القائمة
Prepare source content for RLM processing by discovering files, chunking each one, and writing a unified searchable manifest
يُحدَّد Prompt الذي يراجع المصدر أولًا بشكل افتراضي. يمكنك التبديل إلى أمر مباشر أو تنزيل نسخة محلية.
اقرأ SKILL.md وأي ملفات مرافقة يعرضها SkillsMP قبل أن تقرر التثبيت.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
يتجاوز الأمر المباشر Prompt المخصّص للمراجعة. افحص المصدر قبل تشغيله.
npx skills add https://github.com/jmagly/aiwg --skill rlm-prepيبقى الأمر في سطر واحد. مرّر أفقيًا لمراجعته كاملًا قبل النسخ.
تفضّل نسخة محلية؟ نزّل الملفات المتاحة حاليًا لدى SkillsMP.
عرض SKILL.md
استنادا إلى تصنيف SOC المهني
| namespace | aiwg |
| name | rlm-prep |
| platforms | ["all"] |
| description | Prepare source content for RLM processing by discovering files, chunking each one, and writing a unified searchable manifest |
Prepare source content for RLM processing in one shot: discovers files, chunks each one, builds a searchable index, and writes a unified manifest.json. Run this once on a codebase or document set; then use rlm-search or fanout against the output without re-preparing.
Alternate expressions and non-obvious activations:
docs/| Pattern | Example | Action |
|---|---|---|
| Prep a directory | "prepare src/ for RLM" | rlm-prep src/ |
| Prep a single file | "prep this file for recursive search" | rlm-prep path/to/file.ts |
| Strategy override | "prep with fixed-count chunking" | --strategy fixed-count |
| Size override | "prep in 100-line chunks" | --size 100 |
| Custom output | "prep into tmp/rlm/" | --output tmp/rlm/ |
| Force refresh | "re-prep even if already done" | --force |
| Check status | "is this codebase already prepped?" | Inspect output dir for manifest |
When triggered:
Resolve source — determine whether the input is a single file or a directory. For directories, discover all supported file types (.ts, .js, .py, .go, .md, .txt, .yaml, .json, .sql, and others). Respect .gitignore patterns.
Check for existing prep — look for a manifest in the output directory. If found and --force is not set, report that prep already exists and offer to use it or re-run.
Chunk each file — apply the selected strategy per file. Each file produces its own subdirectory under chunks/, named after the file path (slashes replaced with underscores).
Build index — construct a searchable index (index.json) with:
Write unified manifest — a single manifest.json at the output root that references all chunks across all files. This is what fanout and rlm-search consume.
Report result — print file count, total chunk count, index size, and output path.
.aiwg/rlm-prep/<source-hash>/
├── manifest.json # Unified chunk manifest (all files)
├── index.json # Searchable index with summaries
├── meta.json # Source path, strategy, timestamp
└── chunks/
├── src__auth__middleware.ts/
│ ├── chunk-0001.txt
│ ├── chunk-0002.txt
│ └── chunk-0003.txt
├── src__auth__jwt.ts/
│ ├── chunk-0001.txt
│ └── chunk-0002.txt
└── src__core__parser.ts/
├── chunk-0001.txt
├── chunk-0002.txt
├── chunk-0003.txt
└── chunk-0004.txt
{
"source": "src/auth/",
"source_hash": "sha256:a1b2c3d4...",
"strategy": "semantic-boundary",
"chunk_size": 200,
"overlap": 20,
"created_at": "2026-04-01T14:23:00Z",
"files": 12,
"total_chunks": 47,
"output_dir": ".aiwg/rlm-prep/a1b2c3d4/",
"chunks": [
{
"id": "src__auth__middleware.ts/chunk-0001",
"file_source": "src/auth/middleware.ts",
"chunk_file": ".aiwg/rlm-prep/a1b2c3d4/chunks/src__auth__middleware.ts/chunk-0001.txt",
"start_line": 1
<file|dir> — Source file or directory to prepare (required)--output <dir> — Output directory (default: .aiwg/rlm-prep/<source-hash>/)--strategy semantic-boundary|fixed-count|adaptive — Chunking strategy (default: semantic-boundary)--size N — Target chunk size in lines (default: 200)--overlap N — Overlap lines between adjacent chunks (default: 20)--force — Re-prep even if a manifest already existsUser: "prepare src/ for RLM processing"
Action:
aiwg rlm-prep src/
Response: "Prepped src/ for RLM. 12 files, 47 chunks. Strategy: semantic-boundary (200 lines, 20 overlap). Manifest: .aiwg/rlm-prep/a1b2c3d4/manifest.json"
User: "index the entire repo for RLM, use 100-line chunks"
Action:
aiwg rlm-prep . --size 100 --overlap 15
Response: "Prepped . for RLM. 84 files, 312 chunks. Strategy: semantic-boundary (100 lines, 15 overlap). Manifest: .aiwg/rlm-prep/b3c4d5e6/manifest.json"
User: "get the docs folder ready for recursive search"
Action:
aiwg rlm-prep docs/ --strategy fixed-count --size 150
Response: "Prepped docs/ for RLM. 23 files, 89 chunks. Strategy: fixed-count (150 lines, 20 overlap). Manifest: .aiwg/rlm-prep/c4d5e6f7/manifest.json"
User: "re-prep the auth module, I've made changes"
Action:
aiwg rlm-prep src/auth/ --force
Response: "Re-prepped src/auth/ (previous prep from 2026-03-28 replaced). 4 files, 14 chunks. Manifest: .aiwg/rlm-prep/d5e6f7a8/manifest.json"
User: "is src/ already prepped for RLM?"
Action: Check .aiwg/rlm-prep/ for a manifest matching the source hash of src/.
Response: "Yes — src/ was prepped on 2026-04-01 (47 chunks, strategy: semantic-boundary). Run with --force to re-prep."
If the user's intent is ambiguous: