agentleFS
Arcanea-LabsArcanea-Labs/Starlight-Intelligence-System/skills/memory/insight-distillation/SKILL.md

memory/insight-distillation

Skill10 starsChanged 2 months agoMITView on GitHub

Use during periodic review sessions (weekly/monthly/quarterly) to turn raw captures into reusable frameworks — patterns captured ≥3 times become named, documented, ready-to-reuse frameworks. Default skill for Starlight Second Brain.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Default for: Starlight Second Brain

    line 15
  2. Distinctive:

    Read all captures in the window.

    line 45
  3. Very specific:

    Write proposals to second-brain/genius-evolution/<slug>.md.

    line 98

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Insight Distillation

"Three captures is a pattern. Five is a framework. Seven is a signature. Count."

When This Skill Activates

  • Periodic review sessions — weekly, monthly, quarterly
  • Keywords: distill, synthesize captures, monthly review, what did I learn, connect the dots, pattern across notes
  • Default for: Starlight Second Brain
  • Composes with: memory/knowledge-synthesis, intelligence/pattern-recognition, memory/vault-management

What This Skill Does

Turns raw captures into reusable frameworks. Captures are cheap; frameworks are portable IP. Distillation is the process of noticing the signal across the noise — patterns captured ≥3 times become named, documented, ready-to-reuse frameworks.

Distillation runs on three cadences:

  • Weekly (lightweight, 15 min of the 30-min review): cluster inbox + new captures by theme; identify emerging patterns
  • Monthly (90 min dedicated session): elevate validated patterns to named frameworks; archive raw captures
  • Quarterly (half-day integration): cross-reference against Genius Profile; propose evolution or net-new frameworks

Core Principles

  1. Count occurrences, don't vibe. A pattern is only real at ≥3 occurrences across the window. Two is coincidence. One is anecdote. Inherited from Genius excavation — the rule is the same for personal ongoing capture as it is for one-time excavation.
  2. Distill before you dedupe. Two captures saying similar things is a signal; one "master note" erases the frequency signal. Cluster first, count second, elevate third.
  3. Distillate preserves; raw archives. Once a pattern elevates to a framework, the raw captures can archive — the distillate carries forward.
  4. Not every cluster becomes a framework. Sometimes a theme shows up 3 times and the honest answer is "interesting, not yet useful." Note the theme; don't force elevation.
  5. Cross-reference before naming. A pattern that resembles an existing Genius framework is probably evolution, not net-new. Check before naming.

Procedures

Procedure 1: Load Captures From Window

Scope by cadence:

  • Weekly: Last 7 days of captures + current inbox
  • Monthly: Last 30 days of captures + accumulated weekly clusters
  • Quarterly: Last 90 days of monthly distillates + Genius Profile for cross-reference

Read all captures in the window. Do not skim — clustering requires reading in full. This is why weekly is 15 min, not 60 min: fewer captures, less to read.

Procedure 2: Cluster By Theme

Group captures that touch the same underlying theme. A theme is not a category — it's an emerging thing the person has been thinking about repeatedly.

Example (a practitioner's monthly distillation):

  • Cluster A: 5 captures about client-calls where the framework worked even when the client resisted → theme: "framework resilience under client friction"
  • Cluster B: 3 captures about why one vocabulary word kept returning across unrelated coaching conversations → theme: "vocabulary as diagnostic"
  • Cluster C: 2 captures about a new energy-management observation → theme: still emerging, watch next month

Clusters are named in the person's own voice, not a generic category. "Framework resilience under client friction" is what the practitioner would say; "client-handling patterns" is what a template would say. Voice matters.

Procedure 3: Identify Emerging Patterns (≥3 Threshold)

For each cluster, count occurrences. Apply the threshold:

  • ≥3 occurrences → eligible for elevation to framework (proceed to step 4)
  • 2 occurrences → coincidence; hold for next window
  • 1 occurrence → anecdote; archive or drop

Record the count. The count is part of the distillate — "observed 5 times across March" is stronger evidence than "I noticed this."

Procedure 4: Elevate Patterns To Frameworks

For each eligible cluster, write a framework draft:

### Framework: <name in person's voice>

**Observed:** <N times across <window>>
**Evidence:** <3-5 capture references — dates + one-line excerpts>

**What it says:** <2-3 sentences naming the pattern>

**When it applies:** <context in which this pattern holds>

**When it fails:** <counter-examples or boundary conditions noticed>

**Status:** draft / validated / deprecated

A framework starts as draft. It becomes validated at the next cadence (monthly → quarterly) if still observed. Deprecated if contradicted by later captures.

Procedure 5: Cross-Reference Against Genius Profile

Load the person's Genius Profile (genius/profile-<slug>.md). For each new framework draft, answer:

  • Does this resemble an existing Profile framework?
    • Yes, refinement → propose evolution to that framework; do not create net-new
    • Yes, contradiction → flag for next excavation cycle; the Profile may be outdated
    • No resemblance → candidate for net-new Profile framework at quarterly integration

Write proposals to second-brain/genius-evolution/<slug>.md. Never mutate the Profile directly — Genius agent confirms evolutions.

Procedure 6: Archive Raw Captures

Once a cluster has distilled into a framework draft, archive the raw captures:

  • Raw captures → second-brain/archive/<YYYY-MM>/
  • Distillate (the framework draft) → personal/<slug>/wisdom-vault.md under ## Personal Frameworks section
  • Cluster notes that didn't elevate → second-brain/clusters-<window>-<date>.md (preserves them for next window's counting)

The distillate carries the signal forward. The raw captures are preserved for provenance and re-distillation if the framework later matures or is contested.

Rules

  • Never elevate a pattern to framework below the ≥3 threshold. The threshold is the discipline.
  • Never mutate the Genius Profile directly. Propose; confirm via Genius agent at quarterly review.
  • Never dedupe captures before counting. Frequency is signal — collapsing hides it.
  • Never archive captures whose clusters haven't distilled. Preserve until the next window counts them.
  • Always count in the person's voice. "Observed 5 times" not "n=5."
  • Always cross-reference with Genius Profile before naming net-new frameworks. Most emerging patterns are evolution, not discovery.
  • Always let clusters die honestly. Not every theme becomes a framework. "Interesting, not yet useful" is a valid final state.

Integration Points

  • Vault: Reads second-brain/inbox/, second-brain/captures/; writes second-brain/distillates/, second-brain/archive/, personal/<slug>/wisdom-vault.md, second-brain/genius-evolution/<slug>.md.
  • Agents: Starlight Second Brain (primary). Genius (evolution proposals). Sage (cross-reference with institutional Wisdom).
  • Commands: /distill-insights invokes this skill directly; /orchestrate-brain runs it in the weekly 30-min ritual at lightweight depth.
  • Skills: Composes with memory/knowledge-synthesis (clustering), intelligence/pattern-recognition (threshold recognition), memory/vault-management (persistence), iterative-retrieval (progressive refinement across windows).

Quality Criteria

  • Did every framework elevation meet the ≥3 threshold?
  • Did the monthly distillation produce ≥1 elevation or honestly note "no signal yet"?
  • Were raw captures archived after distillation, preserving the distillate?
  • Was the Genius Profile cross-referenced before naming net-new frameworks?
  • Did the distillate read as the person's voice, not generic template language?
  • Would re-reading the distillate in 6 months still make sense without the raw captures?

Built on SIP — Starlight Intelligence Protocol

  • Substrate: starlightintelligence.org/protocol v1.1.0
  • Layers used: [file-contract, attestation, commands, sovereignty]
  • Verticals: starlight-intelligence-system@v7.4 (SBIS alpha)
  • Generated: 2026-04-24
  • Attestation is compounding, not credit transfer: every composition strengthens every node.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.