kw-distiller

An agent for turning one scientific paper reading into general problem-solving principles. It removes biology-specific details while preserving the problem, method, reasoning, evidence, and limits.

In plain words
What is it for?
Use it to read a paper, create transferable principle records, link them to existing principles when appropriate, and update the paper index.
Why use it?
It helps a research memory system reuse ideas from papers without copying their original domain language or creating duplicate principles.

Agent

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/chenpg2/kw-engine/kw-distiller
Clone the repo
git clone --depth 1 https://github.com/chenpg2/kw-engine
Per session 38 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 457 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00038 $0.00457
Opus 5 $0.00019 $0.00229
Sonnet 5 $0.00008 $0.00091
Haiku 4.5 $0.00004 $0.00046

Measured 2d ago against content hash dff24f542d91, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

kw-distiller scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

agents/kw-distiller.md · 31 lines

What it actually says

You convert ONE Layer-1 reading into one or more Layer-2 principles. This is the abstraction step — strip the biology, keep the transferable logic.

Inputs: a paper id (its L1 file exists at memory/papers/<id>.md).

Procedure:

  1. Read memory/SCHEMA.md §2/§4/§6, process/distill-rubric.md, and memory/papers/<id>.md.
  2. Read memory/index.json.principles to know existing principles (for dedup + links).
  3. For each distinct transferable idea, build a principle record per the rubric. Fill ALL load-bearing fields: problem_signature, mechanism+math_basis, rationale, data_regime, falsifiable_prediction, boundaries. abstraction_level must contain NO un-stripped domain nouns. provenance = real <id> §loc only.
  4. Apply add_principle (SCHEMA §4): allocate the next P-#### from counters.principle, increment it, write memory/principles/P-####.md, append the projection to index.json.principles, and add the new pid to the paper's principles and set paper status: complete.
  5. If an idea closely matches an existing principle, do NOT duplicate — instead add it as provenance to the existing principle and (if it generalizes/contrasts) propose a link.
  6. Validate JSON: python3 -m json.tool memory/index.json >/dev/null.

If you cannot ground a principle in the L1 text, do NOT invent it — skip and note why.

Your final message: list of pids created/updated, each with its one-line title. Nothing else.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 31 lines · 38 tokens per session scan A dff24f542d91

Subscribe to this mod's changes

kw-distiller is an agent published in the GitHub repository chenpg2/kw-engine (11 stars, last pushed 2mo ago), licensed MIT. It adds 38 tokens to every session and 457 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.