agentic-fleet is the centralized, polyglot agent orchestration engine and source of truth for the Autonomous 5-Agent SDLC.
By adding one workflow file, any GitHub repository (Python, TypeScript/Node, Go, Rust, Java, Monorepo) instantly inherits an autonomous engineering department: product framing, architectural design gates, TDD code implementation, adversarial QA, multi-tenant security auditing, and human-in-the-loop sign-off.
[Issue Opened / @fleet]
│
▼
1. 📋 PM-Agent ──► Clarifies requirements & writes Gherkin User Stories
│
▼
2. 🏛️ Architect-Gate ──► Audits design against ADRs before code is written
│
▼
3. 🧑💻 Dev-Agent ──► Implements code, tests & auto-installs dependencies
│
▼
4. 🛡️ Security-Gate ──► Audits SQLi, multi-tenant isolation & secrets
│
▼
5. 🧪 QA-Gate ──► Executes test suite & verifies edge cases
│
▼
6. 🧙♂️ Senior Reviewer ──► Signs off with final APPROVE & ready-for-merge
| Subagent | Persona & Focus | Trigger Event | Output Contract |
|---|---|---|---|
🎯 pm-agent |
Product Strategy & Framing Frames user stories, Gherkin Acceptance Criteria ( Given/When/Then), and RICE prioritization scores. |
issues.opened, @pm-agent |
Structured Issue Spec + label agent:ready-for-design |
📐 architect-agent |
System Architecture & ADRs Audits proposed technical designs against Architecture Decision Records before code is written. |
agent:design-review, @architect-agent |
Design Approval Document + label agent:ready-for-dev |
🧑💻 dev-agent |
TDD Full-Stack Engineer Creates isolated branch ( feat/<issue-id>-<slug>), authors clean code + unit tests, auto-installs manifests, materializes files, and handles self-healing loops. |
agent:ready-for-dev, @dev-agent, PR review feedback |
Materialized source files, Pull Request, and detailed Remediation Summary |
🛡️ security-agent |
Security & Compliance Auditor Audits git diffs for multi-tenant isolation leaks ( tenant_id filters), hardcoded secrets, and OWASP vulnerabilities. |
pull_request.opened, pull_request.synchronize, @security-agent |
Security Audit Report (STATUS: PASSED ✅ or STATUS: BLOCKED 🛑) |
🧪 qa-agent |
Adversarial QA & Test Automation Executes live test suites in the local sandbox ( pytest, npm test, go test, cargo test), tests edge cases, and verifies zero test collection errors. |
ready-for-qa, @qa-agent |
QA Verification Report (100% PASS ✅ or STATUS: FAILED ❌) |
🧙♂️ senior-reviewer-agent |
Principal Reviewer & Gatekeeper Audits diffs against ADRs, validates Sec + QA green checks, submits formal approval ( LGTM ✅), and applies ready-for-merge. |
Sec + QA passed, @senior-reviewer-agent |
Architectural Review (APPROVED LGTM ✅) + ready-for-merge label |
- Auto-detects project manifests:
-
Python:
pyproject.toml,requirements.txt$\rightarrow$ pytest -v -
TypeScript / Node:
package.json,tsconfig.json$\rightarrow$ npm test/vitest/jest -
Go:
go.mod$\rightarrow$ go test ./... -
Rust:
Cargo.toml$\rightarrow$ cargo test -
Taskfile / Makefile:
Taskfile.yml(task test),Makefile(make test)
-
Python:
- Generates language-specific package markers (e.g.,
__init__.pyfor Python only, avoiding polluting Go/TypeScript repos).
- Pluggable LLM execution layer supporting:
- Google Gemini:
gemini-2.0-flash,gemini-2.5-pro(GEMINI_API_KEY) - OpenAI:
gpt-4o,gpt-4o-mini,o3-mini(OPENAI_API_KEY) - Anthropic:
claude-3-5-sonnet,claude-3-7-sonnet(ANTHROPIC_API_KEY) - Local Ollama / DeepSeek:
deepseek-r1:latest(OLLAMA_HOST)
- Google Gemini:
- Automatic failover across models and providers during API rate limits.
SymbolIndexerparses Python AST and TypeScript/Go/Rust regex definitions.- Builds a compact
< 1,000 tokenoutline of classes, methods, endpoints, and types. - Scales effortlessly to large 100k+ line codebases without token bloat or context truncation.
- Deterministic state engine (
StateManager) tracking PR execution history, remediation iterations, and hard budget caps (preventing runaway CI loops). - Automatic token usage calculation and cost estimation ($ USD) posted directly on PRs.
- Optimistic concurrency with exponential fetch-rebase retry loops on git push conflicts.
- Path traversal security guards preventing LLM-generated code from writing outside the workspace.
- Token and secret redaction in all CI logs.
To connect agentic-fleet to any repository:
In Settings
- Add
GEMINI_API_KEY(orOPENAI_API_KEY/ANTHROPIC_API_KEY).
In Settings
- Under Workflow permissions, select "Read and write permissions" and check "Allow GitHub Actions to create and approve pull requests".
Create .github/workflows/agentic-sdlc.yml in your target repository:
name: Autonomous Agentic SDLC
on:
issues:
types: [opened, labeled]
issue_comment:
types: [created, edited]
pull_request:
types: [opened, synchronize, labeled]
pull_request_review:
types: [submitted, edited]
pull_request_review_comment:
types: [created, edited]
workflow_dispatch:
permissions:
contents: write
pull-requests: write
issues: write
jobs:
agentic-orchestrator:
name: "Autonomous SDLC Fleet"
runs-on: ubuntu-latest
steps:
- name: Checkout Target Repository
uses: actions/checkout@v4
with:
fetch-depth: 0
- name: Run Agentic Fleet Action
uses: dipeshsingh2012/agentic-fleet@v1
with:
gemini-api-key: ${{ secrets.GEMINI_API_KEY }}
# Or OpenAI:
# openai-api-key: ${{ secrets.OPENAI_API_KEY }}
# Or Anthropic Claude:
# anthropic-api-key: ${{ secrets.ANTHROPIC_API_KEY }}
github-token: ${{ secrets.GITHUB_TOKEN }}# Clone the orchestration engine
git clone https://github.com/dipeshsingh2012/agentic-fleet.git
cd agentic-fleet
# Install dependencies
pip install -r requirements.txt
# Run full test suite (75+ unit & E2E tests in < 2 seconds)
pytest -v