GitHub adds custom agents to Copilot CLI workflows
GitHub is pushing Copilot CLI from an interactive prompt box toward version-controlled automation, and the new custom agents feature is the mechanism. Announced June 9, 2026 on the GitHub Blog, custom agents let teams define an agent's behavior, allowed tools, standards, and outputs in a Markdown profile file (ending in .agent.md) that lives in the repository alongside code. The practical shift for engineers is that prompt-based workflows become reviewable artifacts: they can be diffed, code-reviewed, and reused across machines instead of being re-typed ad hoc. GitHub also said remote control of Copilot sessions is now generally available on github.com and GitHub Mobile, and that it is piloting an experimental general-purpose accessibility agent.
Why this matters for engineering teams
The interesting part of this release is not that Copilot can run in the terminal - it already could - but that agent behavior now becomes a versioned file in the repository. That moves prompt engineering out of individual chat sessions and into the same review, diff, and CI machinery teams already use for code, which is the real lever for making AI-assisted automation repeatable and auditable.
What GitHub announced
In a June 9, 2026 post on the GitHub Blog, GitHub introduced custom agents for Copilot CLI. A custom agent is defined in a Markdown profile file - GitHub uses the .agent.md extension, for example accessibility.agent.md - that describes how the agent should behave, which tools it may use, what standards to follow, and what outputs to produce. The same post says remote control for Copilot sessions is now generally available on github.com and GitHub Mobile, and that GitHub is piloting an experimental general-purpose accessibility agent.
The practitioner trade-off
File-defined agents let a team encode repository, tooling, and process context once so the same behavior runs wherever the agent is invoked, reducing the need to re-explain context in ad hoc prompts. The cost is that the agent definition becomes another artifact to maintain: as the codebase and tooling change, stale agent profiles can silently produce wrong or low-quality output, so teams will want validation steps and review for agent files just as they have for CI configuration.
What to watch
Watch whether GitHub ships reusable templates, an agent marketplace, or CI patterns for validating agent outputs, and whether third-party tooling standardizes on the .agent.md format - that would signal file-defined agents becoming a portable convention rather than a Copilot-only feature.
Key Points
- 1GitHub added custom agents to Copilot CLI, defined in repository .agent.md Markdown files that specify behavior, tools, standards, and outputs.
- 2File-defined agents turn one-off terminal prompts into version-controlled, code-reviewable workflows that run consistently across machines and team members.
- 3Reproducibility improves, but teams take on new maintenance: agent definitions and their outputs now need validation and review like any other code.
Scoring Rationale
This is a notable tooling update for developer productivity, making prompt-driven terminal work more reproducible. It is not a frontier-model release, but it materially affects workflows for engineers who use the CLI and Copilot.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems