Epic: Feature Comparison - Claude Code vs ForgeCode

[ImBIOS]

Epic: Claude Code vs ForgeCode Feature Comparison

Track features that Claude Code has and their status in ForgeCode. Use this issue to identify gaps and prioritize development.

Legend

• = ForgeCode already has this feature
• = ForgeCode does NOT have this feature yet
• [~] = Partial implementation

---

Core CLI Features

Feature	Claude Code	ForgeCode	Status	Notes
Interactive TUI	Yes	Yes	[x]	ForgeCode has full interactive mode
Non-interactive mode (-p)	Yes (claude -p)	Yes (forge -p)	[x]	Both support piped input
Session resume/continue	Yes (--continue, --resume)	Yes (--conversation-id)	[x]
Session fork	Yes (--fork-session)	Yes (conversation clone)	[x]	conversation clone <id> creates new conversation with new ID. Task tool session_id allows subagent resume.
Named sessions	Yes (--resume "name")	Yes (conversation rename)	[x]	conversation rename <id> <name>
Session list	Yes (claude ls)	Yes (conversation list [--porcelain])	[x]
Pipe input	Yes (cat file | claude -p)	Yes (cat file | forge -p)	[x]
Multi-line input	Yes	Yes	[x]
Output formats (JSON/streaming)	Yes (--output-format json/stream-json/text)	Partial (--porcelain, conversation dump --html)	[~]	--porcelain on most subcommands, JSON/HTML export via conversation dump. No stream-json event streaming.
System prompt customization	Yes (--system-prompt)	Yes (via AGENTS.md)	[x]
Model selection	Yes (--model)	Yes (via config)	[x]
Agent selection	Yes (--agent)	Yes (--agent)	[x]
Budget/cost control	Yes (--max-budget-usd)	No	[ ]
Effort level control	Yes (--effort)	Yes (forge config set reasoning-effort <level>)	[x]	Effort enum: none/minimal/low/medium/high/xhigh/max. Also max_tokens, exclude, enabled flags. crates/forge_config/src/reasoning.rs, crates/forge_domain/src/agent.rs
Permission modes	Yes (ask/auto-plan/auto-edit/full-auto)	Yes (Policy Allow/Deny/Confirm)	[x]	Policy enum with All/Any/Not operators, Local/User scope. crates/forge_domain/src/policies/policy.rs
Verbose/debug mode	Yes (--verbose)	Yes (--verbose)	[x]

Agent & Subagent Features

Feature	Claude Code	ForgeCode	Status	Notes
Subagents	Yes	Yes (task tool)	[x]	task tool with agent_id + session_id for resume. Custom agents in .forge/agents/.
Agent view (parallel management)	Yes (claude agents)	No	[ ]
Agent teams	Yes	No	[ ]
Dynamic workflows	Yes	No	[ ]
Background agents	Yes	No	[ ]
Custom subagent definitions	Yes (.claude/subagents/)	Yes (.forge/agents/)	[x]

Memory & Context

Feature	Claude Code	ForgeCode	Status	Notes
CLAUDE.md / AGENTS.md	Yes	Yes (AGENTS.md)	[x]
Auto memory	Yes	Yes (TitleGenerationHandler)	[x]	Auto-generates conversation title from user prompt. crates/forge_app/src/hooks/title_generation.rs
Memory files per project	Yes (nested CLAUDE.md)	Yes (AGENTS.md)	[x]
Checkpointing (file rewind)	Yes	Yes (fs_undo tool + forge_snaps)	[x]	FSUndo in ToolCatalog, SnapshotService for per-file snapshot/rewind. crates/forge_domain/src/tools/catalog.rs, crates/forge_snaps/src/service.rs
Context window management	Yes (/compact)	Yes (conversation compact <id>)	[x]	Compact + Compactor engine, CompactionStrategy (Evict/Retain/Min/Max), CompactionHandler hook. crates/forge_domain/src/compact/
Prompt caching	Yes (automatic)	No	[ ]

