File: workshop/side-quest-17-06-output-injection.md
Overall Score: 7.85 / 10.0 (corpus mean: 8.63)
Flagged Dimensions:
| Dimension |
Score |
Benchmark |
Delta |
| Scaffolding |
5.0 |
10.0 |
-5.0 |
| Readability |
7.9 |
10.0 |
-2.1 |
Root Cause:
The file has no "Before You Start" or "Prerequisites" section, causing scaffolding score 5.0. The Flesch-Kincaid Grade Level is 12.6 — above the 8–12 target range — driven by long, complex sentences in the security-explanation prose sections. Both dimensions are confirmed genuine deficiencies after reviewing the actual file.
Evidence (quoted from the file):
"If an attacker can control even a small part of what appears in that surface, they can influence downstream human decisions — code reviews, security approvals, or release sign-offs — without ever touching the workflow code itself."
This sentence (Grade Level 14+) exemplifies the readability issue. The file opens directly with "The Attack" without establishing what prior knowledge is expected.
Learning Science Rationale:
Krashen's Input Hypothesis posits that learners acquire new knowledge most efficiently when input is slightly above current comprehension level (i+1), not several levels above it. A Grade Level 12.6 security analysis presented without explicit prerequisite framing forces learners to bridge multiple comprehension gaps simultaneously, increasing extraneous cognitive load and reducing retention of the security concepts themselves.
Improvement Prompt (for an agent):
File: workshop/side-quest-17-06-output-injection.md
Current score: 7.85/10 — scaffolding 5.0 (no prereqs), readability 7.9 (FK grade 12.6)
Improve as follows:
1. Add a "## Before You Start" section after the intro quote:
- You have completed [Step 9: Reading Workflow Output](09-understand-output.md) and understand safe-output records.
- You have completed [Side Quest: Prompt Injection](side-quest-17-03-prompt-injection.md) or are familiar with the concept of prompt injection.
2. Reduce Flesch-Kincaid Grade Level from 12.6 to <= 12.0 by:
a. Breaking the longest sentences (>30 words) into two sentences.
b. Replacing multi-clause constructions like "X, Y, and Z without ever..." with "X and Y. This means an attacker never needs to...".
c. Target 3–4 sentence rewrites in "Why This Matters for Agentic Workflows" and "How AW Defends Against It".
3. Do NOT change the technical accuracy of the security content — only restructure sentence length.
4. Verify the ## Checkpoint has >= 4 verifiable items.
Run markdownlint after editing.
Expected Score After Fix: 8.4 / 10.0
Generated by 🔬 Curriculum Quality Evaluator · 113.6 AIC · ⌖ 8.56 AIC · ⊞ 6.3K · ◷
File:
workshop/side-quest-17-06-output-injection.mdOverall Score:
7.85 / 10.0(corpus mean:8.63)Flagged Dimensions:
Root Cause:
The file has no "Before You Start" or "Prerequisites" section, causing scaffolding score 5.0. The Flesch-Kincaid Grade Level is 12.6 — above the 8–12 target range — driven by long, complex sentences in the security-explanation prose sections. Both dimensions are confirmed genuine deficiencies after reviewing the actual file.
Evidence (quoted from the file):
This sentence (Grade Level 14+) exemplifies the readability issue. The file opens directly with "The Attack" without establishing what prior knowledge is expected.
Learning Science Rationale:
Krashen's Input Hypothesis posits that learners acquire new knowledge most efficiently when input is slightly above current comprehension level (i+1), not several levels above it. A Grade Level 12.6 security analysis presented without explicit prerequisite framing forces learners to bridge multiple comprehension gaps simultaneously, increasing extraneous cognitive load and reducing retention of the security concepts themselves.
Improvement Prompt (for an agent):
Expected Score After Fix:
8.4 / 10.0