claude-compat — Claude Code Compatibility for Copilot CLI

A Copilot CLI plugin + companion wrapper script that lets you use Claude Code CLI syntax with GitHub Copilot CLI, without shadowing your real claude binary.

What's Included

Component	Type	Purpose
plugin.json	Plugin manifest	Registers this as a Copilot CLI plugin
skills/claude-compat/SKILL.md	Plugin skill	/claude-help — full mapping reference inside Copilot sessions
hooks.json	Plugin hook	Shows a reminder on session start
cpc	Companion script	Translates Claude Code CLI flags → Copilot CLI flags
setup-claude-copilot-compat.sh	Setup script (bash)	Symlinks config dirs + installs the plugin
setup-claude-copilot-compat.ps1	Setup script (PowerShell)	Same as above, for Windows/PowerShell
cpc.ps1	PowerShell wrapper	Invokes the cpc Python script from PowerShell
cpc.cmd	CMD wrapper	Invokes the cpc Python script from cmd.exe

Prerequisites

• GitHub Copilot CLI installed (copilot --version)
• Python 3.6+ (ships with macOS; install from python.org on Windows)

Installation

Option A: Plugin only (no cpc wrapper)

If you just want the /claude-help reference skill inside Copilot CLI:

copilot plugin install ./

Option B: Full setup (plugin + wrapper + config sync)

macOS / Linux (bash):

# 1. Make scripts executable
chmod +x cpc setup-claude-copilot-compat.sh

# 2. Put cpc on your PATH (pick one)
cp cpc /usr/local/bin/
# or: ln -s "$(pwd)/cpc" ~/bin/cpc
# or: add this directory to PATH

# 3. Run setup (installs plugin + symlinks config dirs)
./setup-claude-copilot-compat.sh

Windows (PowerShell):

# 1. Put cpc on your PATH (pick one)
Copy-Item cpc, cpc.ps1, cpc.cmd -Destination "$env:USERPROFILE\bin\"
# or: add this directory to your PATH

# 2. Run setup (installs plugin + symlinks config dirs)
#    Note: Run as Administrator if symlink creation requires elevated privileges.
.\setup-claude-copilot-compat.ps1

Option C: Install from a GitHub repo

copilot plugin install abhi-singhs/copilot-cli-claude-code-compat

Usage Examples

# Start interactive session
cpc                                    # → copilot

# Start with a prompt
cpc "explain this project"             # → copilot -i "explain this project"

# Non-interactive mode
cpc -p "summarize README.md"           # → copilot -p "summarize README.md"

# Continue last conversation
cpc -c                                 # → copilot --continue

# Resume a session (-r is the shorthand for --resume in both CLIs)
cpc -r my-session "keep going"         # → copilot --resume=my-session -i "keep going"
cpc --resume my-session                # → copilot --resume=my-session
cpc -r                                 # → copilot --resume (session picker; needs a TTY)

# Auth (copilot login options pass through)
cpc auth login                         # → copilot login
cpc auth login --host https://x.ghe.com  # → copilot login --host https://x.ghe.com
cpc auth login --device-code           # → copilot login --device-code
cpc auth logout                        # → /logout (in interactive session; subcommand removed)

# Update
cpc update                             # → copilot update

# MCP server management
cpc mcp                                # → copilot mcp

# Skip permissions
cpc --dangerously-skip-permissions     # → copilot --allow-all

# Tool permissions (syntax is auto-converted)
cpc --allowedTools "Bash(git log *)" "Bash(npm test)" -p "check history"
# → copilot --allow-tool=shell(git log *) --allow-tool=shell(npm test) -p "check history"

# Read-only Claude Code tools collapse onto Copilot's 'read' permission kind
cpc --allowedTools "Grep" "Glob" -p "find the bug"
# → copilot --allow-tool=read --allow-tool=read -p "find the bug"

# Limit turns
cpc --max-turns 5 -p "fix all lint errors"
# → copilot --max-autopilot-continues=5 -p "fix all lint errors"

# MCP config
cpc --mcp-config ./my-servers.json     # → copilot --additional-mcp-config=@./my-servers.json

# Tool availability
cpc --tools "Bash,Edit,Read" -p "q"    # → copilot --available-tools=bash,edit,view -p "q"

# Delegate to cloud (like Claude Code --cloud "task")
cpc --cloud "Fix the login bug"
# → copilot -i "/delegate Fix the login bug"

