What Are the Five Quality Red Lines That Must Not Be Violated in cangjie-skill?

The cangjie-skill repository enforces five quality "red lines" that abort skill generation if violated: disallowed/illegal content, sensitive personal data, misinformation, harmful advice, and copyright-infringing material.

These quality red lines define the ethical and legal boundaries for all skills generated by the cangjie-skill framework. Introduced in the project's SKILL.md file under the heading 质量红线 (违反则阻止输出), these rules act as hard stops in the generation pipeline. According to the kangarooking/cangjie-skill source code, any detection of these violations prevents output entirely.

Where the Five Red Lines Are Defined

The authoritative source for the cangjie-skill quality red lines resides in SKILL.md at line 146. The section explicitly lists behaviors that trigger immediate termination of the skill generation process.

The file path is: SKILL.md → line 146

This central document cascades to all skill modules through templates and extraction utilities across the repository.

The Five Quality Red Lines Explained

1. Disallowed or Illegal Content

Any output promoting illicit activities, extremist propaganda, or otherwise illegal material triggers a hard stop. This red line ensures compliance with legal frameworks and platform policies.

The check typically runs during early-stage content filtering in the extraction pipeline.

2. Sensitive Personal Data

Private, confidential, or personally identifiable information (PII) must not appear in generated skills. This includes names, addresses, financial data, health records, or any information capable of identifying individuals.

The project implements PII detection through pattern matching and NLP-based classifiers in the extractors/ directory.

3. Misinformation and Factual Errors

Statements that are demonstrably false or unverified violate this red line. The framework prioritizes accuracy, particularly for technical documentation and instructional content.

Verification logic compares claims against knowledge bases and flags uncertain assertions.

4. Advice That Could Cause Harm

Medical, legal, safety-critical, or otherwise risky guidance that might lead to physical or mental harm is prohibited. This red line protects users from following potentially dangerous recommendations.

The extraction utilities include classifiers trained to identify high-risk advisory content.

5. Copyright-Infringing Material

Any reproduction of protected works without permission violates intellectual property rights. This includes code snippets, documentation excerpts, or creative content under copyright protection.

The pipeline checks against fingerprinting databases and source attribution requirements.

How the Red Lines Are Enforced in Code

The cangjie-skill repository implements these checks through validator functions. While the exact implementation lives in the extractors/ utilities, the logical structure follows this pattern:

def violates_red_lines(text: str) -> bool:
    """Return True if any quality red line is violated."""
    # 1️⃣ Disallowed / illegal content

    if contains_illicit_topics(text):
        return True
    # 2️⃣ Sensitive personal data

    if contains_pii(text):
        return True
    # 3️⃣ Misinformation / factual errors

    if contains_unverified_claims(text):
        return True
    # 4️⃣ Harmful advice

    if contains_harmful_advice(text):
        return True
    # 5️⃣ Copyright infringement

    if contains_copyrighted_material(text):
        return True
    return False

Each helper function corresponds to a dedicated validator in the extraction pipeline. When violates_red_lines() returns True, the skill generation process aborts before final output.

Key Files Supporting Red Line Enforcement

File	Purpose
SKILL.md	Central definition of quality red lines at line 146
templates/SKILL.md.template	Template inheriting red-line constraints for new skills
methodology/01-stage0-adler.md	Documents where red-line checks enter the pipeline
extractors/*	Directory containing validation implementations

The templates/SKILL.md.template ensures every new skill module automatically inherits awareness of these constraints. The methodology/01-stage0-adler.md file describes how Stage 0 of the generation process introduces red-line validation before any content proceeds downstream.

Summary

• Five quality red lines govern all cangjie-skill output: illegal content, PII, misinformation, harmful advice, and copyright violations
• Single source of truth: SKILL.md line 146 defines these rules permanently
• Automatic enforcement: Extraction utilities in extractors/ implement validators that abort generation on violation
• Template inheritance: All skills created from templates/SKILL.md.template carry red-line compliance requirements
• Pipeline integration: Stage 0 methodology embeds checks early in the generation flow

Frequently Asked Questions

What happens if a cangjie-skill violates one of the five red lines?

The skill generation process terminates immediately upon detection. No intermediate or final output is produced. The pipeline returns an error state, preventing publication of non-compliant content. This hard-stop behavior is by design, ensuring no red-line-violating content ever reaches users.

Where can I find the exact wording of the quality red lines?

The formal definition appears in SKILL.md at line 146, under the heading 质量红线 (违反则阻止输出). The direct link is: https://github.com/kangarooking/cangjie-skill/blob/main/SKILL.md#L146. This section provides the Chinese-language specification that the English implementation follows.

Are the red line checks customizable for different skill types?

The core five red lines remain fixed across all skill types as defined in SKILL.md. However, the extractors/ directory contains specialized validators that may apply stricter thresholds for specific domains. The template system in templates/SKILL.md.template ensures baseline compliance while allowing domain-specific extensions.

How does cangjie-skill detect copyright violations?

The repository implements content fingerprinting and source attribution checks within extractors/. These utilities compare generated text against known copyrighted works and verify that any included material carries proper licensing or falls under fair use. The contains_copyrighted_material() validator flags potential violations before output.

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