🛠️ Autonomous Code Repair (`FixService`)¶

Poing AI includes an Autonomous Code Repair Engine that reads review findings, compiler errors, or test failures, uses AI agents to formulate drop-in code replacements, applies patches to disk, and executes project test suites (unittest, gdlint, npm test, dotnet test) in a closed-loop validation cycle.

🏛️ How Autonomous Repair Works¶

flowchart TD
    A["📥 Discover Issues<br/><i>(Review Findings, PR comments, or uncommitted diff)</i>"] --> B["🧠 Context & Guidelines Retrieval<br/><i>(Vector RAG + Engine Guidelines)</i>"]
    B --> C["🤖 AI Agent (Antigravity / Gemini / Ollama)<br/><i>(Formulates drop-in snippet replacements)</i>"]
    C --> D["✍️ Apply Patches to Disk<br/><i>(Exact substring replacement)</i>"]
    D --> E{"🧪 Run Test Suite / Linters<br/><i>(unittest, gdlint, npm test, etc.)</i>"}
    E -- "❌ Tests Fail" --> F["🔁 Self-Healing Retry Loop<br/><i>(Feed error trace back to agent)</i>"]
    F --> C
    E -- "✅ Tests Pass" --> G["🚀 Output / Commit"]
    G --> H["Local: Formatted diff & summary<br/>PR: Commit, push to branch, & auto-resolve threads"]

🚀 Running Locally with the CLI¶

1. Auto-Fix with Google Antigravity Managed Agent (Recommended)¶

Powered by Google's Interactions API (antigravity-preview-05-2026) with remote Linux sandboxing:

# Fix uncommitted bugs in your workspace using Antigravity
poing --local --fix --provider antigravity

2. Fast Auto-Fix with Gemini Flash¶

poing --local --fix --provider gemini --model gemini-3.7-flash

3. 100% Offline Auto-Fix with Ollama¶

poing --local --fix --provider ollama --model deepseek-r1:latest

💬 Triggering Auto-Fix in GitHub Pull Requests¶

When Poing AI flags issues or you want automated fixes on a PR:

Comment /fix or @poing-ai fix on the pull request.
Poing AI:
Fetches unresolved review threads.
Generates and tests the required code repairs.
Commits the fixes under poing-ai[bot].
Pushes directly to the PR branch.
Auto-resolves the review threads.

🎯 Autonomous Issue Worker (Opt-in)¶

Poing AI can listen to issues, solve them, and open ready-to-merge Pull Requests. This feature is disabled by default to give repository maintainers full control.

How to Trigger Issue Auto-Fix¶

On-demand via Comment (Default): Comment /fix or /work on any open issue.
Label Opt-in: Add the auto-fix or poing-work label to the issue.

Automatic for All Issues: Enable in .github/poing.json:

"fix": {
  "enabled": true,
  "auto_work_on_issues": true
}

Issue Worker Lifecycle¶

Context & Target Discovery: Extracts referenced file paths from the issue and queries the Vector RAG retriever to identify relevant source files.
Patching & Test Execution: Generates drop-in snippet fixes and executes the project's test suite to verify the repair.
Branch & Pull Request Creation: Creates branch fix/issue-<id>-<slug>, commits the tested changes, pushes to origin, and opens a Pull Request linking to Closes #<id>.
Issue Notification: Posts a confirmation comment on the issue with a link to the generated PR.

⚙️ Test Runner Auto-Detection¶

The fixer automatically detects and executes your repository's test runner to verify fixes before finalizing:

Project Type	Detected Test / Lint Command
Python	`python3 -m unittest discover tests`
Godot Engine	`gdlint .`
Node.js / Web	`npm test`
Custom	Configured via `test_command` in `.github/poing.json`

If tests fail during a fix attempt, Poing AI feeds the failure stack trace back into the agent for up to 3 iterative repair attempts.

🛠️ Autonomous Code Repair (FixService)¶