GitHub Issue and PR Templates for Agentic SDLC

Published 10 May 2026

ai-engineering github software-delivery governance

Agentic SDLC does not start with a new orchestration framework. It starts with sharper work packets.

The issue and pull request become the boundary objects between human intent, agent execution, reviewer judgment, and operational evidence. If those objects are vague, the workflow becomes vague. If they are precise, even a relatively simple agent loop becomes more useful.

Issue template

Use this shape for work that an agent may touch.

## Intent

What outcome should this issue produce?

## Acceptance Criteria

- [ ]
- [ ]

## Non-goals

What should not change?

## Affected Components

- repo:
- service:
- system:
- environment:

## Risk Class

low / medium / high

Why?

## Constraints and Invariants

-
-

## Dependencies / Blockers

- Blocks:
- Blocked by:
- Related:

## Testing Expectations

- Unit:
- Integration:
- Contract:
- Manual:

## Observability / Operations

- Metrics:
- Logs:
- Tracing:
- Dashboards:
- Alerts:

## Rollback Plan

How would this be reverted or disabled?

## Agent Instructions

- May the agent create commits? yes/no
- May the agent open or update PRs? yes/no
- Must ask before changing:
  - CI
  - Terraform
  - IAM
  - authentication
  - data migrations
  - public APIs

This is intentionally more specific than a normal feature issue. The point is not bureaucracy. The point is to prevent the agent from guessing the risk model.

PR template

The pull request should record what the agent did and how a human can check it.

## Resolves

Closes #

## Summary

Short description of the change.

## Implementation Notes

What changed and why?

## Out of Scope

What was intentionally not changed?

## Test Evidence

Commands run:

```bash
# paste commands and summarized results
```

CI:

- [ ] CI passing
- [ ] Relevant checks linked

## Risk / Rollback

Risk class:

Rollback plan:

## Operational Notes

Metrics/logs/tracing/dashboards affected:

## Agent Notes

Assumptions:

Uncertainties:

Follow-up issues:

Why templates matter

Templates do not make agents intelligent. They make work legible. That is the precondition for delegation.

The best template is not the longest one. It is the one that forces the right questions before execution and makes the PR reviewable after execution. If a field is never used, remove it. If a class of incident keeps recurring, add the missing field.

The template is part of the platform.