Obsidian rlm distiller

Skill richfrem/agent-plugins-skills/plugins/obsidian-wiki-engine/skills/obsidian-rlm-distiller

repo for reusable plugins and skills

Install

npx -y skills add richfrem/agent-plugins-skills --skill obsidian-rlm-distiller

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

4 stars4 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Distills wiki source files into the RLM summary layer (summary.md, bullets.md, deep.md) using the cheapest available LLM CLI. Routes to Copilot gpt-5-mini first, then Claude Haiku, then Gemini Flash. Never uses Ollama. Use when wiki nodes need RLM summaries generated or refreshed.

SKILL.md

3.7 KB, as published. Nobody here has run it

Dependencies

Requires Python 3.8+ and at least one CLI installed: copilot, claude, or gemini.

pip install -r requirements.txt

Obsidian RLM Distiller

Status: Active Author: Richard Fremmerlid Domain: Obsidian Wiki Engine Replaces: rlm-distill-ollama (fully deprecated)

Purpose

Distills registered wiki source files into the three-layer RLM summary structure inside {wiki_root}/rlm/{concept}/. Delegates work to the cheapest available LLM CLI — never a local Ollama server.

Cheap-Model Fallback Chain (Strict)

1. copilot CLI available?  → use gpt-5-mini        (fastest, Paid - AI Credits)
2. claude CLI available?   → use claude-haiku-4-5  (fallback, Paid)
3. gemini CLI available?   → use gemini-3-flash-preview (final fallback, Paid)
4. none found              → exit with instructions

rlm-distill-ollama is fully deprecated. Only rlm-distill-agent pointing at cheap cloud models is supported.

Output: Three-Layer RLM Structure

{wiki_root}/rlm/{concept}/
  summary.md    ← 1-5 sentence distilled summary
  bullets.md    ← key idea bullets (6-10 points)
  deep.md       ← full multi-pass distillation

Usage

Distill all stale wiki nodes

python ./scripts/distill_wiki.py --wiki-root /path/to/wiki-root

Distill one named source

python ./scripts/distill_wiki.py --wiki-root /path/to/wiki-root --source arch-docs

Force engine override

python ./scripts/distill_wiki.py --wiki-root /path/to/wiki-root --engine claude
python ./scripts/distill_wiki.py --wiki-root /path/to/wiki-root --engine gemini
python ./scripts/distill_wiki.py --wiki-root /path/to/wiki-root --engine copilot

Use shared .agent/learning/ cache (colocates with rlm-factory)

python ./scripts/distill_wiki.py --wiki-root /path/to/wiki-root \
    --rlm-cache-dir /path/to/project/.agent/learning/rlm_wiki_cache

Dry run

python ./scripts/distill_wiki.py --wiki-root /path/to/wiki-root --dry-run

Engine Detection Logic

distill_wiki.py calls shutil.which() for each CLI in priority order. The first one found and authenticated is used for the entire batch:

ENGINE_PRIORITY = [
    ("copilot", "gpt-5-mini"),
    ("claude",  "claude-haiku-4-5"),
    ("gemini",  "gemini-3-flash-preview"),
]

RLM Cache Storage

distill_wiki.py writes summaries directly into its own RLM cache directory. No cross-plugin script calls are made (ADR-001 compliant).

Default cache: {wiki-root}/rlm/{concept}/

To colocate with rlm-factory under .agent/learning/, pass --rlm-cache-dir:

python ./scripts/distill_wiki.py --wiki-root /path/to/wiki-root \
    --rlm-cache-dir /path/to/project/.agent/learning/rlm_wiki_cache

The cache location is determined by configuration in .agent/learning/rlm_profiles.json (the cache key of the wiki profile) — not by hard-coded cross-plugin paths.

When to Use

After /wiki-ingest populates new wiki nodes
When RLM summaries are missing or stale
Before running /wiki-query for optimal recall
As part of the /wiki-rebuild full pipeline

Related Scripts

distill_wiki.py — cheap-model fallback orchestrator
raw_manifest.py — WikiSourceConfig loader
audit.py — identifies stale/missing RLM summaries

Gives 0 of the 12 instructions most context ai engineering skills give

Counted across 1,193 of the 1,976 authors here whose files we hold, read 2026-08-06

dispatch a fresh implementer subagent per taskin 48 of 1193, across 19 files
dispatch final reviewer after all tasksin 37 of 1193, across 11 files
provide full task text to the subagentin 31 of 1193, across 10 files
review spec compliance before code qualityin 27 of 1193, across 10 files
make the hook script executablein 26 of 1193, across 8 files
re-snapshot after navigation or DOM changesin 25 of 1193, across 17 files
answer subagent questions before proceedingin 22 of 1193, across 7 files
mark task complete in TodoWrite after approvalin 22 of 1193, across 6 files
merge hook into existing settingsin 21 of 1193, across 3 files
read files before editing themin 21 of 1193, across 9 files
ask if installation is global or projectin 20 of 1193, across 2 files
copy the hook script to target locationin 20 of 1193, across 2 files

Said here and by no other author read

install python dependencies
distill stale wiki nodes
distill named source files
use the cheapest available LLM CLI
check copilot CLI first
check claude CLI second

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once.