Other Skills (MANDATORY)
session-context-loader— Integrates deeply to ensure the context map is loaded automatically on every new chat session.token-saver— The map prevents the agent from reading 10 separate files just to find one function, saving massive amounts of tokens.monorepo-architect— Essential for mapping out package boundaries in Turborepo/pnpm workspaces.
Referenced By Orchestrators (MANDATORY)
brainstorming— Add to "Discovery & Audit".zero-to-prod-orchestrator— Phase 1 (Discovery & Architecture).
Bahasa Indonesia
Integrasi Orkestrasi
Terhubung dan mengorkestrasi skill domain yang relevan seperti brainstorming, zero-to-prod-orchestrator, dan project-context-mapper untuk memastikan eksekusi yang kohesif.
Deskripsi
Kelemahan utama agen AI adalah keterbatasan jendela konteks (context window). Skill ini memberi agen "ingatan fotografis" dengan mewajibkannya memelihara file CONTEXT_MAP.md—sebuah indeks super padat yang memetakan arsitektur proyek, lokasi file penting, dan graf dependensi.
Kondisi Pemicu
- Saat agen baru saja dimasukkan ke dalam repositori raksasa yang sudah ada (bukan dari nol).
- Saat agen mulai salah mengedit file atau terlihat kebingungan dengan struktur folder.
Panduan Singkat
- Indeks Tersentralisasi: Buat dan selalu perbarui
CONTEXT_MAP.md. File ini berisi daftar path direktori krusial dan tanggung jawab masing-masing (misal: "Logic pembayaran ada disrc/lib/stripe.ts"). - Ultra-Ringkas: Jangan menaruh baris kode di dalam peta konteks ini. Hanya gunakan format bullet points agar konsumsi token saat dibaca sangat rendah.
- Baca di Awal Sesi: Di setiap awal percakapan, agen tidak perlu lagi meraba-raba melakukan
lsberulang kali. Cukup gunakanview_filepadaCONTEXT_MAP.md, dan agen langsung mengerti seluruh topologi proyek.