AIRD v1.2: The Actionable Layer (L5) & Evolutionary Loop
This document extends the AIRD v1.1 specification to support Agentic Workflows and Self-Evolving Documentation.
1. The L5 Execution Protocol Template
Whenever a document describes a process that an AI Agent should execute, add an ## Execution Protocol section.
Example: Git Branch Cleanup Protocol
## Execution Protocol
- **Tools**: `git`, `shell`
- **Logic Chain**:
1. `git branch --merged main` $\rightarrow$ Identify merged branches.
2. Filter out `main`, `develop`, and protected branches using the `protected-branches.json` list.
3. For each identified branch: `git branch -d <branch-name>`.
- **Success Criteria**:
- All merged feature branches are deleted.
- No protected branches are touched.
- Final count of deleted branches is reported.
- **Fallback**:
- If `git branch -d` fails (non-merged), attempt `git branch -D` ONLY after confirming with the user.
2. The Evolutionary Loop (L4 $\rightarrow$ PR)
In AIRD v1.2, .ai-feedback.md is the trigger for a Documentation Patch.
The Automated Flow:
- Detection: Agent detects a documentation gap while executing an L5 protocol.
- Logging: Agent appends a structured entry to
.ai-feedback.md: ```markdown- Gap ID: GAP-2026-001
- Target:
docs/api.md - Issue: Missing timeout value for
/v1/uploadendpoint. - Proposed Fix: Add “Timeout: 30s” to the
Requestsection. ```
- Target:
- Gap ID: GAP-2026-001
- Action: The Agent (or a dedicated maintenance Agent) creates a new Git branch
ai-patch/GAP-2026-001, applies the fix, and opens a PR. - Human Review: Human merges the PR $\rightarrow$
.ai-feedback.mditem is marked asResolved.
3. Ontology Mapping (The Semantic Layer)
To prevent hallucinations in complex domains, include an ontology.json in the root.
Example ontology.json
{
"domain": "ProjectX-Cloud",
"mappings": {
"EdgeNode": "The physical compute unit located at the customer site, running k3s.",
"ControlPlane": "The central management API hosted in AWS us-east-1.",
"Soothe": "The proprietary internal synchronization protocol used for state delta updates."
}
}
Agent Instruction: Always consult ontology.json when encountering undefined technical terms in L3/L4 docs.