Status: In Progress (Slices 1, 2, 5, 6 complete: Plan/Build toggle, @mention subagents, generic LSP, LiteLLM provider)
ID prefix: CAP
Implementation plan: .plans/PHASE_21_CODING_AGENT_PARITY_V2.md (7 slices, target files, tests, acceptance criteria)
Beep.AI.Code is positioned as a terminal coding agent in the same class as Claude Code and OpenCode (SST/Anomaly). This plan tracks the capability gaps to close for parity. Items cross-reference existing backlogs where they extend shipped work.
| Capability | Claude Code | OpenCode | Beep.AI.Code today | Gap |
|---|---|---|---|---|
| Plan/Build mode toggle | ✓ | ✓ (Tab: build/plan) | Sandbox modes + --mode; no one-key toggle |
Partial |
| Subagents | ✓ (Agent Teams) | ✓ (general/explore/scout, @mention) |
subagents/ (explore, plan, dispatch) |
Partial (no @mention) |
| Persistent client/server sessions | resume from file | ✓ (server + SQLite, survives SSH drop) | Local JSONL sessions; no daemon | Missing |
| Parallel multi-session agents | Agent Teams | ✓ (up to 8, git worktrees) | Single AgentSession |
Missing |
| Session share links | — | ✓ | — | Missing |
| LSP intelligence to LLM | ✓ | ✓ (auto-loads LSPs) | Jedi + tree-sitter (Python/TS); no generic LSP | Partial |
| IDE extensions / ACP | ✓ | ✓ (VS Code, Cursor, Zed, ACP) | — | Missing |
| Provider breadth + local | Anthropic | ✓ (75+, Ollama, models.dev) | OpenAI/Anthropic/OpenRouter/Beep packs | Partial |
| Hooks / automation | ✓ (29-hook system) | permissions rules | hooks.json (event hooks) |
Partial |
| Skills marketplace | ✓ | skill packs | Local skills loader only | Partial (see INT) |
| MCP support | ✓ | ✓ | ✓ (presets, verify-tools) | Parity |
| Project memory | CLAUDE.md | AGENTS.md | .beep.md + AGENTS.md rules |
Parity |
| ID | Item | Priority | Depends / ties | Verification |
|---|---|---|---|---|
| CAP-1 | Plan/Build toggle: one-key switch (REPL hotkey + TUI Tab) mapping to read-only vs workspace-write sandbox, with a visible mode banner | P0 | permissions/manager.py, chat/repl.py, tui/ |
REPL/TUI mode-switch tests |
| CAP-2 | @mention subagents in chat (@explore, @plan, @general) routing to agent/subagents/* |
P0 | subagents/dispatcher.py, chat/commands/ |
Dispatch test per mention |
| CAP-3 | Persistent agent server/daemon: background process holding session state so runs survive terminal/SSH exit; reconnect to in-flight run | P1 | NF-6 beep serve, sessions store |
Reconnect integration test |
| CAP-4 | Parallel multi-agent runs each in an isolated git worktree, with a coordinator and result merge | P1 | AG-1, AG-2; git_tool.py |
Worktree isolation test |
| CAP-5 | Session share/export link (HTML or server-hosted) for review/debug | P2 | NF-11, sessions exporter | Share artifact test |
| CAP-6 | Generic LSP client feeding diagnostics/hover/symbols to the model for any language | P1 | CX-3 | LSP smoke vs sample server |
| CAP-7 | Editor integration via Agent Client Protocol (ACP) + thin VS Code/Cursor/Zed adapters | P2 | NF-6, NF-7 | Protocol conformance test |
| CAP-8 | Provider breadth + local models: Ollama / LiteLLM / models.dev-style catalog; switch model per task | P1 | INT-4, provider_* |
Provider matrix test |
| CAP-9 | Hooks parity: expand lifecycle events (pre/post tool, pre/post run, on-error, on-edit) and document the set | P2 | hooks/manager.py |
Hook event-fire tests |
| CAP-10 | Skills catalog/marketplace with one-command install | P0 | INT-1 (integrations) | See INT plan |
| CAP-11 | /undo + checkpoint timeline for agent edits (revert to any step) |
P2 | editing/rollback.py, graph checkpoints |
Restore test |
| CAP-12 | Cost/limit guardrails surfaced like competitors (per-run budget, model price hints) | P1 | AG-4, NF-8 | Budget stop test |
CAP-1 and CAP-2 first (cheap, high visibility, reuse shipped sandbox + subagents), then CAP-10/INT-1 (skills catalog — the headline ask), then CAP-3/CAP-4 (persistence + parallelism), then CAP-6/CAP-8 (LSP + providers).
Return to ../ENHANCEMENT_PLAN.md · integrations-catalog.md