Quality Red Lines That Prevent Skill Output in cangjie-skill: 5 Critical Blocking Rules

The cangjie-skill pipeline aborts skill generation and moves candidates to the rejected/ folder if they violate any of five hard-coded quality red lines governing verification status, section completeness, quotation length, test coverage, and description specificity.

The kangarooking/cangjie-skill repository maintains strict output standards to ensure every skill consumed by downstream orchestrators like darwin-skill is atomic, testable, and free from false positives. These quality red lines that prevent skill output are defined in the meta-skill definition at SKILL.md and enforced during the Triple Verification and Pressure-Testing stages.

The Five Critical Quality Red Lines

According to the source code in the “质量红线 (Quality Red Lines)” section of SKILL.md (lines 46-53), the pipeline enforces five non-negotiable rules. Violation of any single rule stops the skill from being emitted.

1. Triple Verification Must Pass Completely

All three Triple-Verification checks must be passed before a skill can proceed. As implemented in methodology/03-stage1.5-triple-verify.md, the pipeline validates:

• Cross-domain evidence – substantiation across different contexts
• Predictive power – ability to answer unseen questions
• Uniqueness – distinction from trivial common-sense knowledge

If any check fails, the candidate unit is blocked from output.

2. Mandatory Six-Section Structure

Every skill must contain the six required sections defined in templates/SKILL.md.template:

• R (Reading) – the source evidence
• I (Interpretation) – restatement in new terms
• A1 (Past Application) – historical usage example
• A2 (Future Trigger) – activation conditions
• E (Execution) – step-by-step instructions
• B (Boundary) – explicit limitations

Missing any section triggers immediate rejection.

3. Original Quotation Length Constraints

Original quotation length must be ≤ 150 characters per segment (or ≤ 100 words for English content). This prevents bloated quoting that hinders prompt size limits and ensures concise, focused skills.

4. Required Test Files With Bait Prompts

Every skill must ship a test-prompts.json file containing:

• At least one bait/negative test (a scenario that should not trigger the skill)
• One cross-skill confusion test explicitly referencing a brother skill from the same book

This requirement, detailed in methodology/06-stage4-pressure-test.md and scaffolded by templates/test-prompts.json.template, ensures robustness against false positives and clean integration with the testing harness.

5. Concrete Trigger Descriptions

The description field must spell out a concrete trigger condition. Vague descriptions like “a skill about X” fail this red line. The description must specify exact activation criteria to make the skill discoverable by orchestrating agents.

Code Examples: Violations vs. Compliance

Non-Compliant Skill (Multiple Red Line Violations)

This example fails red lines #2, #3, #4, and #5:

# SKILL.md (invalid)

R: "In the long passage the author says many things about leadership and management..."   
I: Short reinterpretation.
A1: Example from chapter 3.
A2: Trigger when needed.
E: Step-by-step instructions.
B: When not applicable.

description: "A skill about leadership."

Problems identified:

• R exceeds the 150-character limit (violates red line #3)
• description is vague and lacks concrete trigger (violates red line #5)
• No test-prompts.json provided (violates red line #4)
• Missing explicit Triple-Verification evidence markers (violates red line #1)

Compliant Skill (Passes All Red Lines)

This example satisfies all enforcement criteria:

# SKILL.md (valid)

R: "The author notes, 'Align early and often.'"
I: I restate the principle in my own words.
A1: Case: The author applied the framework in XYZ scenario.
A2: Trigger: When planning a product roadmap that requires stakeholder alignment.
E: 1. Identify stakeholders. 2. Map dependencies. 3. Draft timeline.
B: Not suitable for solo projects lacking cross-functional teams.

description: "Triggers when a team needs a structured stakeholder-alignment roadmap."

Accompanying test-prompts.json:

{
  "positive": [
    "How can I align my product team with the executive board?"
  ],
  "negative": [
    "Give me a summary of the book."
  ],
  "cross-skill": [
    "When should I use the 'risk-assessment' skill from the same book?"
  ]
}

All six sections are present, the quotation is concise, the description specifies a clear trigger, and the test file contains the required bait prompt referencing a sibling skill.

Where Red Lines Are Enforced in the Source Code

• SKILL.md – Central specification containing the red-line checklist in the “质量红线 (Quality Red Lines)” section (lines 46-53)
• methodology/03-stage1.5-triple-verify.md – Implements red line #1 (Triple Verification logic)
• methodology/06-stage4-pressure-test.md – Implements red line #4 (test-prompt creation and bait requirements)
• templates/SKILL.md.template – Enforces red line #2 by ensuring sections R/I/A1/A2/E/B are present
• templates/test-prompts.json.template – Scaffolds the required bait prompts to satisfy red line #4

Summary

• Triple Verification must pass completely—cross-domain evidence, predictive power, and uniqueness are mandatory.
• Six-section structure (R/I/A1/A2/E/B) is required for every skill output.
• Quotation limits of ≤150 characters per segment prevent prompt bloat.
• test-prompts.json must include bait tests and cross-skill confusion tests to ensure robustness.
• Concrete descriptions with explicit trigger conditions are required for agent discoverability.
• Violations move candidates to the rejected/ folder with audit reasons, blocking pipeline output.

Frequently Asked Questions

What happens if a skill fails only one quality red line?

The pipeline aborts skill generation for that unit immediately. According to the enforcement logic in SKILL.md, any single violation prevents output, and the candidate is moved to the rejected/ folder with a specific reason for audit.

Why is the cross-skill confusion test required in test-prompts.json?

The cross-skill test, which references a brother skill from the same book, prevents false positive triggers when multiple skills from the same source could activate simultaneously. This ensures clean integration with the darwin-skill testing harness and verifies that the skill activation logic is precisely scoped.

Where is the 150-character quotation limit defined?

The limit is specified in the quality red lines section of SKILL.md (lines 46-53). This constraint keeps skills concise and prevents token limit issues when downstream agents consume multiple skills in a single context window.

Can a skill pass Triple Verification but still be rejected?

Yes. Passing Triple Verification (red line #1) satisfies only one of five requirements. A skill must also meet structural requirements (red line #2), length constraints (red line #3), testing standards (red line #4), and description specificity (red line #5) to be emitted from the pipeline.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}