Tools & Integrations

Feature	Claude Code	ForgeCode	Status	Notes
MCP support	Yes	Yes	[x]	McpServerConfig::{Stdio, Http} transports. crates/forge_domain/src/mcp.rs
MCP server mode	No	Yes	[x]	crates/forge_infra/src/mcp_server.rs — Forge can expose itself as MCP server. ForgeCode advantage
MCP OAuth	No	Yes (mcp login/mcp logout)	[x]	OAuth device/code/PKCE, refresh, custom headers. crates/forge_domain/src/auth/oauth_config.rs ForgeCode advantage
Shell/command execution	Yes	Yes	[x]
File read/write	Yes	Yes	[x]
Web search	No (external MCP)	No	[x]	Neither has built-in web search. ForgeCode has Fetch tool for URL content retrieval.
Computer use (GUI)	Yes (--chrome)	No	[ ]
Browser automation	Yes (Chrome integration)	No	[ ]
GitHub Actions integration	Yes	No	[ ]
GitLab CI/CD integration	Yes	No	[ ]
Slack integration	Yes	No	[ ]
Hooks system	Yes (user-configurable)	Yes (internal EventHandle/Hook)	[~]	on_start, on_end, on_request, on_response, on_toolcall_start, on_toolcall_end. Built-in: CompactionHandler, DoomLoopDetector, TitleGenerationHandler. crates/forge_domain/src/hook.rs. Not user-configurable yet.
Custom tools	Yes	Yes (16 tools in ToolCatalog)	[x]	Read, Write, FsSearch, SemSearch, Remove, Patch, MultiPatch, Undo, Shell, Fetch, Followup, PlanCreate, SkillFetch, TodoWrite, TodoRead, Task. crates/forge_domain/src/tools/catalog.rs
Skills/Commands	Yes	Yes	[x]	Skills in .forge/skills/, commands in .forge/commands/.
Plugins	Yes (plugin marketplace)	Yes (skills + shell plugin)	[~]

IDE & Platform

Feature	Claude Code	ForgeCode	Status	Notes
VS Code extension	Yes	Yes (forge vscode install-extension)	[x]
JetBrains integration	Yes	No	[ ]
Desktop app	Yes	No	[ ]
Web interface	Yes (claude.ai/code)	No	[ ]
Mobile access	Yes (Remote Control)	No	[ ]
Deep links (URL launch)	Yes (claude-cli://)	No	[ ]

SDK & Automation

Feature	Claude Code	ForgeCode	Status	Notes
Agent SDK (Python)	Yes	No	[ ]
Agent SDK (TypeScript)	Yes	No	[ ]
Structured output	Yes (--json-schema)	Yes (forge data --schema)	[~]	forge data runs JSONL batches constrained by --schema JSON schema. No --json-schema flag on -p itself.
Headless/programmatic mode	Yes	Yes (forge -p + forge data)	[x]
Scheduled tasks/routines	Yes (Desktop)	No	[ ]
Cron/scheduled execution	Yes (/loop)	No	[ ]
Channels (push events)	Yes	No	[ ]

Code Review & Quality

Feature	Claude Code	ForgeCode	Status	Notes
Code review	Yes (/code-review)	Yes (via subagents)	[x]
Multi-agent cloud review	Yes (ultrareview)	No	[ ]
Cloud planning	Yes (ultraplan)	No	[ ]
PR auto-fix	Yes (/autofix-pr)	No	[ ]
Security review plugin	Yes (security-guidance)	No	[ ]

Worktree & Isolation

Feature	Claude Code	ForgeCode	Status	Notes
Worktree support	Yes (--worktree)	Yes (--sandbox <name>)	[x]	Creates git worktree. crates/forge_main/src/sandbox.rs
Dev containers	Yes	No	[ ]
Docker sandbox	Yes	No	[ ]
Sandboxed bash	Yes	Yes (Policy system)	[x]

Enterprise & Admin

Feature	Claude Code	ForgeCode	Status	Notes
Multi-provider support	Anthropic-only*	Yes (43 providers)	[x]	ForgeCode advantage: 43 providers including OpenAI, Anthropic, Google, GitHub Copilot, Bedrock, Vertex AI, Azure, OpenRouter, Cerebras, Ollama, LM Studio, llama.cpp, vLLM, Jan AI, etc. crates/forge_repo/src/provider/provider.json
Enterprise settings	Yes (server-managed)	No	[ ]
Usage analytics	Yes (analytics dashboard)	Yes (forge info, conversation stats)	[x]	Usage struct with cost field. crates/forge_main/src/info.rs
Cost tracking	Yes (--max-budget-usd)	Yes (Usage + stats)	[~]	Has cost display, no budget cap.
Team management	Yes	No	[ ]
Amazon Bedrock support	Yes	Yes	[x]
Google Vertex AI support	Yes	Yes	[x]
Azure support	Yes	Yes	[x]
Zero data retention	Yes	No	[ ]
Managed MCP policies	Yes	No	[ ]

UX Features

Feature	Claude Code	ForgeCode	Status	Notes
Voice dictation	Yes	No	[ ]
Vim mode	Yes	No	[ ]
Custom keybindings	Yes	Partial (ZLE widgets)	[~]	shell-plugin/lib/bindings.zsh registers custom ZLE widgets (forge-accept-line, forge-completion, bracketed-paste).
Custom color themes	Yes (via plugins)	Yes (forge.theme.zsh)	[x]
Fullscreen rendering	Yes	No	[ ]
Status line customization	Yes	Yes (forge zsh rprompt)	[x]	Shows model + conversation stats in right-prompt.
Interactive mode keyboard shortcuts	Yes	Yes	[x]
Shell completion (zsh)	Yes	Yes (full shell plugin)	[x]	shell-plugin/ with plugin, theme, keyboard, dispatcher, actions, context, completion, highlight, config, helpers, bindings.
Auto-update	Yes	Yes (forge update)	[x]
Doom-loop detection	No	Yes	[x]	DoomLoopDetector hook detects [A,A,A,A] and [A,B,C][A,B,C][A,B,C] patterns. ForgeCode advantage. crates/forge_app/src/hooks/doom_loop.rs
Interactive fuzzy picker	No	Yes (forge select)	[x]	forge select {model, agent, provider, reasoning-effort, command, conversation, file}. ForgeCode advantage

---

Summary

ForgeCode already has: ~45 features
ForgeCode is missing: ~25 features
Partial implementation: ~6 features

Top Priority Missing Features (by impact)

1. Agent SDK (Python/TypeScript) - Critical for automation
2. Streaming JSON output (--output-format stream-json) - Critical for CI/CD
3. Budget/cost cap (--max-budget-usd) - Important for enterprise
4. Agent view (parallel management) - High impact for productivity
5. ACP (Agent Client Protocol) - See [Feature]: Support ACP (Agent Client Protocol) #2968
6. Desktop app - Important for UX
7. Web interface - Important for accessibility
8. GitHub Actions / GitLab CI / Slack integrations - Important for workflows
9. JetBrains integration - Important for developer reach
10. Computer use (--chrome) - Important for GUI workflows

ForgeCode Unique Advantages (not in Claude Code)

• 43 provider support (vs Claude Code Anthropic-only)
• MCP server mode (Forge can act as MCP server)
• MCP OAuth (device/code/PKCE login/logout)
• Semantic code search (workspace indexing via embeddings)
• Doom-loop detection (automatic stuck-pattern breaker)
• Interactive fuzzy picker (forge select for models, agents, providers, etc.)
• Full shell plugin (ZSH theme, plugin, keyboard bindings, dispatcher, actions)
• Custom commands/skills system
• Plan tool (PlanCreate for structured planning)
• forge data (JSONL schema-constrained data processing pipeline)
• forge commit (AI-generated commit messages)
• forge suggest (NL to shell command)
• Conversation lifecycle (list, new, dump, compact, retry, resume, show, info, stats, clone, rename, delete)
• Config migration (forge config migrate for ~/forge → ~/.forge)
• Diagnostics (forge doctor, forge zsh doctor, forge logs)

Cross-references

• Related: Epic Epic: Feature Comparison - Codex CLI vs ForgeCode #3501 (Codex CLI comparison)
• Related: Epic Epic: Feature Comparison - OpenCode vs ForgeCode #3502 (OpenCode comparison)
• Related: Epic Epic: Feature Comparison - Cursor CLI vs ForgeCode #3503 (Cursor CLI comparison)
• Related: Issue [Feature]: Support ACP (Agent Client Protocol) #2968 (ACP support request)

Co-Authored-By: ForgeCode noreply@forgecode.dev