Skip to content

ROADMAP.md

Latest commit

 

History

History
229 lines (156 loc) · 9.59 KB

ROADMAP.md

File metadata and controls

229 lines (156 loc) · 9.59 KB

Roadmap

Feature gap analysis between the Tool Manager and Claude Code's current capabilities. Organized by priority.

High Priority

Core features users will expect to configure through the Tool Manager.

Hooks: Missing Event Types (Done)

All 14 missing hook events added — types, metadata, store ordering, templates, sound presets, and tests.

  • InstructionsLoaded — fires when CLAUDE.md or rules are loaded (matcher: load reason)
  • PostToolUseFailure — fires after a tool fails (separate from PostToolUse success path)
  • SubagentStart — fires when a subagent spawns (matcher: agent type name)
  • TaskCompleted — fires when a background task completes
  • TeammateIdle — fires before a teammate idles (agent teams)
  • StopFailure — fires on error (matcher: rate_limit, authentication_failed, server_error)
  • ConfigChange — fires when settings change (matcher: user_settings, project_settings, policy_settings)
  • CwdChanged — fires when the working directory changes
  • FileChanged — fires on file changes (matcher: filename)
  • WorktreeCreate — fires when a git worktree is created
  • WorktreeRemove — fires when a git worktree is removed
  • PostCompact — fires after context compaction
  • Elicitation — fires on MCP user input request
  • ElicitationResult — fires on MCP elicitation response

Hooks: Missing Hook Types (Done)

Added http and agent hook types with full DB, backend, and UI support.

  • http — POST to a URL with headers and env var substitution ($VARIABLE)
    • Fields: url, headers, allowedEnvVars
  • agent — spawn a subagent to process the hook event

Hooks: Missing Hook Fields (Done)

Added 5 universal hook fields with DB columns, settings.json serialization, and form UI.

  • if — permission rule syntax filter (e.g., Bash(rm *))
  • statusMessage — custom spinner text while hook runs
  • once — run only once per session
  • async — run in background (command hooks)
  • shell — bash (default) or powershell

Permission Modes (Done)

The app exposes default, allowEdits, bypassPermissions. Added:

  • plan — read-only exploration mode
  • auto — background safety checks (research preview)
  • dontAsk — auto-deny unless pre-approved

Effort Level Configuration (Done)

Added effortLevel scoped setting with full DB, backend, settings writer, and UI support. Also added effort to skill frontmatter.

  • Add effortLevel setting (low, medium, high, max)
  • Surface in model/settings UI
  • Support effort in subagent frontmatter
  • Support effort in skill frontmatter
  • Env var toggle: CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING

Subagent Frontmatter: New Fields (Done)

Added 9 missing frontmatter fields with full DB, backend, writer, scanner, and UI support.

  • disallowedTools — tool denylist (complement of existing tools allowlist)
  • maxTurns — limit iterations
  • memory — persistent memory scope (user, project, local)
  • background: true — always run in background
  • effort — effort level override
  • isolation: worktree — run in a git worktree
  • hooks — scoped hooks (run only within this subagent)
  • mcpServers — scoped MCP servers (available only to this subagent)
  • initialPrompt — auto-submitted first prompt

Skill Frontmatter: New Fields (Done)

Added 6 missing frontmatter fields with full DB, backend, writer, scanner, and UI support.

  • context: fork — run in subagent context
  • agent — which subagent type to use
  • hooks — lifecycle hooks scoped to this skill
  • paths — glob patterns for auto-loading (comma-separated or list)
  • shell — bash or powershell
  • once — run only once per session

Rules Directory Management (Done)

Added full rules management with DB, backend, filesystem scanning, and UI.

  • Manage .claude/rules/ (project) and ~/.claude/rules/ (user) directories
  • Create/edit/delete rule files (markdown with frontmatter)
  • Support paths frontmatter for glob-based conditional loading
  • Show which rules are active for a given file context
  • Symlink support for shared rules across projects

WebSocket MCP Transport (Done)

Added ws as a 4th MCP transport type with full DB, config writer, and UI support.

  • Add ws transport type for MCP servers (wss:// URLs)
  • UI support in MCP creation/edit forms

Medium Priority

Power user and advanced configuration features.

Auto Mode Configuration (Done)

Added Auto Mode settings tab with prose fields and disableAutoMode toggle (moved from Session tab).

  • autoMode.environment — trusted infrastructure descriptions (prose)
  • autoMode.allow — actions to allow (prose descriptions)
  • autoMode.soft_deny — actions to block (prose descriptions)

Model Overrides & Extended Context (Done)

Added ModelOverridesEditor key-value component and extended context model shortcuts.

