[curriculum-eval] side-quest-17-07-repo-poisoning.md: cognitive_load — four security mechanisms introduced back-to-back with no interleaved checks
维护者通常 1 天内回复
还没有人认领这个 Issue。
评估
- 难度
- 2/5
- 预计耗时
- 1-3 小时
- 新手友好度
- 75/100
- Issue 类型
- 文档
- 描述清晰度
- 描述清楚
- 活跃度
- 活跃
- 技术栈
- javascript
调研方向
阅读 workshop/side-quest-17-07-repo-poisoning.md，重点关注“How AW Defends Against It”和现有的两个练习。在四个防御措施小节之后分别添加一个简短的可折叠自测，保持 capstone 练习不变，并避免重复比较 toolsets。四个自测全部添加完成，且指定的 markdownlint 命令通过后，即为完成。
由索引模型根据 Issue 内容生成。
描述
File: workshop/side-quest-17-07-repo-poisoning.md
Overall Score: 4.95 / 10.0 (corpus mean: 6.09)
Flagged Dimensions:
| Dimension | Score | Benchmark | Delta |
|---|---|---|---|
| cognitive_load | 5.8 |
≤800 words, ≤15 new concepts → 10.0 |
-4.2 (1,234 words, 23 new concepts) |
| active_learning | 2.7 |
activity density ≥ 3 → 10.0 |
-7.3 |
Root Cause (≤ 2 sentences):
The page introduces 23 distinct bolded/code terms (contents: read, contents: write, safe-outputs, allowed-files, protected-files.exclude, protected-files.policy, network.allowed-domains, toolsets: [everything], etc.) across four stacked defence sections before the learner ever applies any of them, and its two <details> exercises are the only interactive elements in 1,234 words of prose (activity_density 0.81). This front-loads an unusually high concept count for a single side quest relative to the ≤15 ideal.
Evidence (quoted from the file):
gh-aw gives you three layers to prevent repository poisoning.
Declare [read-only permissions]... ### Route writes through a pull request ... ### Restrict which paths can change ... ### Limit network destinations
Learning Science Rationale:
Sweller's Cognitive Load Theory predicts that introducing four related-but-distinct security mechanisms (permissions, safe-outputs, protected-files, network allow-lists) back-to-back as pure exposition — each with its own YAML snippet and terminology — exceeds working-memory capacity before any single mechanism is rehearsed, increasing extraneous load and reducing transfer to the "Harden Your Workflow" exercise that follows. Spacing retrieval practice between each new concept (rather than clustering all four, then testing all four at once) is a well-established mitigation for this kind of sequential concept pile-up.
Improvement Prompt (for an agent):
Open workshop/side-quest-17-07-repo-poisoning.md. The "How AW Defends Against It" section
currently presents four defence mechanisms (contents: read, safe-outputs: create-pull-request,
protected-files, network.allowed-domains) back-to-back with no interleaved check-for-
understanding. Do the following:
1. After each of the four ### subsections under "How AW Defends Against It" (Declare read-only
permissions / Route writes through a pull request / Restrict which paths can change / Limit
network destinations), insert one short self-check question in a collapsed <details> block,
e.g. after "Declare read-only permissions":
<details>
<summary>Quick check: what can the agent NOT do with contents: read?</summary>
It cannot call any GitHub API that creates or modifies repository content, even if a hijacked
prompt instructs it to.
</details>
2. Keep the existing "Spot the Dangerous Frontmatter" and "Harden Your Workflow" exercises
unchanged — they remain the capstone application step after the four mini-checks.
3. Verify the total new-concept count stays bounded by considering whether "toolsets:
[everything]" vs "toolsets: [default]" can be mentioned once (in the exercise) rather than
introduced twice.
4. Re-run `npx --yes markdownlint-cli2 "workshop/side-quest-17-07-repo-poisoning.md"` after
editing.
Expected Score After Fix: 6.3 / 10.0 (active_learning and cognitive_load improvements; checkpoint_quality already correct via shortcode once the tooling regex fix lands).
Generated by 🔬 Curriculum Quality Evaluator · copilot · auto · 124.3 AIC · ⌖ 8.34 AIC · ⊞ 10.4K · ◷
- expires on Oct 14, 2026, 4:56 AM UTC
- 主要语言
- JavaScript
- 星标
- 53
- 派生
- 35
- 平均合并
- 13 小时 50 分钟
- 30 天内合并 PR
- 14
环境准备
在浏览器里用你自己的 GitHub 账号启动这个项目的开发容器。
- 没有 Dockerfile 或 Docker Compose 文件
- 没有 Pull Request 模板
- 没有贡献指南
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
githubnext/gh-aw-workshop 的其他 Issue
-
[curriculum-eval] 04-github-actions-intro.md: cognitive_load — 28 new concepts in a 5-minute intro未关闭curriculum documentation quality
难度 2/5 1-3 小时 新手友好度 66/100
githubnext/gh-aw-workshop#4603 ·
维护者通常 1 天内回复
-
curriculum documentation quality
难度 2/5 1-3 小时 新手友好度 75/100
githubnext/gh-aw-workshop#4601 ·
维护者通常 1 天内回复
-
curriculum documentation quality
难度 2/5 1-3 小时 新手友好度 72/100
githubnext/gh-aw-workshop#4578 ·
维护者通常 1 天内回复
-
documentation
难度 2/5 1-3 小时 新手友好度 72/100
githubnext/gh-aw-workshop#4577 ·
维护者通常 1 天内回复
-
[guidelines] side-quest-21-01-sub-agent-syntax.md: checklist checkbox count exceeds 10-item limit未关闭documentation guidelines
难度 2/5 1-3 小时 新手友好度 62/100
githubnext/gh-aw-workshop#4409 ·
维护者通常 1 天内回复
查看 githubnext/gh-aw-workshop 的全部 Issue
相似的 Issue
-
area:install bug needs-triage
难度 2/5 1-3 小时 新手友好度 68/100
维护者通常 2 天内回复
-
难度 2/5 1-3 小时 新手友好度 62/100
维护者通常 1 天内回复
-
[BUG]: Prompt compressor crashes with TypeError when a past chat has a non-string response.text可能已有人在做 @ManojDevarajulu 今天认领。 未关闭possible bug
难度 2/5 1-3 小时 新手友好度 82/100
Mintplex-Labs/anything-llm#6737 ·
维护者通常 1 天内回复
-
check:passed feeds:add
难度 2/5 1 小时以内 新手友好度 62/100
-
Remove the dead tree 'loadfail' handler in browser.js, which calls a non-existent tree.isSelected()未关闭Housekeeping
难度 2/5 1-3 小时 新手友好度 72/100
pgadmin-org/pgadmin4#10601 ·
维护者通常 1 天内回复