# --remote is a deprecated Claude Code alias for --cloud
cpc --remote "Fix the login bug"
# → copilot -i "/delegate Fix the login bug"

# Resume a cloud session locally (like Claude Code --teleport)
cpc --teleport abc123                  # → copilot --resume=abc123
cpc --teleport                         # → copilot --resume (session picker)

# Select model
cpc --model sonnet "fix the bug"       # → copilot --model sonnet -i "fix the bug"
cpc --model gemini-3.6-flash "fix it"  # → copilot --model gemini-3.6-flash -i "fix it"

# Shell sandbox for this session only (Copilot CLI-only, experimental mode)
cpc --sandbox -p "run the test suite"  # → copilot --sandbox -p "run the test suite"
cpc --no-sandbox -p "run the build"    # → copilot --no-sandbox -p "run the build"

# Plan mode
cpc --permission-mode plan             # → copilot --plan

# Code review (Claude Code /review is an alias for /code-review; its effort
# level, --fix, and --comment have no Copilot equivalent and are stripped)
cpc -p "/review high --fix pr#123"     # → copilot -p "/review pr#123"

# Isolated Git worktree (experimental in Copilot CLI)
cpc --worktree feature-auth            # → copilot --worktree=feature-auth
cpc -w                                 # → copilot --worktree (auto-generated name)

Debugging

Use --dry-run to see what copilot command would be executed:

cpc --dry-run --dangerously-skip-permissions --max-turns 3 -p "fix tests"
# Output: copilot --allow-all --max-autopilot-continues=3 -p fix tests

Interactive Slash Commands

Slash commands are in-session only and can't be aliased externally. Type /claude-help inside a Copilot CLI session to see the full mapping table.

Quick reference for the most common ones:

