page:docs-harness-features-backlog-gaps-tools-capabilities-tools-coverage-map
Tools Coverage Map: CC (42 tools) vs Babysitter Agentic Tools (16) reference
This document maps every CC tool to its babysitter equivalent (if any) and identifies
Tools Coverage Map: CC (42 tools) vs Babysitter Agentic Tools (16)
This document maps every CC tool to its babysitter equivalent (if any) and identifies the gaps. Babysitter's agentic tools are defined in packages/sdk/src/harness/agenticTools.ts and injected into Pi sessions. When delegating to CC or other harnesses, those harnesses provide their own tools -- this map is specifically about what babysitter can provide when it IS the executing runtime (Pi sessions, direct execution).
Coverage Matrix
File Operations (6 CC tools)
| CC Tool | Babysitter Agentic Tool | Status | Gap? |
|---|---|---|---|
| FileReadTool | read | Near parity | Minor -- CC uses absolute paths + pages param for PDFs; babysitter uses workspace-relative paths, no PDF page support |
| FileEditTool | edit | Parity+ | Babysitter has replace_all param that CC lacks |
| FileWriteTool | write | Parity | No |
| GlobTool | find | Parity | No -- different impl, same functionality |
| GrepTool | grep | **Partial** | **GAP-TOOLS-035** |
| NotebookEditTool | notebook | Near parity | Minor -- CC uses cell_id, babysitter uses cell_index; babysitter has read action CC lacks |
**GAP-TOOLS-035**: CC's GrepTool has output_mode (content/files_with_matches/count), separate -B/-A/-C params for before_context/after_context/combined context, -n line number toggle, and head_limit (default 250). Babysitter only has context (equivalent to -C), limit, and offset. Missing: output mode selection, separate before/after context lines, line number toggle.
Execution (3 CC tools)
| CC Tool | Babysitter Agentic Tool | Status | Gap? |
|---|---|---|---|
| BashTool | bash | **Partial** | **GAP-TOOLS-036** |
| PowerShellTool | (none -- bash covers shell) | Partial | Minor |
| REPLTool | python (Python only) | Partial | **GAP-TOOLS-007** |
**GAP-TOOLS-007**: CC has JS/TS REPL. Babysitter has python tool only. Missing JS/TS REPL for interactive code evaluation in orchestrated tasks.
**GAP-TOOLS-036**: CC's BashTool has run_in_background and description params that babysitter lacks. Babysitter's bash has env and cwd params that CC lacks. Neither is a superset of the other.
Web (2 CC tools)
| CC Tool | Babysitter Agentic Tool | Status | Gap? |
|---|---|---|---|
| WebSearchTool | (none) | Missing | **GAP-TOOLS-008** (exists) |
| WebFetchTool | fetch | **Partial** | **GAP-TOOLS-037** |
**GAP-TOOLS-008**: CC has web search. Babysitter has no equivalent.
**GAP-TOOLS-037**: CC's WebFetchTool requires a prompt param that processes and summarizes fetched content before returning it. Babysitter's fetch returns raw response data with timeout and raw params. Different content processing model.
Note: CC does NOT have a WebBrowserTool. Babysitter's browser tool is babysitter-only (see Babysitter-only tools section below).
Agent/Delegation (3 CC tools)
| CC Tool | Babysitter Agentic Tool | Status | Gap? |
|---|---|---|---|
| AgentTool | (effect system) | Different model | See GAP-AGENT-001 |
| SendMessageTool | (none) | Missing | **GAP-TOOLS-015** merged into GAP-AGENT-005 |
| SyntheticOutputTool | (none) | Missing | **GAP-TOOLS-029** |
**GAP-TOOLS-029**: CC's SyntheticOutputTool creates structured output blocks (images, code blocks, downloadable files) that appear in the conversation. No equivalent for producing structured output from orchestrated tasks.
Task Management (6 CC tools)
| CC Tool | Babysitter Agentic Tool | Status | Gap? |
|---|---|---|---|
| TaskCreateTool | (effect system via ctx.task) | Different model | **GAP-TOOLS-014** (exists) |
| TaskGetTool | (task:show CLI) | CLI-only | **GAP-TOOLS-014** (exists) |
| TaskListTool | (task:list CLI) | CLI-only | **GAP-TOOLS-014** (exists) |
| TaskUpdateTool | (task:post CLI) | CLI-only | **GAP-TOOLS-014** (exists) |
| TaskStopTool | (none) | Missing | **GAP-TOOLS-030** |
| TaskOutputTool | (task:show CLI) | CLI-only | **GAP-TOOLS-014** (exists) |
**GAP-TOOLS-030**: No ability to cancel/stop a running effect. Once dispatched, effects run to completion or timeout. CC can cancel background tasks mid-execution.
