agentleFS
lavine888lavine888/career-alpha/skills/proof/SKILL.md

proof

Skill0 starsChanged 3 months agoMITView on GitHub

Career Alpha 证据层。用于把项目、PR、实习、比赛、研究或真实工作整理成可核验的 Evidence Card 和 Claim–Evidence Ledger;当用户希望把经历变得可证明、整理作品集证据、补齐简历强主张、审计数字/ownership、或为面试准备事实底稿时使用。

17 other projects use a version of this.

Adopted by

Most recently taken up by

The file

Files
SKILL.md

/proof — Evidence Layer

/proof 的目标是把:

“我做过这个”

变成:

“这条主张由哪些证据支持、我本人负责什么、能安全说到哪一步、最容易在哪被追问穿。”

它是 Career Alpha 的事实层。下游 /position 和 /interview 都不应该绕过它随意升级主张。


1. Collect raw material first

优先收集原始材料,而不是先写简历:

  • repo / commit / branch;
  • PR / issue / review;
  • benchmark / CSV / logs;
  • demo / deployment;
  • architecture / design docs;
  • screenshots;
  • competition result;
  • user / customer feedback;
  • internal docs;
  • emails / messages;
  • timeline notes;
  • 用户口述。

按 Evidence Taxonomy 区分:

  • direct public evidence;
  • direct private evidence;
  • reproducible technical evidence;
  • external review evidence;
  • self-reported evidence;
  • planned evidence。

不要把所有材料都视为同等强度。


2. Build atomic claims

不要直接审计一句漂亮 bullet,要先拆成 atomic claims。

例如:

Led development of an agent platform that improved task completion by 28%.

至少包含:

  1. 平台存在;
  2. 用户承担 lead ownership;
  3. task completion 被测量;
  4. 数字是 28%;
  5. 改进与该工作的关系成立。

每个 atomic claim 可以有不同 confidence。

完整规则见 Claim Audit Protocol。


3. Assign confidence

统一使用:

  • VERIFIED — 有直接、足够强的证据支持该 atomic claim;
  • SUPPORTED — 有材料支持,但仍有不能完全核验 / 因果 / ownership 不完整的部分;
  • SELF-REPORTED — 目前主要来自用户陈述;
  • PLANNED — 尚未完成。

Important

一个项目可以是 VERIFIED,其中的“提升 28%”仍然只是 SUPPORTED。

Confidence 属于 claim,不是整个项目。


4. Numerical audit

所有精确数字优先审计:

  • metric 定义;
  • baseline;
  • sample / denominator;
  • measurement date;
  • raw data;
  • repeated trials;
  • model / dependency version;
  • percentage vs percentage-point;
  • external vs internal benchmark;
  • exact vs approximate。

Unsafe pattern

Improved accuracy by 32%.

但没有:baseline、metric、raw output。

Safer pattern

Observed completion increase from 61% to 78% on a fixed 50-task internal benchmark after introducing retry and context policies.

如果 causal evidence 不够,不要用过强的“caused / drove / improved because of”。


5. Ownership audit

把 ownership 拆开:

  • problem definition;
  • decision ownership;
  • implementation;
  • integration;
  • coordination;
  • evaluation;
  • result responsibility;
  • team contribution;
  • AI / coding-tool assistance。

Strong verbs

以下动词需要更高证据门槛:

  • led;
  • owned;
  • architected;
  • designed;
  • drove;
  • founded;
  • managed。

AI 辅助写代码不自动取消 ownership。关键是用户是否能解释:

  • 为什么这样定义问题;
  • 为什么选这个架构;
  • 怎么验收;
  • 怎么 debug;
  • 哪些方案被放弃;
  • 如何判断结果可信。

6. Status audit

严格区分:

  • planned;
  • prototype;
  • local demo;
  • internal deployment;
  • public deployment;
  • PR open;
  • PR merged;
  • finalist;
  • track placement;
  • overall placement。

不要为了叙事顺畅把不同状态合并。

例如:

PR open ≠ contributed a merged change
prototype ≠ production system
track #2 ≠ competition runner-up overall

7. Causal / scale audit

Causal words

improved / reduced / increased / drove / caused

要求 baseline 或合理 comparison。

Scale words

production-grade / enterprise / large-scale / real-time / high-concurrency / autonomous

要求具体 scale evidence。

如果证据不够,删掉形容词通常比补一个“听起来合理”的数字更好。


8. Create Evidence Cards

每条高价值 claim 形成:

claim:
why_it_matters:
atomic_claims:
evidence:
  - type:
    reference:
result:
ownership:
confidence:
resume_safe_wording:
aggressive_but_defensible_wording:
strong_wording_requires:
interview_risks:
missing_evidence:
privacy_boundary:

需要结构化保存时,使用:

  • ledger template
  • JSON Schema
  • Claim–Evidence Ledger

模板默认 local-only。


9. Evidence conflicts

如果两个来源冲突:

  1. 不要取平均;
  2. 同时保留;
  3. 判断哪个更接近原始事件;
  4. 标记 unresolved;
  5. 在解决前降低 claim 强度。

例如用户记得“约 40%”,benchmark CSV 显示 31.6%,优先使用原始数据,并说明口述与记录有差异。


10. Privacy

敏感材料包括:

  • 招聘邮件;
  • 私人联系方式;
  • 内部链接;
  • 未公开项目;
  • 客户数据;
  • 薪资;
  • NDA 内容。

可以用于当前证据判断,但不要默认写进公开 repo、README 或 public ledger。

公开证据与私有证据分开记录。


11. Find missing evidence

每个 Strong / Future claim 都回答:

还差什么最小证据,才能提高一个 confidence level?

示例:

SUPPORTED → VERIFIED
Need: raw benchmark + pinned config + reproducible run command
SELF-REPORTED ownership → SUPPORTED
Need: commit history / design doc / teammate-visible decision record
PLANNED → actual evidence
Need: /build or /contributor

12. Claims to downgrade or remove

默认主动寻找:

  • 过度精确数字;
  • 无法证明的 ownership;
  • team result 冒领;
  • status 夸大;
  • 无依据规模词;
  • AI buzzword 堆叠;
  • 跟目标岗位无关、占简历空间的弱 claim。

/proof 不应为了“让简历更强”保留一个无法防守的句子。


13. Default output

A. Evidence Summary

哪些经历证据最强、哪些最弱。

B. Claim–Evidence Ledger

ClaimEvidenceOwnershipConfidenceRiskSafe wording

C. Safe Claims

可以现在直接写。

D. Aggressive-but-defensible Claims

可以使用,但必须明确边界。

E. Claims to Downgrade / Remove

说明为什么。

F. Missing Evidence Checklist

按 evidence ROI 排序,不要只说“再丰富项目”。


14. Handoff

按 Cross-skill Handoff Contract 传递:

goal:
claims:
  - claim:
    evidence:
    ownership:
    confidence:
    safe_wording:
    strong_wording:
    interview_risks:
missing_evidence:
open_questions:
recommended_next_skill: /position | /interview | /build | /contributor
privacy_boundary:

规则:下游 Skill 可以改写表达,但不能静默升级 confidence、ownership 或完成状态。

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.