Claude Code	Copilot CLI	Same?
/compact	/compact [FOCUS-INSTRUCTIONS]	✅ Both accept optional focus instructions
/autocompact [auto|<tokens>]	—	❌ Claude Code-only (v2.1.221+, sets the auto-compact threshold; Copilot's /compact only compacts on demand)
/clear	/clear	✅ — note: Claude Code's optional [name] labels the previous conversation in /resume; Copilot's optional [PROMPT] starts the new conversation
/context [all]	/context	✅ Claude Code's optional all expands the per-item breakdown in fullscreen; Copilot's /context takes no arguments
/diff	/diff	✅
/model	/model [--session|--global|--repo|--local] [MODEL] (/models)	✅ Copilot CLI scopes the change: --session (default) is the current session, --repo/--local pin a repository default, --global sets it for future sessions
/plan	/plan	✅
/resume	/resume (/continue)	✅
/review [low|medium|high|xhigh|max|ultra] [--fix] [--comment] [target]	/review [PROMPT]	⚠️ Same name, different arguments — Claude Code's /review is now an alias for /code-review; Copilot's takes a free-form prompt, so cpc strips effort levels, --fix, and --comment
/tasks	/tasks	✅
/agents	/agent (/subagents)	⚠️ Renamed — /agent browses agents; /subagents (/agents) configures per-agent subagent models. Claude Code removed the /agents wizard in v2.1.198
/security-review	/security-review [PROMPT]	✅ Direct match — both run a security review agent over pending changes
/background (/bg)	—	❌ Claude Code-only (detach to background agent; closest: Ctrl+X then b)
/fork [prompt]	/fork [NAME] (experimental)	✅ Aligned in Claude Code v2.1.212+ — both copy the session into a new independent session; the forked-subagent behavior moved to /subtask
/subtask <task>	/fleet <task>	⚠️ Best-effort — Claude Code v2.1.212+ spawns a forked subagent that inherits the conversation; Copilot's /fleet runs parallel subagents (not 1:1)
/branch [name]	/branch [NAME]	✅ Aligned — fork the session into a new one (experimental in Copilot CLI)
/btw [question]	/ask (experimental)	⚠️ Renamed — side question without adding to history; the question argument is optional in Claude Code v2.1.212+
/code-review (/simplify, /review)	/review	⚠️ Renamed — /simplify and /review are now aliases; --fix, --comment, and effort levels have no Copilot equivalent
/cost	/usage	⚠️ Renamed
/permissions	/permissions [default|assisted|allow-all|show] (/permissions reset)	⚠️ Aligned — Copilot's /permissions now switches permission modes (default/assisted/allow-all) in addition to show; /permissions reset clears in-memory tool and path approvals
/allow-all (/yolo)	/allow-all [off|auto|show] (/yolo)	⚠️ Alias for /permissions allow-all — the on option was replaced with auto
/sandbox	/sandbox [config|status|policy|enable|disable]	⚠️ Aligned — both now have /sandbox; Copilot adds config (settings dialog), status, and policy alongside enable/disable
/voice [hold|tap|off]	/voice [on|off|models|devices]	⚠️ Aligned — both have voice mode; Copilot's models and devices browse voice models and pick the microphone, and it has no push-to-talk hold/tap
/deep-research <question>	/research TOPIC	⚠️ Best-effort — Claude Code fans out web searches; Copilot uses GitHub + web sources
/export	/share [link|off|file|html|gist|research] (/export)	⚠️ Renamed — /export is a Copilot alias; bare /share generates a shareable GitHub link (falling back to a Markdown export) and off stops sharing
/config [key=value]	/config [--repo|--local] [show KEY|KEY|KEY VALUE] (/settings)	✅ Name match — Copilot CLI documents /config as an alias of /settings; only the inline syntax differs (space instead of =)
/extra-usage → /usage-credits	—	❌ Claude Code-only (configure usage credits; closest: /usage for stats only)
/goal	/autopilot [OBJECTIVE] (/goal [OBJECTIVE])	⚠️ Aligned — Copilot's /autopilot [OBJECTIVE] (alias /goal, experimental) starts/refocuses autopilot mode with an optional objective (--max-ai-credits N caps spend); /goal on//goal off toggle without an objective, closing the gap for Claude Code's /goal <condition> / /goal clear
/radio	—	❌ Claude Code-only (Claude FM lo-fi radio)
/remote-control	/remote [on|off]	⚠️ Renamed — no args shows status; on/off toggles
/run, /run-skill-generator, /verify	—	❌ Claude Code-only (build/launch/drive the project's app; v2.1.145+)
/memory	/memory [on|off|show]	✅ Both have it — Copilot CLI added /memory on|off|show in v1.0.49
/scroll-speed	—	❌ Claude Code-only (interactive scroll speed adjustment)
/stop	—	❌ Claude Code-only (stop current background session)
/autofix-pr	—	❌ Not available
/import [codex|gemini] [--dry-run] [--yes]	—	❌ Claude Code-only (v2.1.213+, import Codex or Gemini CLI config into Claude Code)
/list-agents (/peers)	—	❌ Claude Code-only (v2.1.224+, list subagents and sessions Claude can message; /subagents only configures models)
/doctor (/checkup)	—	❌ Claude Code-only (installation health check)
/web-setup	—	❌ Not available
/team-onboarding	—	❌ Not available
/fewer-permission-prompts	—	❌ Not available
/loop (/proactive)	—	❌ Not available
—	/ask QUESTION	🆕 Copilot CLI only (experimental)
—	/env	🆕 Copilot CLI only — show loaded environment details
—	/chronicle [standup|tips|improve|reindex|skills create|skills review|skills status]	🆕 Copilot CLI only (experimental) — session history tools, insights, and repository skill proposals
—	/research TOPIC	🆕 Copilot CLI only
—	/rubber-duck [PROMPT]	🆕 Copilot CLI only — rubber duck agent for a second opinion
—	/update (/upgrade)	🆕 Copilot CLI only
—	/version	🆕 Copilot CLI only
—	/streamer-mode (/on-air)	🆕 Copilot CLI only — hide preview model names and quota details for streaming
—	/after [DELAY PROMPT] / /every [INTERVAL PROMPT]	🆕 Copilot CLI only (experimental) — schedule a one-shot or recurring prompt/skill for this session
—	/app	🆕 Copilot CLI only — launch the GitHub Copilot app (or show its download URL)
—	/extensions (/extension) [manage|mode]	🆕 Copilot CLI only — manage CLI extensions
—	/settings [--repo|--local] [show KEY|KEY|KEY VALUE] (/config)	🆕 Copilot CLI only — open the settings dialog, focus it on a key, set a setting inline, or print a key's value
—	/pr [view|create|fix|auto|automerge]	🆕 Copilot CLI only — manage pull requests for the current branch
—	/limits / /limits set max-ai-credits VALUE	🆕 Copilot CLI only — per-response AI credit limits
—	/move [branch|task]	🆕 Copilot CLI only (experimental) — move uncommitted changes into a new Git worktree and switch to it
—	/refine [TEXT]	🆕 Copilot CLI only — rewrite a rough prompt into a clear one for review before sending (no args cleans up the current input box)
—	/tuikit [colors|icons|select|tabbar]	🆕 Copilot CLI only — preview TUIkit design-system components and color tokens
—	/worktree [branch|task] / /worktree new [PROMPT]	🆕 Copilot CLI only (experimental) — move the conversation to a new Git worktree, or start a new conversation in one; Claude Code's /fork and /branch are partial analogs

