RIA-TV++ Methodology Pipeline: A 7-Stage Agent Workflow Explained

The RIA-TV++ pipeline is a seven-stage agent-oriented workflow that transforms raw textual sources into actionable skill artefacts through progressive extraction, verification, structuring, and testing.

The RIA-TV++ methodology, implemented in the kangarooking/cangjie-skill repository, provides a systematic approach to converting unstructured text—such as books, transcripts, and podcasts—into machine-readable skill files. This end-to-end pipeline enforces six strict invariants to ensure that the resulting artefacts are both agent-ready and human-usable.

Stage 0: Structural Analysis (整书理解 – Adler)

The workflow begins with Stage 0, which applies Adler’s four-step reading methodology to produce a structural, interpretive, and critical analysis of the entire source work. This phase generates a high-level overview that guides subsequent extraction.

According to methodology/01-stage0-adler.md, this stage outputs books/<slug>/BOOK_OVERVIEW.md, establishing the conceptual foundation before any chunking occurs.

Stage 1: Parallel Extraction (并行提取)

Stage 1 deploys five sub-agents running in parallel to segment the source into candidate units. Each unit is extracted with a minimal set of mandatory fields and stored as raw candidates.

As documented in methodology/02-stage1-parallel-extract.md, these extracts populate the candidates/ folder, serving as the initial pool of raw material for the pipeline.

Stage 1.5: Triple Verification

Between extraction and structuring lies Stage 1.5, a rigorous quality gate implementing three independent verification checks:

• V1 (Cross-domain check): Validates applicability across different contexts
• V2 (Predictive power): Assesses the unit’s ability to forecast outcomes
• V3 (Exclusivity): Ensures the unit is distinct and non-redundant

Only candidates passing all three checks proceed, with verification records saved as verified.md per unit (see methodology/03-stage1.5-triple-verify.md).

Stage 2: RIA++ Construction (RIA++ 构造)

Stage 2 transforms verified units into the six-dimensional RIA++ schema. Each dimension serves a specific purpose in making knowledge actionable:

Dimension	Field	Purpose
R	Reading	Original quote or excerpt
I	Interpretation	Re-interpretation in accessible terms
A1	Past Application	Historical case or example
A2	Future Trigger	Activation condition for the skill
E	Execution	Concrete, actionable steps
B	Boundary	Limitations and scope constraints

This stage produces individual SKILL.md files as defined in methodology/04-stage2-ria-plus.md.

# Skill: 《xxx》中的时间管理框架

## R — Reading (原文)

> “时间管理的核心在于把大块任务拆解为可执行的微步骤……”

## I — Interpretation (自述)

将作者关于“把大块任务拆解为微步骤”的论述，重新表述为：先拆分任务，再分配时间块。

## A1 — Past Application (书中案例)

作者在第 3 章举例说明如何在项目启动阶段使用此框架来制定里程碑。

## A2 — Future Trigger ★

当用户在日历中看到“未完成的任务”时，触发该技能，提示立即拆分为子任务。

## E — Execution (可执行步骤)

1. 在待办列表中新建任务。  
2. 将任务拆分为不超过 30 分钟的子任务。  
3. 为每个子任务分配具体时间块。

## B — Boundary (边界)

不适用于需要持续关注的大型宏观目标，例如“提升个人品牌”。此类目标应交由更高层的规划技能处理。

Stage 3: Zettelkasten Linking (链接 – Zettelkasten)

Stage 3 atomizes the skills and establishes bidirectional links between related units, creating a navigable knowledge graph. This process generates an INDEX.md file that serves as the entry point for the skill set.

The atomization and linking rules are specified in methodology/05-stage3-zettelkasten.md, enabling non-linear exploration of the extracted knowledge.

Stage 4: Pressure Testing (压力测试)

Stage 4 subjects the skills to automated validation. The system generates test-prompts.json containing blind test scenarios, executes them against the skills, and marks pass/fail results.

If a skill fails testing, it returns to previous stages for revision. The testing protocol and prompt format are detailed in methodology/06-stage4-pressure-test.md.

{
  "skill_id": "time-management-framework",
  "prompt": "User wants to schedule a new project. How should they apply the skill?",
  "expected_output": [
    "Create a task entry",
    "Break it into ≤30‑minute subtasks",
    "Assign each subtask to a specific time block"
  ]
}

Stage 5: Delivery (交付)

The final Stage 5 packages the validated skills into a reader-friendly DIGEST.md and installs the artefacts into the host skills/ directory. This makes the knowledge immediately available for consumption by downstream agents such as darwin-skill.

The delivery specifications reside in methodology/07-stage5-deliver.md.

Pipeline Invariants

The RIA-TV++ methodology enforces six invariants (documented in methodology/00-overview.md) that govern every stage:

• Atomicity: Each skill represents a single, indivisible unit of knowledge
• Traceability: Every artefact links back to its source material
• Verifiability: All units must pass objective quality checks
• Evolvability: Skills can be updated as source material or requirements change
• User Participation: Human oversight remains integral to the process
• Deliverability: Final outputs must be immediately usable by target agents

Summary

• The RIA-TV++ pipeline consists of seven sequential stages (0 through 5, including 1.5) that progressively refine raw text into structured skills.
• Stage 0 creates structural overviews using Adler’s methodology, while Stage 1 extracts parallel candidates and Stage 1.5 applies triple verification.
• Stage 2 structures verified content into the six-field RIA++ schema (R, I, A1, A2, E, B) producing SKILL.md files.
• Stage 3 builds Zettelkasten-style knowledge graphs with INDEX.md, and Stage 4 validates through automated test-prompts.json scenarios.
• Six invariants (atomicity, traceability, verifiability, evolvability, user participation, deliverability) ensure the output is both agent-ready and human-usable according to the kangarooking/cangjie-skill implementation.

Frequently Asked Questions

What is the RIA++ six-field schema used in Stage 2?

The RIA++ schema is a structured format that transforms raw excerpts into actionable skills. It comprises six dimensions: R (Reading) for the original quote, I (Interpretation) for rephrasing, A1 (Past Application) for historical examples, A2 (Future Trigger) for activation conditions, E (Execution) for concrete steps, and B (Boundary) for scope limitations. This structure ensures each skill is immediately applicable and contextually bounded.

How does the triple verification in Stage 1.5 ensure quality?

Stage 1.5 implements three independent checks defined in methodology/03-stage1.5-triple-verify.md: V1 validates cross-domain applicability, V2 assesses predictive power, and V3 confirms exclusivity against existing units. A candidate must pass all three verifications to proceed to Stage 2, effectively filtering out redundant or weakly supported content before resource-intensive structuring occurs.

What files are generated by the end of the RIA-TV++ pipeline?

The pipeline produces several concrete artefacts: BOOK_OVERVIEW.md from Stage 0, candidates/ directories from Stage 1, verified.md files from Stage 1.5, individual SKILL.md files from Stage 2, an INDEX.md knowledge graph from Stage 3, test-prompts.json from Stage 4, and finally DIGEST.md with installed skills in Stage 5.

How does Stage 4 pressure testing work?

Stage 4 generates automated test prompts stored in test-prompts.json that simulate real-world usage scenarios. Each skill undergoes blind testing against these prompts, with results marked pass or fail. Failed skills return to earlier stages for revision, creating a feedback loop that ensures only validated, robust skills reach final delivery.

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