Sling — Autonomous Development Pipeline

The STIV Pipeline

Every issue goes through a structured, auditable pipeline before a PR is created.

Spec

Refines the issue into a testable specification with acceptance criteria

Test

Generates a test suite (TDD red phase). Tests fail initially and drive implementation

Implement

Iterative code generation to pass tests. Up to 3 attempts per file with error feedback

Validate

Full test suite, code review, quality gates. PR created with evidence chain

Dual Execution Modes

Choose the right mode for the task. Both produce tested, reviewed PRs.

STIV Mode

Structured TDD pipeline

+ Auditable per-phase progression
+ Tests written before implementation
+ Phase checkpointing (resume on failure)
+ Best for well-scoped, single-file changes
+ API-billed (pay per token via OpenRouter)

Recommended

Claude Code Mode

Headless Claude Code CLI

+ Full codebase exploration
+ Handles complex multi-file changes
+ Installs dependencies, runs commands
+ Best for refactors and large features
+ Flat-rate (Max subscription, no API cost)

Everything Built In

From governance to observability — Sling handles the full autonomous development lifecycle.

🛡

Governance Gates

Risk-tiered approval system. Low-risk changes auto-approve. High-risk changes require human approval via Slack or dashboard.

🌙

Night Shift

Autonomous batch processing overnight. Scans repos, processes eligible issues, stops at budget cap or dawn.

📊

Pipeline Dashboard

Real-time visibility into every task. Step timing, test results, review scores, PR links — all in one place.

🔀

PR Automation

Auto-merge safe Dependabot bumps. Detect duplicate PRs with AI. Flag stale PRs. Daily maintenance scan.

💬

AIA Slack Bot

Ask for status, approve tasks, diagnose issues, and create tickets — all from Slack. Thin intent router, not another AI agent.

🏥

HIPAA Compliance

Repos with HIPAA mode get elevated risk tiers, mandatory human approval, and PHI-adjacent change detection.

🧪

E2E Testing

Playwright E2E runs on staging after deployment. Videos, traces, screenshots, and AI analysis of failures.

🧠

Model Routing

Multi-model AI with champion/challenger scoring. Circuit breakers, spend limits, and weekly eval harness.

📋

Multi-Tenant

Organization-scoped projects, fair-scheduled task queues, per-org concurrency limits, and API token management.

Risk-Based Governance

Every task is classified by risk before autonomous execution begins.

Tier 0

Auto-Approved

Docs, tests, config, READMEs. Zero risk — executes immediately.

Tier 1

Low Risk

Single-file code changes. Auto-approved, PR is the governance gate.

Tier 2

Medium Risk

Shared code, complex logic. Held for human approval via Slack.

Tier 3

High Risk

Security, DB migrations, payment logic. Rejected — requires manual implementation.

Get Started in 3 Steps

From GitHub issue to merged PR — without touching a keyboard.

      # 1. Write a clear GitHub issue in a supported repo

      gh issue create --title "fix: login fails with + in email" \

        --body "POST /api/auth/login returns 500 for user+test@example.com..."

      # 2. Label it for Sling

      gh issue edit 123 --add-label "auto-implement"   # STIV mode

      gh issue edit 123 --add-label "claude-code"     # Claude Code mode

      # 3. Or use the Claude Code slash command to evaluate + label

      /sling-prep owner/repo#123

      → Sling picks it up, runs the pipeline, and opens a PR.

      → Monitor at the dashboard or ask @AIA in Slack.

Issues In. PRs Out.