Config Sharing

The setup script symlinks these directories so both tools share the same files:

Directory	Shared?
agents/	✅ Symlinked between ~/.claude/ and ~/.copilot/
skills/	✅ Symlinked (Copilot also reads ~/.claude/skills/ natively)
commands/	✅ Copilot reads .claude/commands/ natively
AGENTS.md	✅ Copilot reads natively
Settings (~/.claude/settings.json / ~/.copilot/settings.json)	❌ Different schemas
MCP config	❌ Different formats

Limitations

• Slash commands can't be aliased — use /claude-help for the reference
• System prompts (--system-prompt, --append-system-prompt) don't exist in Copilot CLI — use .github/copilot-instructions.md or .instructions.md files
• Subagent system prompts (--append-subagent-system-prompt) can't be injected in Copilot CLI; the flag and its text are ignored with a warning
• Cloud web sessions (--cloud, and its deprecated alias --remote) map to /delegate when a task is provided; bare web-session launch has no Copilot CLI equivalent
• MCP configs have different JSON schemas — migrate manually
• Settings (~/.claude/settings.json vs ~/.copilot/settings.json) have different formats. Copilot CLI moved user settings from ~/.copilot/config.json to ~/.copilot/settings.json and migrates the old file automatically; repository settings live in .github/copilot/settings.json and .github/copilot/settings.local.json
• Windows symlinks may require running PowerShell as Administrator or enabling Developer Mode
• Budget limits (--max-budget-usd) aren't available in Copilot CLI
• Plugin URL loading (--plugin-url) is a Claude Code-only feature — Copilot CLI only supports local plugins via copilot plugin install <dir>
• /team-onboarding is a Claude Code–only command (generates team onboarding guides from session history) — no Copilot CLI equivalent
• /loop (/proactive) is a Claude Code–only command (runs a prompt repeatedly while the session stays open) — no Copilot CLI equivalent
• --bg / --background flag (start session as a background agent) is Claude Code-only — closest: Ctrl+X then b to promote a running task to the background
• --teammate-mode flag (set how agent team teammates display: in-process (default), auto, tmux, or iterm2 (added in v2.1.186); default changed from auto to in-process in v2.1.179) is Claude Code-only — Copilot CLI has no agent team display modes, so cpc drops the flag with a warning
• --advisor <model> flag (enable the server-side advisor tool; accepts opus, sonnet, or a full model ID — the fable alias was removed in v2.1.212 and now errors out) is Claude Code-only — no direct Copilot CLI equivalent
• --safe-mode flag (start with all customizations disabled for troubleshooting: CLAUDE.md, skills, plugins, hooks, MCP servers, custom commands/agents, output styles, etc.) is Claude Code-only — no direct Copilot CLI equivalent; closest is --no-custom-instructions, though the semantics differ significantly
• Background agent session management (attach, logs, respawn, rm, stop subcommands) is Claude Code-only — Copilot CLI manages sessions via /session and --resume. Note: claude respawn restarts a running or stopped background session (--all restarts every running session)
• claude daemon status is Claude Code-only — reports the state of Claude Code's background-session supervisor (version, socket directory, worker count); no Copilot CLI counterpart
• /code-review (which replaces /simplify in Claude Code; /simplify and /review are now aliases) maps to Copilot CLI /review. The --fix flag (apply the suggested fixes), the --comment flag (post inline PR comments), and effort levels (low|medium|high|xhigh|max|ultra) have no Copilot equivalent. Copilot CLI's /review [PROMPT] takes a free-form prompt, so cpc strips those arguments when /review ... is passed as the initial prompt (e.g. cpc -p "/review high --fix pr#123" → copilot -p "/review pr#123")
• /ultraplan was removed from Claude Code — use plan mode (/plan), which exists in both CLIs
• /autocompact [auto|<tokens>] (Claude Code v2.1.221+, sets how full the context window gets before Claude Code compacts automatically and saves it to user settings) has no Copilot CLI equivalent — Copilot's /compact is a one-shot compaction command, not a threshold setting
• --autocompact <auto|tokens> (the session-only launch-flag form of /autocompact) is Claude Code-only — cpc drops the flag and its value with a warning
