Possible improvements
We keep exploring ways to improve the template. The ideas below cover security, planning, testing, agent isolation, and support for larger teams.
This is not a committed roadmap. A new feature has to add clear value without making the template harder to use for a single developer.
Partially available means related functionality already exists. Under consideration means the idea was discussed and is not built.
Security and approvals
| Improvement | What it would add | Status |
|---|---|---|
| GitHub approval rules | Code owners and repository rules, so that protected files and important changes need the right approvals | Partially available the scripts classify a change; GitHub does not enforce the approvals yet. Issue 99 |
| Separate agent identities | GitHub identities with limited permissions for coding agents, in place of the owner's permissions | Under consideration Issue 99 |
| More thorough reviews for risky changes | Additional checks for changes to security, infrastructure, or important architecture decisions | Partially available risk classification exists; specialized reviews do not |
Planning and task execution
| Improvement | What it would add | Status |
|---|---|---|
| Feature clarification | Focused questions when a feature issue lacks important details | Under consideration Issue 16 |
| Dependency-aware planning | Dependencies between tasks and features in a form that automation can use | Partially available the roadmap describes dependencies in words |
| Ready-to-run task queue | A list of the tasks whose dependencies are complete and that an agent can start | Under consideration |
| Parallel execution | Several agents working on separate features without conflicts | Under consideration separate worktrees already allow concurrent work that you start yourself |
| Better failure recovery | Diagnosing common failures and recovering without redoing finished work | Partially available retries and resuming are supported |
Reviews, testing, and quality
| Improvement | What it would add | Status |
|---|---|---|
| Two independent reviewers | One review of the behavior and one of the engineering quality | Under consideration Issue 66 |
| Faster, more selective CI | Running the tests a change affects, not everything every time | Under consideration a successful verification is already reused for unchanged content |
| Requirements-to-test tracking | Which implementation and which tests provide the evidence for each requirement | Under consideration |
| Browser-based validation | Browser tests with screenshots or traces for user-interface changes | Under consideration a project can configure its own browser tests |
| Specialized quality checks | Security, performance, API, or infrastructure checks where they are relevant | Under consideration |
Agent environments and infrastructure
| Improvement | What it would add | Status |
|---|---|---|
| Full environment isolation | For each agent its own working files, database, cache, services, queues, ports, and credentials. It has three parts:
| Under consideration only the working files are separated today |
| Remote execution | Agent runs that continue on a server when the developer's laptop is closed | Under consideration |
Skills and model selection
| Improvement | What it would add | Status |
|---|---|---|
| Reusable skills registry | Selecting, versioning, and managing specialist agent skills per project and role | Under consideration |
| Dynamic model selection | Choosing an AI model by what a task needs and what it costs | Under consideration a model is already configured per role and can be set per run |
| Workflow improvement proposals | Suggested updates to tests, prompts, or skills, drawn from failures that recur | Partially available analyze-runs.sh proposes improvements when you start it (learning from your runs); an analysis on a schedule is not built yet (Issue 131) |
| Run analytics | Duration, retries, model usage, and cost of runs | Partially available every run records its duration, its retries, and the tokens per role and model, and report-runs.sh reports on them; a cost in money is not calculated |
Developer experience and larger teams
| Improvement | What it would add | Status |
|---|---|---|
| Web dashboard | Starting, monitoring, and resuming agent tasks in a browser | Under consideration |
| Expand the documentation | Guides for the planning process, for unattended runs, and for approvals and the merge gate, beside the pages that exist | Partially available this website covers the workflow, getting started, the research, and the history |
| Convert an application into a template | Reusable configuration and modules extracted from an existing application | Under consideration |
| Event-driven maintenance | Controlled maintenance tasks that start after a CI failure, a security alert, or a dependency update | Under consideration |
| Support for multiple repositories | Work coordinated across projects, with shared scheduling and resource limits | Under consideration |
To see where these ideas came from, read Research and design choices. To see what the template supports today, read The workflow.