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.
npx agentmods add skills/chenpg2/kw-engine/kwnpx skills add chenpg2/kw-engine --skill kwgit clone --depth 1 https://github.com/chenpg2/kw-engineWhat 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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00065 | $0.01681 |
| Opus 5 | $0.00032 | $0.00840 |
| Sonnet 5 | $0.00013 | $0.00336 |
| Haiku 4.5 | $0.00006 | $0.00168 |
Grade A, and why
kw 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.
How it starts
The opening of the file, as written. The whole thing — 110 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/kw — knowledge engine (Loop 1)
You are the orchestrator. Do NOT do fetching/reading/distilling/synthesis yourself —
dispatch the subagents. Read .kw/config.yaml and memory/index.json first.
CLI substrate: The kw CLI (from kw-engine package) provides deterministic commands
for mutations. Subagents should use these instead of hand-editing index.json:
kw add-paper <id>— register a paper (creates scaffold md + index entry)kw add-principle --title … --sig … --math … …— allocate P-####kw add-link <from> <to> <type>— add link (preserves formatting)kw search "<query>"— find relevant principleskw fetch <id>— acquire PDF + validate + registerkw reindex— rebuild index.json + SQLite from markdownkw verify— check SCHEMA §6 invariantskw status— show engine state
Hard boundary (cost + separation): never call read_*/search_*/download_* MCP
tools or read a paper PDF in this (opus) loop — that wastes opus on text-ingestion and
bypasses the engine. Acquisition runs on kw-fetcher; reading runs on kw-reader. The
ONLY fetching the orchestrator may do itself is /browse for a NEEDS-BROWSE paywalled
paper (interactive, uses the user's institutional access).
Model enforcement — MUST pass model explicitly on every Agent dispatch:
| subagent | model | subagent_type |
|---|---|---|
| kw-fetcher | sonnet | kw-fetcher |
| kw-reader | sonnet | kw-reader |
| kw-distiller | opus | kw-distiller |
| kw-synthesizer | opus | kw-synthesizer |
| kw-verifier | sonnet | kw-verifier |
Frontmatter model: alone does NOT guarantee the model — the Agent tool inherits the
parent session model by default. You MUST pass BOTH subagent_type AND model in every
Agent call. Example:
Agent({ subagent_type: "kw-reader", model: "sonnet", prompt: "..." })
If you omit model, the reader/fetcher/verifier will run on opus and waste tokens.
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.
- 2d ago First seen · 110 lines · 65 tokens per session scan A 29786288b409
kw is a skill published in the GitHub repository chenpg2/kw-engine (11 stars, last pushed 2mo ago), licensed MIT. It adds 65 tokens to every session and 1,681 once invoked, about $0.0003 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.
Other skills, from other repositories
paper-deep-reader
Skill "paper-deep-reader" from Linwei-Chen/paper-deep-reader-skill, covering paper deep reader, 1. 路由交付模式, 2. 按需加载参考材料, 3. 先读懂，再写报告 and 4. 强制执行双层讲解.
Datacore Methodology
Core knowledge management principles — capture, journal, learn, compound.
PBE Extractor
Extract invariant principles from any text — find the ideas that survive rephrasing.
flux-analyzer
Analyse FBA flux distributions to extract biological insights. Covers gene essentiality, phenotypic phase planes, flux sampling, pathway-level aggregation, secretion product prediction, and production of publication- quality figures.
a-evolve
Apply A-Evolve's agentic evolution methodology to improve AI agent performance across runs. Use when the user wants to diagnose agent failures, generate targeted skills from error patterns, evolve system prompts, or accumulate episodic knowledge. Works standalone or inside AutoResearchClaw pipelines. Triggers on…
statistical-method-design
Design statistical methods, baselines, diagnostics, variants, and ablations that directly address a formal problem formulation.