• --environment <environment-id> (with -p, create a new Claude Code cloud session on a named environment and exit; self-hosted IDs look like ccpool_...; cannot be combined with --cloud) and --ref <branch> (base that cloud session's checkout on a named ref instead of local HEAD) are Claude Code-only — cpc drops both flags and their values with a warning; the closest Copilot CLI workflow is /delegate inside a session
• /claude-api [migrate|managed-agents-onboard|prompt-audit] is a Claude Code-only skill (prompt-audit, v2.1.221+, flags instructions written for older models in prompts, skills, and tool descriptions and proposes fixes as a diff) — no Copilot CLI equivalent
• /usage-credits (renamed from /extra-usage in Claude Code) is Claude Code-only — configure usage credits to keep working when you hit a limit; closest in Copilot CLI is /usage (stats only)
• /run, /run-skill-generator, /verify are Claude Code-only skills (v2.1.145+) that build, launch, and drive the project's app to observe a change running — no Copilot CLI equivalent
• /radio is a Claude Code-only command (opens Claude FM lo-fi radio in the browser) — no Copilot CLI equivalent
• /dataviz [request] is a Claude Code-only skill (v2.1.198+) offering design guidance for charts, graphs, and dashboards (picks the chart form, assigns color by role, validates the palette for colorblind safety) — no Copilot CLI equivalent
• /design-login is a Claude Code-only command (authorize design-system access for /design-sync with your claude.ai account; requires a claude.ai subscription) — no Copilot CLI equivalent
• /design-sync [hint] is a Claude Code-only skill (convert your repo's React design system and upload it to Claude Design so generated designs use real components) — only available on the Anthropic API (not on Amazon Bedrock, Google Cloud Vertex AI, or Microsoft Foundry) and no Copilot CLI equivalent
• /background (/bg) slash command (detach current session to background) is Claude Code-only — closest: Ctrl+X then b
• /goal [condition|clear] (Claude Code: set a goal for a multi-turn agentic loop) maps to Copilot CLI's /autopilot [OBJECTIVE] (alias /goal, experimental) — with no objective it infers intent from context, and --max-ai-credits N caps autopilot spend; /goal on//goal off toggle autopilot without setting an objective, so Claude Code's clear/stop/off/cancel variants map to Copilot's /goal off. This closes a previous gap where Copilot had no equivalent for Claude Code's autonomous goal-tracking
• /stop slash command (stop current background session while attached) is Claude Code-only — no Copilot CLI equivalent
• /scroll-speed is a Claude Code-only UI command — no Copilot CLI equivalent
• claude install [version] is Claude Code-only — use copilot update (no version pinning)
• claude setup-token is Claude Code-only — use gh auth token for CI/script authentication
• claude project purge [path] is Claude Code-only — use /session cleanup or /session prune in Copilot CLI
• /tui, /focus, /heapdump, /recap are Claude Code–only UI/debugging commands — no Copilot CLI equivalents
• /ultrareview [PR] is a Claude Code–only command (deep cloud-based code review) — use /review in Copilot CLI for local reviews
• /keep-alive [on|off|busy|DURATION] (/caffeinate) is a Copilot CLI-only slash command (prevent machine sleep; duration accepts bare numbers, 30m, 2h, 1d) — no Claude Code equivalent
• /research, /update (/upgrade), /version are Copilot CLI-only slash commands — Claude Code's closest analog to /research is /deep-research <question> (best-effort mapping; the research pipelines differ)
• /bug [report] is now the primary bug-reporting command in Claude Code, with /share as its alias (before v2.1.212 both /bug and /share were aliases of /feedback); /feedback still opens the same dialog, and in the VS Code extension /bug opens the extension's own feedback dialog (v2.1.229+). The /share alias collides in name with Copilot CLI's /share [link|off|file|html|gist|research] [...] (session sharing and export). Same name, different action — /share submits a bug report in Claude Code but shares or exports the session in Copilot CLI
• /import [codex|gemini] [--dry-run] [--yes] is a Claude Code-only command (v2.1.213+) that imports another coding agent's configuration — instruction files, MCP servers, commands, subagents, and skills — from OpenAI Codex or Google Gemini CLI. Not available on Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry, or Claude Platform on AWS. --dry-run previews without writing, --yes skips the picker. Copilot CLI is not a supported source, so there is nothing for cpc to map