  • modelOverrides setting — map Anthropic model IDs to provider-specific IDs (Bedrock, Vertex, Foundry)
  • Extended context suffixes: sonnet[1m], opus[1m]
  • opusplan alias (Opus in plan mode, Sonnet otherwise)
  • availableModels setting — restrict which models users can select
  • Custom model env vars (ANTHROPIC_CUSTOM_MODEL_OPTION, _NAME, _DESCRIPTION) — added to Known Env Vars picker
  • Third-party provider pinning env vars (ANTHROPIC_DEFAULT_OPUS_MODEL, etc.) — added to Known Env Vars picker

Expanded Sandbox Settings (Done)

Added SandboxFilesystemEditor component and allowManagedDomainsOnly toggle. Proxy ports, weaker sandbox, and unsandboxed commands were already in UI.

  • sandbox.filesystem.allowRead — re-allow reading within denied regions
  • sandbox.filesystem.denyRead — paths subprocesses cannot read
  • sandbox.filesystem.allowUnixSockets — allow specific Unix sockets (e.g., Docker)
  • sandbox.network.httpProxyPort — custom HTTP proxy port
  • sandbox.network.socksProxyPort — custom SOCKS proxy port
  • sandbox.network.allowManagedDomainsOnly — only allow managed domain lists
  • sandbox.enableWeakerNestedSandbox — weaker mode for Docker compatibility
  • sandbox.allowUnsandboxedCommands — allow dangerouslyDisableSandbox parameter

Permission Rule Syntax Expansion (Done)

Added Agent and Skill tool types to permission rule builder with templates.

  • Agent(subagent-name) — match specific subagent spawning
  • Skill(skill-name) / Skill(skill *) — match skill invocation
  • WebFetch(domain:example.com) — domain-level filtering for web fetches

Agent Memory (Done)

Added per-subagent persistent memory system with full backend service, Tauri commands, store, and UI.

  • Manage .claude/agent-memory/ (project) and ~/.claude/agent-memory/ (user) directories
  • Per-agent MEMORY.md index files
  • Browse/edit agent memory from the subagent detail view

New Settings Keys (Done)

Added 6 new settings keys with full DB, backend, settings writer, and UI support.

  • autoMemoryEnabled / autoMemoryDirectory — auto memory toggle and custom path
  • claudeMdExcludes — glob patterns to skip CLAUDE.md files (monorepo support)
  • agent — default subagent for a project
  • attribution.enabled / attribution.rules — contribution tracking with file patterns
  • disableBypassPermissionsMode — already existed in managed settings
  • disableAutoMode — prevent auto mode

Lower Priority

Enterprise, niche, or session-scoped features.

Agent Teams (Done)

Added agentTeamEnabled setting with toggle UI and moved teammateMode into Agent Teams card.

  • agentTeamEnabled setting
  • UI to visualize/configure team composition
  • Team size, model per agent, display mode settings

Managed Settings (Enterprise)

Admin-deployed, read-only configuration:

  • allowManagedPermissionRulesOnly — only enforce managed permission rules
  • allowManagedHooksOnly — only load managed hooks
  • allowManagedMcpServersOnly — only use managed MCP servers
  • forceLoginMethod / forceLoginOrgUUID — lock authentication
  • allowedChannelPlugins — restrict channel plugins
  • sandbox.filesystem.allowManagedReadPathsOnly — restrict to managed read paths
  • Better read-only display for managed settings in the Admin tab

Cron / Scheduled Tasks (Done)

Added CLI & Scheduling settings tab with /loop, /schedule, and CronCreate/List/Delete reference cards.

  • Expose /loop interval configuration
  • Expose /schedule for cloud/desktop persistent scheduling
  • CronCreate/CronList/CronDelete integration
  • Visual cron expression builder

Plugin Marketplace: npm Source (Done)

Already implemented — npm source type exists in MarketplaceSourceForm with @scope/package support.

  • Add npm as a marketplace source type (@scope/package)
  • npm package resolution and installation flow

CLI Startup Flags (Done)

Added CLI & Scheduling settings tab with flag reference table and interactive command builder.

  • --agent — run session as specific subagent
  • --baremode — minimal plugin startup
  • --system-prompt / --append-system-prompt — custom system prompts
  • --permissions / --allowedTools / --disallowedTools — runtime overrides

Tracking

Last audited: 2026-03-29 Audited against: Claude Code capabilities as of March 2026