Tutorials & How-Tos
Install Agency Agents in Copilot: Full Guide
Step-by-step guide to installing agency-agents in GitHub Copilot, including workarounds for Copilot's single-instructions limitation.
Step-by-step guide to installing agency-agents in GitHub Copilot, including workarounds for Copilot's single-instructions limitation.
GitHub Copilot is everywhere. If you work in a shop that uses Copilot by default, you might feel locked out of the agents ecosystem — Copilot supports only one custom instructions file per repository. But there's a workaround, and once you know it, you can still get most of the benefit of the agency-agents library.
This guide walks through installing agents in Copilot, handling the single-file limitation, and approximating multi-agent workflows.
.github/copilot-instructions.md as its single instruction file per repoStart with a local copy of agency-agents:
git clone https://github.com/msitarzewski/agency-agents.git ~/agency-agents
In your project root:
mkdir -p .github
Copilot reads .github/copilot-instructions.md by default. No other configuration needed.
Because Copilot only reads one file, you need to decide which agent is "default." Pick the one that matches the majority of your work. For most developers, that's Frontend Developer or Backend Developer from the 26 engineering agents roundup.
cp ~/agency-agents/engineering/frontend-developer.md .github/copilot-instructions.md
Commit this file. Your team will inherit the same agent configuration.
Open a file and start typing. Copilot's suggestions should now reflect the agent's priorities — accessibility considerations, modern patterns, performance awareness. If you're not sure whether it's working, ask Copilot Chat a question that a specialist would answer well and compare the response to baseline.
Copilot only loads one file, but nothing stops you from swapping files based on the task. Here's a shell script that does exactly that:
#!/bin/bash
# ~/bin/copilot-agent
AGENT_NAME=$1
AGENT_FILE=~/agency-agents/**/${AGENT_NAME}.md
cp $AGENT_FILE .github/copilot-instructions.md
echo "Loaded agent: $AGENT_NAME"
Usage:
copilot-agent frontend-developer
# ... work on frontend ...
copilot-agent security-engineer
# ... do security review ...
Restart Copilot Chat after each swap for the new instructions to take effect.
Another option: concatenate multiple agents into a single instruction file with clear section headers. Copilot reads all of them but you guide which to use by mentioning the agent name in your chat:
# Copilot Multi-Agent Instructions
## Default Persona
Act as a senior engineer...
## When I say "frontend mode"
(paste Frontend Developer agent content)
## When I say "security review"
(paste Security Engineer agent content)
## When I say "database review"
(paste Database Engineer agent content)
This works reasonably well for 2-4 agents. Beyond that, context gets crowded and the agents blur together.
GitHub is rolling out Copilot Workspace, which supports more flexible instruction hierarchies including per-folder rules. If you have access, it's significantly better for multi-agent workflows. The setup is similar to Cursor's rules system — see Install Agency Agents in Cursor for the pattern.
Pitfall 1: Forgetting to commit. The instructions file should be committed so your teammates get the same agent. Otherwise you get inconsistent behavior across the team.
Pitfall 2: Over-editing the agent. It's tempting to rewrite the agent to match your preferences. Resist. The agents are carefully tuned, and heavy edits often hurt performance.
Pitfall 3: Long instruction files. Copilot has a context budget. Files over 10KB start degrading suggestion quality. Keep it focused.
Pitfall 4: Confusing Copilot with Cursor patterns. Cursor has globs and scoped rules. Copilot doesn't. Don't try to port Cursor configs directly.
Pitfall 5: Expecting full agent fidelity. Copilot doesn't follow instructions as precisely as Claude Code. The agent's voice and framework carry through, but some nuances get lost.
Honestly, Copilot with agents is a step down from Claude Code or Cursor with agents. The suggestions improve meaningfully, but the chat experience feels less "in character" than on other platforms. If you have a choice of tools, pick one of the other two for agent workflows.
That said, if your company is standardized on Copilot for policy or billing reasons, you're not stuck with baseline. A good agent still improves quality, and the concatenation or swap approach gets you most of the benefit.
Claude Code's .claude/agents/ directory supports multiple agents natively and lets you invoke them by name without swapping files. It's simply a better fit for multi-agent workflows. For the comparison, see How to Install Agents Across 8 AI Platforms.
Yes. Copilot Business allows organization-level policies including standardized instructions. If your org is rolling out agents, this is the way.
No native support. Use the swap-based workaround or migrate to a platform with better agent orchestration support.
Both, but the effect is stronger in chat. Inline completions are shaped more by the local context than by instructions.
Yes. The .github/copilot-instructions.md file works across all Copilot-supported editors.
Copilot Enterprise supports more instruction options including custom model routing. Check your admin panel for available features.
Copilot wasn't designed for multi-agent workflows, but with a little creativity, you can approximate them. Start with a single primary agent, use the swap script for occasional specialist tasks, and upgrade to Workspace or Claude Code when your team is ready.
Browse all 150 agents at aiskill.market/agents or submit your own skill.
AI soul and personality tracing skill — gives your AI agent a consistent persona, tone, and conversational identity across all interactions. 319.2K installs.
Official Microsoft bundle: Azure AI, deploy, hosted Copilot SDK, observability, and cloud migration. 600K+ combined installs.
Design short-term, long-term, and graph-based memory architectures for agents
Official Agent Skills suite from ElevenLabs: text-to-speech, speech-to-text, voice agents, sound effects, music, and voice isolation. Works in Claude Code, Cursor, and other compatible agents.