• /list-agents (/peers) is a Claude Code-only command (v2.1.224+) that lists the subagents and other Claude Code sessions Claude can message, and is only available when cross-session messaging is enabled. Copilot CLI has no equivalent — /subagents (/agents) configures subagent models, and the list_agents tool only covers background agents started from the current session
• /doctor (alias /checkup) is a Claude Code-only command (installation and auto-updater health check) — no Copilot CLI equivalent
• /deep-research <question> is a Claude Code workflow (fan out web searches, cross-check sources, synthesize a cited report) — the cpc wrapper treats it as a best-effort mapping to Copilot CLI's /research TOPIC, which uses GitHub search + web sources
• /advisor [model|off] is a Claude Code-only command (enable/disable the server-side advisor tool; accepts opus, sonnet, or a full model ID — the fable alias was removed in v2.1.212) — no Copilot CLI equivalent
• /cd <path> (Claude Code v2.1.169+, move the session to a new working directory) maps to Copilot CLI's /cd [PATH] (combined with /cwd)
• /reload-skills (Claude Code v2.1.152+, re-scan skill/command directories without restarting) maps to Copilot CLI's /skills reload
• /fork changed semantics again in Claude Code v2.1.212 — /fork [prompt] now copies the current conversation into a new background session while the original keeps running (the two are independent afterwards), which matches Copilot CLI's own /fork [NAME] / /branch [NAME] (experimental). The forked-subagent behavior it had on v2.1.161–v2.1.211 moved to /subtask
• /subtask <task> (Claude Code v2.1.212+) spawns a forked subagent that inherits the full conversation and works on the task while you keep working; the result returns to the conversation when it finishes. Copilot CLI has no 1:1 equivalent — the closest is /fleet <task> (parallel subagent execution)
• /search [QUERY] (/find) is a Copilot CLI-only experimental command (search the conversation timeline) — no Claude Code equivalent
• --effort / --reasoning-effort is now supported in both CLIs and passes straight through — Copilot CLI accepts the same five levels as Claude Code (low, medium, high, xhigh, max; max is the highest-depth tier for Anthropic models), so cpc forwards it to Copilot's --effort=LEVEL unchanged. Claude Code's ultracode level (v2.1.203+: xhigh reasoning + automatic workflow orchestration) has no Copilot equivalent — cpc maps it to max with a warning
• --ax-screen-reader (Claude Code v2.1.181+, renders screen-reader-friendly output and forces the classic renderer) maps to Copilot CLI's --screen-reader
• --worktree / -w [NAME] (create or reuse an isolated Git worktree) now maps to Copilot CLI's --worktree[=NAME] (-w) flag — cpc forwards claude --worktree NAME → copilot --worktree=NAME. Copilot's worktree flag is experimental, so enable Copilot CLI's experimental mode for it to take effect; cpc prints a warning as a reminder
• /worktree is a Copilot CLI-only slash command (experimental). /worktree [branch|task] creates a new Git worktree and switches the current conversation to it, leaving uncommitted changes behind; /worktree new [PROMPT] starts a new conversation in a new worktree and leaves the current conversation and its working directory unchanged (new is a reserved keyword and can't be a literal branch name). Claude Code has no exact equivalent — /fork and /branch copy the conversation but don't create a worktree
• /voice now exists in both CLIs with different arguments — Claude Code's /voice [hold|tap|off] selects the input style, while Copilot CLI's /voice [on|off|models|devices] toggles voice mode, browses voice models, or picks the input device (microphone). Only on/off overlap, so drop hold/tap when switching
• -r is now documented as the shorthand for --resume in Copilot CLI, matching Claude Code. cpc forwards both forms as --resume=SESSION (Copilot spells the flag --resume[=VALUE], so the value must be attached with =); bare --resume / -r opens the session picker, which requires a TTY
• /config gained inline key=value support in Claude Code v2.1.181 (e.g. /config thinking=false, also works in -p mode). Copilot CLI now documents /config as an alias of /settings, so the command name matches — use a space instead of = (/config thinking false)
• /clikit [COMPONENT] is a Copilot CLI-only internal/debug command — no Claude Code equivalent
• /tuikit [colors|icons|select|tabbar] is a Copilot CLI-only internal/debug command (preview TUIkit design-system components and color tokens) — no Claude Code equivalent
• /env is a Copilot CLI-only slash command (show loaded environment details) — no Claude Code equivalent
• /rubber-duck [PROMPT] is a Copilot CLI-only slash command (consult the rubber duck agent for a second opinion on plans, code, and tests) — no Claude Code equivalent
• /sandbox exists in both CLIs but with different syntax — Claude Code's /sandbox toggles sandbox mode, while Copilot CLI's /sandbox [config|status|policy|enable|disable] manages OS-level sandboxing for shell commands, MCP/LSP servers, and built-in file/web tools: config (or bare /sandbox) opens the settings dialog, status reports whether sandboxing is enabled, policy shows the effective path grants, denials, and network access, and enable/disable toggle it directly (the older enable/disable usage still works). Copilot CLI also has --sandbox / --no-sandbox launch flags (experimental mode only) that enable or disable the OS-level shell sandbox for a single session without changing the saved setting (handy with -p) — Claude Code has no launch-flag equivalent, and cpc passes both through unchanged
• /permissions differs between the CLIs — Claude Code manages persistent allow/ask/deny rules with /permissions; Copilot CLI's /permissions [default|assisted|allow-all|show] now switches between permission modes (default, assisted, allow-all), and /permissions reset clears in-memory tool and path approvals. /allow-all (/yolo) is documented as an alias for /permissions allow-all — its on option was replaced with auto (/allow-all [off|auto|show])
• /compact [FOCUS-INSTRUCTIONS] now accepts optional focus instructions in both CLIs (e.g. /compact focus on the auth module); cpc passes in-session slash commands through unchanged
• /chronicle is a Copilot CLI-only experimental command — standup summarizes recent work, tips and improve surface usage insights and propose instruction-file changes, reindex rebuilds the session index, and skills create|review|status propose, review, and report on repository skills. No Claude Code equivalent
• /streamer-mode (/on-air) is a Copilot CLI-only command that hides preview model names and quota details for streaming — no Claude Code equivalent
• --connect[=SESSION-ID] is a Copilot CLI-only flag for remote session joining — no direct Claude Code equivalent (see --remote and --teleport)
• --attachment PATH is a Copilot CLI-only flag that attaches a file to the initial prompt (image files accepted if model/org policy allows vision input; repeatable). cpc passes it through and consumes its value, e.g. cpc --attachment image.png "describe this". Claude Code has no launch-flag equivalent
• --context default|long_context is a Copilot CLI-only flag that sets the context window tier (overrides the persisted setting). cpc passes it through unchanged; Claude Code manages context via /compact and /context instead
• -C DIRECTORY is a Copilot CLI-only flag that changes the working directory before launch. cpc passes it through unchanged; Claude Code uses --add-dir / --worktree instead
• --acp, --allow-all-mcp-server-instructions, --enable-memory, and --extension-sdk-path DIRECTORY are Copilot CLI-only flags with no Claude Code equivalent — cpc passes them through as-is
• --mode=MODE and --plan are Copilot CLI-only flags — cpc maps --permission-mode plan → --plan; Claude Code's --permission-mode manual alias maps like default and needs no Copilot flag
• COPILOT_SUBAGENT_MAX_DEPTH (default 4, range 1–128) and COPILOT_SUBAGENT_MAX_CONCURRENT (default 32, range 1–256) are Copilot CLI-only environment variables for tuning subagent behavior
• GITHUB_COPILOT_PROMPT_MODE_EXTENSIONS, GITHUB_COPILOT_PROMPT_MODE_REPO_HOOKS, and GITHUB_COPILOT_PROMPT_MODE_WORKSPACE_MCP control whether prompt mode (-p) loads repository-provided extensions, hooks, and MCP sources (all disabled by default for security). Set to true explicitly when using cpc -p with repo hooks or MCP servers
• copilot logout subcommand has been removed — use /logout in an interactive session instead
• /security-review is now a direct match — both Claude Code and Copilot CLI (/security-review [PROMPT]) run a security review agent that analyzes pending changes for vulnerabilities
• /after [DELAY PROMPT] and /every [INTERVAL PROMPT] are Copilot CLI-only experimental commands (schedule a one-shot or recurring prompt, skill, or schedulable slash command for the session; no args opens the schedule manager) — Claude Code's /schedule (/routines) only partially overlaps
• /extensions (/extension) [manage|mode] is a Copilot CLI-only command for managing CLI extensions — Claude Code's /plugin is a conceptually similar but distinct system
