RIA++ Skill Format Explained: Reading‑Interpretation‑Application‑Boundary for AI Agents

The RIA++ skill format is a six-segment structure that transforms verified knowledge into machine-executable skills combining human-readable narrative (R, I, A1, A2) with agent-ready contracts (E, B).

The RIA++ format is the core architecture used by the kangarooking/cangjie-skill repository to convert extracted knowledge from books into functional, dispatchable capabilities for AI agents. It extends the classic RIA (Reading → Interpretation → Application) framework with Execution and Boundary dimensions to create skills that both humans understand and machines execute accurately.

Understanding the Six Dimensions of RIA++

Each skill in the cangjie-skill project must populate six mandatory sections, documented in methodology/04-stage2-ria-plus.md.

R — Reading (原文)

The direct quotation from the source material. This preserves the original insight verbatim, establishing provenance and preventing drift.

## R — Reading

> "When you're stuck, try to think of the exact opposite of your current approach."
> — *The Art of Thinking Differently*, p. 42

I — Interpretation (自述)

The author rephrases the quotation in their own words. This step verifies comprehension and creates a bridge between source language and operational context.

A1 — Past Application (书中案例)

Concrete examples already demonstrated in the source material. These ground the skill in proven use cases rather than theoretical possibility.

A2 — Future Trigger ★ (未来触发场景)

The most critical component of the entire format. This section defines the exact scenario that should activate the skill:

• When: The user's expressed situation
• Then: The agent's proposed action
• Because: The underlying rationale

The A2 summary (≤300 characters) populates the description front-matter that agents like Claude read to determine skill activation.

E — Execution (可执行步骤)

Step-by-step instructions the agent follows once triggered. These are machine-executable procedures, not suggestions:

## E — Execution

1. Present the reverse-thinking quote.
2. Ask the user to state their current approach.
3. Prompt the user to describe the exact opposite.
4. Collect the opposite ideas and suggest a concrete next step.

B — Boundary (边界与盲点)

Explicit constraints that prevent inappropriate skill invocation. This includes:

• Known limitations
• Blind spots where the skill fails
• Edge cases that should suppress activation

## B — Boundary

- Do **not** use this skill for purely factual queries (e.g., "What is the capital of France?").
- If the user's problem is not creative or design-related, the trigger should be ignored.

How RIA++ Fits Into the Skill Pipeline

The cangjie-skill project implements RIA++ through a three-stage workflow:

1. Stage 1 — Verification: Raw knowledge extracts from source books into verified.md

2. Stage 2 — RIA++ Construction: Each verified unit maps to the six-section format defined in methodology/04-stage2-ria-plus.md

3. Stage 3+ — Zettelkasten Linking: Skills interconnect; A2 and B refine through testing

The completed skill renders through templates/SKILL.md.template into a standalone SKILL.md file with this exact structure—enforced by the template itself.

Key Files Defining the RIA++ Format

File	Purpose
methodology/04-stage2-ria-plus.md	Complete specification of all six segments with authoring guidelines
templates/SKILL.md.template	Enforces R/I/A1/A2/E/B structure for every generated skill
SKILL.md	Example rendered skill showing populated template
methodology/00-overview.md	Visual table mapping all pipeline stages to RIA++ columns
README.en.md	High-level project documentation introducing RIA++ as the "Construction" phase

Why RIA++ Matters for AI Agent Systems

The RIA++ format solves a critical problem in agent architecture: the gap between human knowledge representation and machine actionability.

• R + I + A1 + A2 create interpretability—humans can audit why a skill exists and when it should fire
• E + B create reliability—agents execute deterministically while respecting guardrails

Without the Boundary dimension, agents over-generalize. Without Execution, agents hallucinate implementation. The A2 trigger—marked with ★ in the methodology—prevents the "always-on" skill problem that degrades agent performance.

Summary

• RIA++ stands for Reading, Interpretation, Application (Past), Application (Future Trigger), Execution, and Boundary—six mandatory sections per skill
• The format lives in methodology/04-stage2-ria-plus.md and enforces through templates/SKILL.md.template
• A2 (Future Trigger) is the most critical section: it defines activation conditions and populates the description front-matter
• E (Execution) provides step-by-step agent instructions; B (Boundary) defines when to suppress the skill
• This structure makes skills simultaneously human-auditable and machine-executable

Frequently Asked Questions

What does the "++" in RIA++ stand for?

The "++" represents the two additions to classic RIA: E (Execution) and B (Boundary). Classic RIA covers knowledge extraction and human understanding; the plus components make that knowledge runnable by agents with appropriate guardrails.

Why is A2 (Future Trigger) marked as the most critical component?

According to the cangjie-skill methodology, A2 determines whether a skill fires appropriately. Poorly defined triggers cause either missed opportunities (false negatives) or inappropriate interruptions (false positives). The A2 summary populates the description field that agents read for activation decisions.

How does the Boundary section prevent agent errors?

The B section explicitly lists failure modes, edge cases, and inappropriate contexts. This acts as a negative filter: even when the A2 trigger matches superficially, Boundary conditions can suppress execution if the situation falls outside the skill's valid operating envelope.

Can RIA++ skills be generated automatically?

Yes. The repository provides templates/SKILL.md.template to standardize structure. Stage 2 of the pipeline consumes verified.md outputs and prompts authors to complete each RIA++ section, producing consistent SKILL.md files ready for agent consumption.

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