[curriculum-eval] side-quest-17-07-repo-poisoning.md: cognitive_load/active_learning — lowest-scored page in corpus, four defenses crammed into on
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
- 半天
- 新手友好度
- 78/100
- Issue 类型
- 文档
- 描述清晰度
- 描述清楚
- 活跃度
- 活跃
- 技术栈
- github-actions, markdown
调研方向
阅读 workshop/side-quest-17-07-repo-poisoning.md，并审查学习目标、摘要表、攻击叙述和两个练习。按照描述重新组织并缩短页面，同时保留两个练习和三项防御措施。重新运行课程 rubric 脚本，并确认词数、概念数和活动密度朝着所述目标值有所改善。
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 / 10.0, lowest score in the entire corpus)
Flagged Dimensions:
| Dimension | Score | Benchmark | Delta |
|---|---|---|---|
| cognitive_load | 5.8 |
≤800 words, ≤15 concepts (has 1234 words, 23 concepts) | -4.2 |
| active_learning | 2.7 |
density ≥3 (has 0.81) | -6.3 |
Note: this page's checkpoint_quality and scaffolding scores are also suppressed by the regex bug documented in the companion "curriculum_assessment.py" issue (the page has a proper ## :white_check_mark: Checkpoint and ## :clipboard: Before You Start); this finding covers only the two dimensions that are genuinely low regardless of that bug.
Root Cause (≤ 2 sentences):
This "create"-level side quest crams four distinct defensive mechanisms (contents: read, safe-outputs: create-pull-request, allowed-files/protected-files, network.allowed-domains) plus a full attack-scenario narrative and two separate frontmatter-reading exercises into 1,234 words with only 4 code blocks — well above the corpus's ideal word count and concept ceiling, and its activity density (0.81) sits far below the ≥3 target expected for a page whose own learning objectives promise the reader will "Apply the three gh-aw defences."
Evidence (quoted from the file):
By the end of this side quest you will be able to: Explain what repository poisoning is and why agentic workflows are uniquely vulnerable to it. Identify dangerous
permissions:andtoolsets:values in a workflow frontmatter. Apply the three gh-aw defences:contents: read,safe-outputs: create-pull-request, andnetwork.allowed-domains.
Learning Science Rationale:
Bloom's Taxonomy classifies this page at "create" (the highest cognitive level), which is appropriate for its content, but Cognitive Load Theory (Sweller) warns that high-order tasks are especially sensitive to extraneous load — asking a learner to apply three separate defensive mechanisms while simultaneously introducing the attack narrative, a permissions table, and two "spot the danger" exercises in one continuous page risks overwhelming working memory before the learner reaches the hands-on "Harden Your Workflow" exercise, which is the page's actual highest-value activity.
Improvement Prompt (for an agent):
In workshop/side-quest-17-07-repo-poisoning.md, reduce cognitive load without losing security content:
1. Move the "What You Can Do as a Workflow Author" summary table to immediately after the "Learning Objectives" section (near the top) so readers have a compact reference before reading the three long defensive-mechanism subsections — this lets them skim the table first and treat the detailed prose as optional deep-dive rather than required linear reading.
2. Shorten "The Attack" narrative paragraph and the "Why This Matters" paragraph by combining them into a single paragraph — both currently make a similar point (agents reason over untrusted content, so write access is dangerous) using different words.
3. Keep both "Exercise: Spot the Dangerous Frontmatter" and "Exercise: Harden Your Workflow" — these are the strongest active-learning components on the page and should not be cut; if anything, consider adding one more short code-block example between the three defensive-mechanism subsections to raise activity_density above the current 0.81 without adding narrative words.
4. Target a final word count closer to 900-1000 (from 1234) and confirm new_concepts count drops from 23 toward the corpus median of ~12-15 by re-running the rubric script after edits.
Expected Score After Fix: 6.4 / 10.0 (after also applying the checkpoint/scaffolding regex fix from the companion issue, likely closer to 8.1 / 10.0)
Generated by 🔬 Curriculum Quality Evaluator · copilot · auto · 134.5 AIC · ⌖ 17.6 AIC · ⊞ 9K · ◷
- expires on Sep 30, 2026, 11:52 AM UTC
- 主要语言
- JavaScript
- 星标
- 55
- 派生
- 36
- 平均合并
- 2 天 8 小时
- 30 天内合并 PR
- 17
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
-
[ordering-review] workshop order graph: 1 confirmed ordering problem, 3 cycles are false positives未关闭
难度 2/5 1-3 小时 新手友好度 78/100
githubnext/gh-aw-workshop#4943 · 1 条评论 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 72/100
githubnext/gh-aw-workshop#3109 · 11 条评论 ·
维护者通常 1 天内回复
-
accessibility bug
难度 2/5 1-2 天 新手友好度 68/100
githubnext/gh-aw-workshop#2744 · 10 条评论 ·
维护者通常 1 天内回复
-
feedback simulation workshop
githubnext/gh-aw-workshop#4974 ·
维护者通常 1 天内回复
-
[workshop-sim] Repair: Resolve billing-path contradiction and duplicate access check in Step 07/07d未关闭feedback simulation workshop
githubnext/gh-aw-workshop#4973 ·
维护者通常 1 天内回复
查看 githubnext/gh-aw-workshop 的全部 Issue
相似的 Issue
-
难度 2/5 1-3 小时 新手友好度 76/100
bigcapitalhq/bigcapital#1544 · 1 条评论 ·
维护者通常 1 天内回复
-
help wanted kind/bug
难度 2/5 1-3 小时 新手友好度 78/100
meshery/meshery.io#3112 ·
维护者通常 2 天内回复
-
area:docs type:chore
难度 2/5 1-3 小时 新手友好度 82/100
bulkhead-labs/tailsuite#31 ·
-
dependencies javascript
难度 2/5 1-3 小时 新手友好度 70/100
gems-uff/noworkflow#470 ·
维护者通常 3 天内回复
-
product / auth product / console product / sites product / storage
难度 1/5 1 小时以内 新手友好度 84/100
维护者通常 1 天内回复