• /app is a Copilot CLI-only command (launch the GitHub Copilot desktop app, or show its download URL) — no Claude Code equivalent
• /settings [--repo|--local] [show KEY|KEY|KEY VALUE] (alias /config) opens the Copilot CLI settings dialog, opens it focused on a specific key, sets a setting inline, or prints a key's current value (secret-named values are masked). --repo and --local write to .github/copilot/settings.json / settings.local.json instead of user settings, and only repo-overridable keys can be set that way
• /subagents (alias /agents) configures default and per-agent subagent models in Copilot CLI — a richer counterpart to Claude Code's /agents than /agent
• /mcp gained a search subcommand in Copilot CLI (/mcp [show|add|edit|delete|disable|enable|auth|reload|search] [SERVER-NAME]) for searching available MCP servers
• /refine [TEXT] is a Copilot CLI-only slash command (rewrite a roughly composed prompt into a clear one for review before sending; no args — via Ctrl+X then /refine — cleans up the current input box) — no Claude Code equivalent, and cpc can't translate it because it is in-session only
• $ (a lone $ on an empty prompt) is a Copilot CLI-only shortcut that hands the terminal over to a real interactive shell ($SHELL on Unix, %COMSPEC% on Windows), suspending the CLI UI so job control, full-screen apps, tab completion, and colors work natively. Disabled by default — enable it with the shellShortcut setting; only available for local, trusted, idle sessions on a real TTY, and it can be disabled in enterprise managed settings
• copilot plugins [list|enable|disable|remove] [--plugin|--mcp|--skill] NAME (note the plural) is a Copilot CLI-only subcommand for non-interactively inspecting, enabling, disabling, or uninstalling plugins, MCP servers, and skills. enable/disable persist to configuration; remove can only delete personal and project skills (disable plugin-provided or built-in ones instead). cpc plugins ... passes through to copilot plugins ... (cpc plugin ... still maps to copilot plugin ...)
• mai-code-1-flash is a Copilot CLI model (fast, adaptive coding tasks); pass it through with cpc --model mai-code-1-flash
• gemini-3.6-flash is a Copilot CLI model (fast Google Gemini responses), joining gemini-3.1-pro-preview and gemini-3.5-flash; pass it through with cpc --model gemini-3.6-flash
• COPILOT_MCP_TOOL_CACHE (set to false to disable loading and persisting local MCP server tool snapshots for the whole process; existing cache files are left untouched) and COPILOT_TASK_WAIT_TIMEOUT_SECONDS (default 600; maximum seconds -p, including -p --autopilot, waits for pending background agents or shell commands before exiting — 0 exits immediately) are Copilot CLI-only environment variables. COPILOT_TASK_WAIT_TIMEOUT_SECONDS is worth setting for scripted cpc -p pipelines
• COPILOT_COMPUTER_USE_LINUX (opt in to the computer-use MCP server on supported Linux distributions; not available on Alpine Linux / musl libc) and COLORFGBG (fallback for dark/light terminal background detection) are no longer listed in the Copilot CLI reference docs — they may still work, but treat them as undocumented. COPILOT_ENABLE_HTTP2 (set to 1/true to opt into HTTP/2 transport; HTTP/1.1 is the default) remains a documented Copilot CLI-only environment variable
• copilot help [TOPIC] now supports a sandbox topic in addition to billing, config, commands, environment, logging, monitoring, permissions, and providers

Architecture

This project has two parts:

Copilot CLI Plugin (plugin.json)

Installed via copilot plugin install. Provides:

• /claude-help skill — type it in any Copilot session for the full mapping reference
• sessionStart hook — shows a reminder that the plugin is active

cpc Companion Wrapper

A standalone Python script (not part of the plugin — plugins can't intercept CLI invocation). It:

1. Checks for subcommands (update, auth login, plugin, etc.)
2. Translates flags that differ between the two CLIs
3. Warns on flags with no Copilot equivalent
4. Passes through all unknown/matching flags unchanged
5. On Unix, calls os.execvp("copilot", ...) to replace the process — TTY, stdin, and signals are inherited transparently. On Windows, uses subprocess.run() and forwards the exit code

Why can't cpc be a plugin? Copilot CLI plugins can provide skills, agents, hooks, and MCP servers — but they cannot intercept or modify how copilot itself is invoked from the shell. Flag translation must happen before copilot starts, which only an external wrapper can do.