File: workshop/side-quest-17-07-repo-poisoning.md
Overall Score: 4.97 / 10.0 (corpus mean: 6.14)
Flagged Dimensions:
| Dimension |
Score |
Benchmark |
Delta |
| active_learning |
2.7 |
≥ 3 activity density (scaled ×3.3, capped 10) |
-7.3 |
| cognitive_load |
5.9 |
≤ 800 words, ≤ 15 new concepts |
-4.1 (concept_overload) |
Root Cause (≤ 2 sentences):
The page introduces 23 new concepts (permissions scoping, toolsets, safe-outputs, create-pull-request, allowed-files, protected-files, policy: request_review, network.allowed-domains, plus the attack-scenario vocabulary) in 1,224 words with only 4 code blocks and 6 checklist items, giving an activity density of 0.82 — well below the 3.0 benchmark that would let learners practice each new term as it's introduced. Most of the page is expository "why this matters" prose and reference tables rather than hands-on practice, so learners read about many defenses in a row without applying each one individually.
Evidence (quoted from the file):
Repository poisoning is what happens when a misdirected agent with write access commits changes an attacker designed — not changes the workflow author intended.
Declare read-only permissions
Route writes through a pull request
Restrict which paths can change
Limit network destinations
Learning Science Rationale:
Mayer's multimedia and active-processing principles hold that learners retain material better when each new concept is immediately followed by a generative activity (practice, application, or retrieval) rather than a long uninterrupted sequence of exposition. Four distinct defensive techniques are presented back-to-back as prose/table content before a single combined exercise appears; this violates the "coherence" and "generative processing" guidance and increases extraneous cognitive load (Sweller), since learners must hold all four techniques in working memory before they get to apply any of them.
Improvement Prompt (for an agent):
Edit workshop/side-quest-17-07-repo-poisoning.md to raise its activity density above the
3.0 benchmark. After each of the four "How AW Defends Against It" subsections
(Declare read-only permissions, Route writes through a pull request, Restrict which paths
can change, Limit network destinations), add a short inline micro-exercise: a 1-2 line
"Quick check" fenced code block or fill-in-the-blank frontmatter snippet that asks the
learner to apply just that one defense before moving to the next subsection. Keep each
micro-exercise under 3 lines so total word count does not grow substantially. Also trim
the "The Attack" and "Why This Matters" narrative sections by roughly 150-200 words,
consolidating overlapping explanation, to reduce new_concepts and word_count pressure on
the cognitive_load score. Preserve the existing :white_check_mark: Checkpoint section and
its checklist items unchanged. Run npx --yes markdownlint-cli2 "**/*.md" after editing to
confirm no lint regressions.
Expected Score After Fix: 6.6 / 10.0
Generated by 🔬 Curriculum Quality Evaluator · copilot · auto · 96.3 AIC · ⌖ 12.2 AIC · ⊞ 9.5K · ◷
File:
workshop/side-quest-17-07-repo-poisoning.mdOverall Score:
4.97 / 10.0(corpus mean:6.14)Flagged Dimensions:
Root Cause (≤ 2 sentences):
The page introduces 23 new concepts (permissions scoping,
toolsets,safe-outputs,create-pull-request,allowed-files,protected-files,policy: request_review,network.allowed-domains, plus the attack-scenario vocabulary) in 1,224 words with only 4 code blocks and 6 checklist items, giving an activity density of 0.82 — well below the 3.0 benchmark that would let learners practice each new term as it's introduced. Most of the page is expository "why this matters" prose and reference tables rather than hands-on practice, so learners read about many defenses in a row without applying each one individually.Evidence (quoted from the file):
Learning Science Rationale:
Mayer's multimedia and active-processing principles hold that learners retain material better when each new concept is immediately followed by a generative activity (practice, application, or retrieval) rather than a long uninterrupted sequence of exposition. Four distinct defensive techniques are presented back-to-back as prose/table content before a single combined exercise appears; this violates the "coherence" and "generative processing" guidance and increases extraneous cognitive load (Sweller), since learners must hold all four techniques in working memory before they get to apply any of them.
Improvement Prompt (for an agent):
Expected Score After Fix:
6.6 / 10.0