File: workshop/05-agentic-workflows-intro.md
Overall Score: 7.91 / 10.0 (corpus mean: 8.74)
Flagged Dimensions:
| Dimension |
Score |
Benchmark |
Delta |
| Cognitive Load |
2.7 / 10 |
≤ 800 words, ≤ 15 concepts |
−7.3 |
| Style Compliance |
8.5 / 10 |
≤ 3 callout blocks |
−1.5 |
Root Cause (≤ 2 sentences):
At 1702 words and 26 new concepts this step is more than twice the recommended cognitive load budget — the scoring formula docks 1 point per 100 words over 800 (−9 pts) and 1 point per 2 concepts over 15 (−5.5 pts), driving the dimension to 2.7. The introductory step attempts to explain triggers, trust models, hybrid workflows, and agent reasoning loops all in one pass, creating an extraneous load spike at exactly the moment learners are orienting.
Evidence (quoted from the file):
"Classic Actions handles deterministic CI/CD. Agentic workflows fill the gap for tasks that need judgment — or you can mix both in a single hybrid workflow."
(This sentence arrives mid-page after the learner has already encountered: task brief, frontmatter, lock.yml, agent loop, tool calls, sandbox, firewall, trigger, permission, runner, AIC, natural-language schedule, and three classification tasks.)
Learning Science Rationale:
Sweller's Cognitive Load Theory identifies the introductory step of a skill sequence as the highest-risk moment for germane load overflow: learners have no prior schema to anchor new concepts, so each additional term competes for the same limited working-memory slots. Presenting 26 new concepts in a single 1702-word step creates an extraneous load spike that research consistently links to reduced transfer and increased dropout.
Improvement Prompt (for an agent):
Edit workshop/05-agentic-workflows-intro.md to reduce cognitive load to ≤ 900 words and ≤ 16 concepts.
1. Move the "Why not just use a standard Actions workflow?" <details> block content into a new side quest (e.g., side-quest-05-01-actions-vs-agentic.md) and replace it with a one-sentence teaser linking to that side quest.
2. Trim the "Three things to know" section to bullet points only — remove the surrounding prose paragraphs that repeat information already in the bullets.
3. Remove or relocate the trust model paragraph ("If you already trust GitHub Actions...") — it belongs in a security side quest, not the intro.
4. Shorten the two classification tasks to a single task with one example — move the second task to a side quest.
5. Target: ≤ 900 words, ≤ 16 new concepts. Run `npx --yes markdownlint-cli2 "**/*.md"` before committing.
Expected Score After Fix: 8.6 / 10.0
Generated by 🔬 Curriculum Quality Evaluator · 123.5 AIC · ⌖ 9.72 AIC · ⊞ 6.4K · ◷
File:
workshop/05-agentic-workflows-intro.mdOverall Score:
7.91 / 10.0(corpus mean:8.74)Flagged Dimensions:
Root Cause (≤ 2 sentences):
At 1702 words and 26 new concepts this step is more than twice the recommended cognitive load budget — the scoring formula docks 1 point per 100 words over 800 (−9 pts) and 1 point per 2 concepts over 15 (−5.5 pts), driving the dimension to 2.7. The introductory step attempts to explain triggers, trust models, hybrid workflows, and agent reasoning loops all in one pass, creating an extraneous load spike at exactly the moment learners are orienting.
Evidence (quoted from the file):
(This sentence arrives mid-page after the learner has already encountered: task brief, frontmatter, lock.yml, agent loop, tool calls, sandbox, firewall, trigger, permission, runner, AIC, natural-language schedule, and three classification tasks.)
Learning Science Rationale:
Sweller's Cognitive Load Theory identifies the introductory step of a skill sequence as the highest-risk moment for germane load overflow: learners have no prior schema to anchor new concepts, so each additional term competes for the same limited working-memory slots. Presenting 26 new concepts in a single 1702-word step creates an extraneous load spike that research consistently links to reduced transfer and increased dropout.
Improvement Prompt (for an agent):
Expected Score After Fix:
8.6 / 10.0