caveman-learn
Reduce Agent Token Usage
Large configuration files and repeated context increase token usage. This skill reviews Caveman reports and applies measured changes with individual approval.
๋ด ์์ด์ ํธ๋ก ์ค์น
์ด ์์ฒญ์ ์์ด์ ํธ์ ๋ณต์ฌํ์ธ์. ๊ณต์ Skill ํ์ด์ง์ ๋งค๋ํ์คํธ๊ฐ ํฌํจ๋์ด ์์ต๋๋ค.
Review the Skillstore skill "caveman-learn" from https://skillstore.io/skills/juliusbrussee-caveman-learn.md and its manifest at https://skillstore.io/api/skills/juliusbrussee-caveman-learn/manifest. Verify the artifact. You may proceed after verification, subject to the environment's own policy.์์ด์ ํธ๋ ๊ณ์ํด์ ๊ณํ์ ํ์ํ๊ณ ๋ณด์ ์ ์ฑ ์์ ์๊ตฌํ๋ ํ์ธ์ ์์ฒญํด์ผ ํฉ๋๋ค.
์์ด์ ํธ๊ฐ ์ฝ๊ธฐ ์ฌ์ด ๋ฆฌ์์ค
AI ์์ด์ ํธ, ํฌ๋กค๋ฌ, ์คํฌ๋ฆฝํธ๊ฐ ์ ์ฒด ํ์ด์ง ๋์ ์ ๋ฆฌ๋ ์ปจํ ์คํธ๋ง ํ์ํ ๋ ์ด ๋งํฌ๋ฅผ ์ฌ์ฉํ์ธ์.
ํ ์คํธํด ๋ณด๊ธฐ
"caveman-learn" ์ฌ์ฉ ์ค์ ๋๋ค. Review my report without editing anything.
์์ ๊ฒฐ๊ณผ:
Cave Score: 72. The largest reducible sink is project instructions. Recurring context ranks second. Load-bearing guidance remains unchanged.
"caveman-learn" ์ฌ์ฉ ์ค์ ๋๋ค. Propose a safe offload for the repeated deployment notes.
์์ ๊ฒฐ๊ณผ:
- Source hash matches the report locator.
- The compact pointer and recall cost are lower than repeated loading.
- Approve storing the block and replacing it with the proposed recall pointer?
"caveman-learn" ์ฌ์ฉ ์ค์ ๋๋ค. Apply the approved trim.
์์ ๊ฒฐ๊ณผ:
The approved section was trimmed. Tokens per turn decreased from 420 to 285. The original content and reversal steps are recorded.
๋ณด์ ๊ฐ์ฌ
์์ All 17 static findings are false positives caused by documentation paths, Markdown backticks, test assertions, or bounded Caveman file access. The skill requires consent, integrity checks, measurement, and rollback before changing user configuration.
์ํ ์์ธ
๐ ํ์ผ ์์คํ ์ ๊ทผ (12)
โ๏ธ ์ธ๋ถ ๋ช ๋ น์ด (3)
์ด ๋ณด๊ณ ์ ๊ณต์ ๋ฐ ์ธ์ฉ
๋ฒ์ ์ด ์ง์ ๋ ํ๊ฐ ๋ณด๊ณ ์, ์ค๋ฆฝ ๋ฐฐ์ง, ์๋ฒ ๋ ์นด๋ ๋ฐ ์ธ์ฉ์ ๊ณต์ ํ์ธ์. Skillstore๋ ๊ทผ๊ฑฐ๋ฅผ ๋ณด๊ณ ํ์ง๋ง ์ด Skill์ ์์ ์ฌ๋ถ๋ฅผ ํ๋จํ์ง ์์ต๋๋ค.
