feat: complete core 5 elements and risk layer architecture

This commit is contained in:
Antigravity Agent
2026-06-06 21:11:16 +02:00
commit 96f7643f8a
29 changed files with 12336 additions and 0 deletions

85
GITHUB_WORKFLOW.md Normal file
View File

@@ -0,0 +1,85 @@
# Antigravity Automated Issue-Resolution Workflow
This guide details the automated pipeline for listening to, processing, resolving, and verifying GitHub issues using the Antigravity agentic system.
---
## 1. Workflow Architecture
```mermaid
sequenceDiagram
participant GH as GitHub Webhook / Action
participant OR as Orchestrator Agent
participant IP as Implementation Planner Agent
participant DEV as Developer Agent
participant CI as CI Build & Browser Verifier
GH->>OR: Webhook: Issue Labeled ('agent-resolve')
Note over OR: Parses issue details &<br/>checks config
OR->>OR: Git Branch Checkout (feature/issue-N)
OR->>IP: Spawns Planner Agent with prompt
IP->>IP: Researches codebase & generates plan
IP-->>OR: Returns plan
OR->>DEV: Invokes Developer Agent to write code
DEV->>DEV: Modifies files & implements feature
DEV-->>OR: Code complete
OR->>CI: Runs verification (npm run build & test)
CI-->>OR: Compilation / Test Success
OR->>GH: Git Commit & Push. Auto-creates Pull Request
Note over GH: GitHub Action checks pass.<br/>Human reviews & merges PR.
```
---
## 2. Webhook Payload Specifications
When an issue is labeled (e.g. `agent-resolve`, `enhancement`, `bug`), GitHub triggers a webhook with the following JSON structure sent to the Orchestrator endpoint:
```json
{
"action": "labeled",
"issue": {
"number": 12,
"title": "IMPLEMENT RISK MANAGEMENT UPGRADE: #ISSUE-005 - Kelly Allocation",
"body": "Detailed objective description...",
"state": "open",
"labels": [
{
"name": "agent-resolve"
}
]
},
"repository": {
"full_name": "jannr/investment-sandbox",
"html_url": "https://github.com/jannr/investment-sandbox"
},
"sender": {
"login": "jannr"
}
}
```
---
## 3. Step-by-Step Execution Lifecycle
### Step 3.1: Parsing and Routing
The Orchestrator Agent parses the webhook payload, maps labels using `github-agent-config.json`, and triggers the workflow.
### Step 3.2: Branching Strategy
The system automatically creates a branch:
- For enhancements: `feature/issue-{issue_number}`
- For bugs: `bugfix/issue-{issue_number}`
### Step 3.3: Planning & Coding
1. The Orchestrator launches an **Implementation Planner Agent** to search the codebase and write an `implementation_plan.md` draft.
2. Once the plan is approved (autonomously or by a maintainer), the **Developer Agent** modifies the target code files.
### Step 3.4: Automated Verification
Before pushing changes, the Orchestrator runs:
- `npm run build` to verify Next.js/TypeScript compilations.
- Virtual browser tests (Playwright/Puppeteer stubs) to simulate UI button interactions, state modifications, and graph renderings.
### Step 3.5: PR Creation and Merge
1. The agent pushes the branch to remote and creates a PR using the GitHub API:
`POST /repos/{owner}/{repo}/pulls`
2. Once merged by a human, the branch is deleted and the issue is closed.