AI‐Team‐Workflow

AI Team Workflow

AxiomID uses a zero-cost AI engineering team built on GitHub Actions + Gemini free tier. No paid Copilot plan required.

Overview

New Issue/PR
     |
     v
[ai-issue-triage.yml]  [ai-pr-health.yml]
     |                        |
     v                        v
Labels + Priority       Size Labels + Security Detection
     |                        |
     v                        v
[gemini-review.yml]    [ci-intelligence.yml]
     |                        |
     v                        v
AI Code Review         CI Anomaly Detection
     |
     v
[Human Review Required]
CODEOWNERS + Branch Protection
     |
     v
[Merge]

The 5 Workflows

1. ai-issue-triage.yml

• Trigger: issues: opened
• Model: Gemini 2.0 Flash (free tier)
• Actions:
  • Classifies issue as bug, feature, docs, security, perf
  • Assigns priority:high/medium/low
  • Tags area: auth, wallet, ui, api, infra
  • Adds welcome comment with next steps

2. ai-pr-health.yml

• Trigger: pull_request: opened, synchronize
• Model: Gemini 2.0 Flash (free tier)
• Actions:
  • Applies size labels: size:XS (<50) size:S (50-200) size:M (200-500) size:L (500-1000) size:XL (1000+)
  • Detects security-sensitive files by path: auth/, crypto/, pi-sdk/, soul/, secret, token, key, password
  • Adds needs-human-approval + priority:high for security paths
  • Adds ai-generated label when opened by bot

3. gemini-review.yml

• Trigger: pull_request: opened, synchronize
• Model: Gemini 2.0 Flash (free tier)
• Actions:
  • Full code review on PR diff
  • Posts review comment on PR
  • Informational only — does not block merge
  • Status check: Gemini AI Review

4. ci-intelligence.yml

• Trigger: workflow_run: completed
• Actions:
  • Analyzes CI failures
  • Detects anomalies in PR patterns
  • Vectors CI data for trend analysis
  • Posts summary comment on failed PRs

5. loops.yml

• Trigger: cron: 0 8 * * * (daily at 8am UTC)
• Actions:
  • Generates daily changelog
  • Marks stale issues/PRs
  • Sends Telegram notification with daily summary

Human-in-the-Loop

AI agents operate within strict boundaries:

Action	AI Can	Human Required
Open PR	Yes	No
Add Labels	Yes	No
Review Code	Yes (informational)	No
Merge PR	No	Yes (always)
Modify workflows	No	Yes
Approve security PR	No	Yes

Labels Used

Label	Applied By	Meaning
size:XS/S/M/L/XL	ai-pr-health.yml	PR size
priority:high/medium/low	ai-issue-triage.yml	Priority
needs-human-approval	ai-pr-health.yml	Security-sensitive
ai-generated	ai-pr-health.yml	Bot-opened PR
needs-rebase	Manual	Conflicts with main

Cost

• GitHub Actions: Free (public repo = unlimited minutes)
• Gemini API: Free tier (1500 requests/day, 15 RPM)
• Total: $0/month