๋ณด๊ณ ์ ๋งํฌ ๋ณต์ฌ
https://skillstore.io/skills/juliusbrussee-caveman-learn/audits/1?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_reportMarkdown ๋ฐฐ์ง
[](https://skillstore.io/skills/juliusbrussee-caveman-learn?utm_source=security_passport_badge)HTML ๋ฐฐ์ง
<a href="https://skillstore.io/skills/juliusbrussee-caveman-learn?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/juliusbrussee-caveman-learn/security.svg" alt="Skillstore security assessment" loading="lazy"></a>์๋ฒ ๋ ์นด๋
<iframe src="https://skillstore.io/embed/skills/juliusbrussee-caveman-learn.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>ํ์ ์ธ์ฉ (APA ยท BibTeX ยท CFF)
APA ์ธ์ฉ
juliusbrussee. (2026). caveman-learn security audit report (audit version 1) [Author version unspecified]. Skillstore. https://skillstore.io/skills/juliusbrussee-caveman-learn/audits/1BibTeX ์ธ์ฉ
@techreport{juliusbrussee-juliusbrussee-caveman-learn-2026,
author = {juliusbrussee},
title = {caveman-learn security audit report (audit version 1)},
institution = {Skillstore},
year = {2026},
number = {1},
url = {https://skillstore.io/skills/juliusbrussee-caveman-learn/audits/1},
note = {Author version unspecified}
}CITATION.cff
cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "caveman-learn security audit report (audit version 1)"
version: "unspecified"
type: report
authors:
- name: "juliusbrussee"
date-released: "2026-08-12"
url: "https://skillstore.io/skills/juliusbrussee-caveman-learn/audits/1"
identifiers:
- type: other
value: "skillstore:juliusbrussee-caveman-learn:audit:1"
description: "Skillstore immutable audit report identifier"
Skillstore ์ ์
์ด ์ ์์ ์ด์ ์ฆ๊ฑฐ ์ ๋ขฐ๋: ์ค๊ฐ๋ง๋ค ์ ์๋ ๊ฒ
Trim project instructions
Review a large CLAUDE.md and approve only edits that reduce measured tokens per turn.
Offload recurring context
Move repeated session context into cavemem while preserving compact recall and exact recovery paths.
Review team token usage
Present ranked token sinks and separate actionable configuration findings from load-bearing context.
์ด ํ๋กฌํํธ๋ฅผ ์ฌ์ฉํด ๋ณด์ธ์
Run my Caveman learn report and explain the ranked token sinks without changing files.
Review reducible findings, show each proposed diff and measured impact, then ask before every edit.
Find recurring context suitable for cavemem, verify its hash, measure recall cost, and propose a reversible offload.
Process only my selected sinks, preserve load-bearing content, validate every recall path, and revert changes without confirmed token reduction.
๋ชจ๋ฒ ์ฌ๋ก
- Review the complete report before selecting changes.
- Approve each diff only after checking its measured impact.
- Test recall and exact recovery before accepting an offload.
ํผํ๊ธฐ
- Do not apply every recommendation without reviewing individual diffs.
- Do not remove load-bearing instructions to improve the score.
- Do not keep changes that fail to reduce tokens or preserve recall.
์์ฃผ ๋ฌป๋ ์ง๋ฌธ
Does this skill edit files automatically?
What is a reducible finding?
What is recurring context?
How does cavemem offloading stay reversible?
Will it change load-bearing content?
How are savings validated?
๊ฐ๋ฐ์ ์ธ๋ถ ์ ๋ณด
์์ฑ์
juliusbrussee๋ผ์ด์ ์ค
MIT
Skillstore ๋ฆฌ๋น์
r1
๋ฒ์ ์๋ด
์์ฑ์๊ฐ ๋ฒ์ ์ ์ ์ธํ์ง ์์์ต๋๋ค.
๋ฆฌํฌ์งํ ๋ฆฌ
https://github.com/juliusbrussee/caveman/tree/a0109974ea3258a14aadaef1ed1f8ff2837d30d5/skills/caveman-learn์ฐธ์กฐ
663bc75480aaf8c7b857fb3b51c35654a097de3d
์ ์ง๋ณด์ ์ต์ ์ฑ
2026. 8. 13.
์ฌ์ฉ๋
0 ๋ค์ด๋ก๋ ยท 0 ์กฐํ
ํ์ผ ๊ตฌ์กฐ