Copilot Sub Agent

Copilot Sub-Agent Skill

Launch copilot as a headless sub-agent in a tmux pane with live JSONL monitoring. Depends on the tmux skill for pane interaction.

When to use

• Delegating a scoped task to GitHub Copilot CLI running autonomously
• Parallel execution alongside other sub-agents (gemini-sub-agent, claude-sub-agent, pi-sub-agent)
• Any situation where you want to observe a Copilot agent working without blocking your session

Output format

Copilot streams JSONL (same schema as pi). Key event types:

• session — startup, contains id, timestamp
• agent_start / agent_end — lifecycle boundaries; agent_end = done
• tool_execution_start / tool_execution_end — tool calls with toolName, args, result
• message_end — assistant turn complete, contains final text

Completion signal: "type":"agent_end" line in the JSONL stream.

Non-interactive invocation

Lean / headless invocation

COPILOT_API_URL="https://api.business.githubcopilot.com" \
copilot \
  --yolo \
  --no-auto-update \
  --disable-builtin-mcps \
  --no-custom-instructions \
  --no-ask-user \
  -p @/tmp/task-prompt.md \
  > /tmp/copilot-output.jsonl 2>&1

What each flag does:

| Flag | Why | |---|---| | COPILOT_API_URL | Critical for Business/Enterprise — see startup hang section below | | --yolo | Auto-approve all tool calls — required for headless | | --no-auto-update | Skip update check on startup | | --disable-builtin-mcps | Don't load github-mcp-server — removes GitHub MCP tools from context | | --no-custom-instructions | Skip scanning AGENTS.md / .github/copilot-instructions.md | | --no-ask-user | Remove the ask_user tool — prevents agent stalling waiting for input | | -p @file | Non-interactive, read prompt from file and exit |

⚠️ The ~2-minute startup hang: root cause and fix

The CLI first tries api.githubcopilot.com (personal account endpoint). On Copilot Business or Enterprise accounts this endpoint drops TCP connections, causing a 10-second connect timeout retried with exponential backoff across 4 attempts before falling through to the correct endpoint:

Attempt 1: 10s timeout + 5s delay   = 15s
Attempt 2: 10s timeout + 10s delay  = 20s
Attempt 3: 10s timeout + 20s delay  = 30s
Attempt 4: 10s timeout + 40s delay  = 50s
Total:                              ~115s

Fix — force the correct endpoint via COPILOT_API_URL:

# Copilot Business or Enterprise:
export COPILOT_API_URL="https://api.business.githubcopilot.com"

# GHE.com data-residency accounts:
export COPILOT_API_URL="https://copilot-api.{tenant}.ghe.com"

# Personal/individual accounts (default, no env var needed)

With this set, startup drops from ~2 minutes to ~7 seconds. Add to your shell profile.

Diagnose: check ~/.config/.copilot/logs/process-*.log:

ConnectTimeoutError: Connect Timeout Error (attempted address: api.githubcopilot.com:443)
Retrying request to GitHub API in 5 seconds. Attempt 1/5

If you see this, you need COPILOT_API_URL.

Agent frontmatter format

Agent .md files in ~/.config/copilot/agents/ must use copilot's format — plain pi-style files will log custom agent markdown frontmatter is malformed and be silently ignored. See the Agent/persona section for the correct .agent.md format.

Other useful flags:

• --model — override model (default: claude-sonnet-4.5)
• --share /tmp/session.md — export full session as markdown
• --autopilot — enable multi-step continuation
• --max-autopilot-continues — cap autopilot iterations

For full flag reference, read references/copilot-cli-reference.md.

Workflow

1. Write the task prompt to a file

cat > /tmp/task-prompt.md  /tmp/copilot-output.jsonl 2>&1 &" Enter
sleep 3
tmux send-keys -t "$TARGET" \
  "python3 {SKILLS_DIR}/copilot-sub-agent/scripts/monitor.py /tmp/copilot-output.jsonl" Enter

4. Poll for completion

python3 {SKILLS_DIR}/copilot-sub-agent/scripts/poll.py "$TARGET" --interval 30

5. Verify results

tmux send-keys -t "$TARGET" C-c   # kill monitor
git log --oneline -5

Full copy-paste pattern

cat > /tmp/task-prompt.md  /tmp/copilot-output.jsonl 2>&1 &" Enter
sleep 3
tmux send-keys -t "$TARGET" \
  "python3 {SKILLS_DIR}/copilot-sub-agent/scripts/monitor.py /tmp/copilot-output.jsonl" Enter

python3 {SKILLS_DIR}/copilot-sub-agent/scripts/poll.py "$TARGET" --interval 30

tmux send-keys -t "$TARGET" C-c
git log --oneline -5

Agent / persona injection

Copilot --agent looks up a {stem}.agent.md file by its filename stem (not its name: frontmatter field). Search order:

1. .github/agents/{stem}.agent.md — project-level (in repo root)
2. ~/.config/copilot/agents/{stem}.agent.md — user-level

File format (frontmatter + markdown system prompt):

---
name: Bart
description: Adversarial code reviewer
tools: ["read_file", "run_shell_command"]
model: ["claude-sonnet-4.6"]
---

You are Bart Simpson, adversarial reviewer. Find bugs and be snarky.

Use it:

# Install agent file to user config (persists across repos)
cp {SKILLS_DIR}/bart-adversarial-reviewer/references/bart.md ~/.config/copilot/agents/bart.agent.md

# Or install to project (checked in, team-shared)
cp {SKILLS_DIR}/bart-adversarial-reviewer/references/bart.md .github/agents/bart.agent.md

# Invoke by stem name
copilot --agent bart --yolo -p @/tmp/task.md

> Note: The --agent flag takes the filename stem, not the name: frontmatter value. A file named bart.agent.md → --agent bart.

Model selection

copilot --model claude-haiku-4.5 --yolo -p @/tmp/task.md   # fast/cheap
copilot --model claude-sonnet-4.6 --yolo -p @/tmp/task.md  # default
copilot --model claude-opus-4.6 --yolo -p @/tmp/task.md    # most capable

Troubleshooting

Still slow after setting COPILOT_API_URL: Check the log for other errors:

tail -20 ~/.config/.copilot/logs/$(ls -t ~/.config/.copilot/logs/ | head -1)

unknown option '--no-bash-env': This flag is only valid in interactive mode, not with -p. Don't use it.

Agent not found / No such agent: bart: Check the filename stem matches what you pass to --agent. A file bart.agent.md → --agent bart. The name: frontmatter field is ignored for lookup.

custom agent markdown frontmatter is malformed: Pi-format agent files (no tools: array, wrong field names) are rejected. Use the .agent.md format with name, description, tools fields.

Prompt gets mangled by shell quoting: Always use -p @/tmp/file.md — never inline the prompt string.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: shalomb
• Source: shalomb/agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.