MCP Tools (4 CC tools)
| CC Tool | Babysitter Agentic Tool | Status | Gap? |
|---|---|---|---|
| MCPTool | (none) | Missing | **GAP-TOOLS-025** (exists) |
| ListMcpResourcesTool | (none) | Missing | **GAP-TOOLS-031** |
| ReadMcpResourceTool | (none) | Missing | **GAP-TOOLS-031** |
| McpAuthTool | (none) | Missing | **GAP-TOOLS-032** |
**GAP-TOOLS-031**: MCP resource browsing and reading. CC can list and read resources from connected MCP servers. Babysitter has no MCP resource support.
**GAP-TOOLS-032**: MCP authentication. CC handles OAuth flows for MCP servers that require auth (e.g., Slack, Gmail, Calendar). No auth in babysitter's MCP.
User Interaction (1 CC tool)
| CC Tool | Babysitter Agentic Tool | Status | Gap? |
|---|---|---|---|
| AskUserQuestionTool | ask | **Partial** | **GAP-TOOLS-038** |
**GAP-TOOLS-038**: CC's AskUserQuestionTool is a single free-text question model. Babysitter's ask is a structured multi-question tool with options, multi-select, and recommended index. Different interaction models -- babysitter is richer but not a drop-in replacement for CC's simple question flow.
Planning (2 CC tools)
| CC Tool | Babysitter Agentic Tool | Status | Gap? |
|---|---|---|---|
| EnterPlanModeTool | (none) | Missing | **GAP-TOOLS-018** (exists) |
| ExitPlanModeTool | (none) | Missing | **GAP-TOOLS-018** (exists) |
Worktree (2 CC tools)
| CC Tool | Babysitter Agentic Tool | Status | Gap? |
|---|---|---|---|
| EnterWorktreeTool | (none) | Missing | **GAP-TOOLS-017** (exists) |
| ExitWorktreeTool | (none) | Missing | **GAP-TOOLS-017** (exists) |
Configuration (2 CC tools)
| CC Tool | Babysitter Agentic Tool | Status | Gap? |
|---|---|---|---|
| ConfigTool | (none) | Missing | **GAP-TOOLS-033** |
| ToolSearchTool | (none) | Missing | **GAP-TOOLS-034** |
**GAP-TOOLS-033**: Runtime configuration tool. CC can modify settings during a session. Babysitter's configure command exists but not as an agentic tool.
**GAP-TOOLS-034**: Tool discovery/search. CC has ToolSearchTool for deferred tool loading. Babysitter's Pi sessions have all tools loaded upfront -- no dynamic discovery. This becomes important with MCP tools.
Scheduling (2 CC tools)
| CC Tool | Babysitter Agentic Tool | Status | Gap? |
|---|---|---|---|
| ScheduleCronTool | (none) | Missing | **GAP-TOOLS-020** (exists) |
| RemoteTriggerTool | (none) | Missing | **GAP-TOOLS-021** (exists) |
CC-Specific (5 CC tools -- removed as host harness concerns)
| CC Tool | Babysitter Status | Reason |
|---|---|---|
| BriefTool | Removed | CC proactive mode feature |
| TodoWriteTool | Removed | CC in-session task list |
| MonitorTool | Removed | CC resource monitoring |
| TeamCreateTool | Removed | CC agent teams (covered by GAP-AGENT-001 sub-harness isolation) |
| TeamDeleteTool | Removed | CC agent teams (covered by GAP-AGENT-001 sub-harness isolation) |
Special (4 CC tools)
| CC Tool | Babysitter Agentic Tool | Status | Gap? |
|---|---|---|---|
| SkillTool | (none as agentic tool) | Missing | **GAP-TOOLS-027** (exists) |
| SleepTool | (ctx.sleepUntil effect) | Different model | **GAP-TOOLS-028** (exists) |
| LSPTool | (none) | Missing | **GAP-TOOLS-012** (LSP Integration for Code-Aware Routing) |
| DiscoverSkillsTool | (none) | Missing | **GAP-TOOLS-027** (Skill Discovery/Invocation from Processes) |
Babysitter-only tools (no CC equivalent)
| Babysitter Tool | Description |
|---|---|
browser | Headless browser via puppeteer -- navigate, click, type, evaluate, screenshot, close |
ssh | SSH remote execution |
calc | Calculator/math evaluation |
ast_grep | AST-aware code search |
ast_edit | AST-aware code transformation |
render_mermaid | Mermaid diagram rendering |
Summary
| Category | CC Tools | Babysitter Has | Gap Count |
|---|---|---|---|
| File ops | 6 | 6 (1 partial, 2 near-parity, 1 parity+, 2 parity) | 1 new (grep partial) |
| Execution | 3 | 2 (both partial) | 1 existing + 1 new |
| Web | 2 | 1 (partial) | 1 existing + 1 new |
| Agent/Delegation | 3 | 0 (different model) | 1 new |
| Task Management | 6 | 0 (CLI only) | 1 existing + 1 new |
| MCP | 4 | 0 | 1 existing + 2 new |
| User Interaction | 1 | 1 (partial) | 1 new |
| Planning | 2 | 0 | 1 existing |
| Worktree | 2 | 0 | 1 existing |
| Configuration | 2 | 0 | 2 new |
| Scheduling | 2 | 0 | 2 existing |
| CC-specific | 5 | N/A (removed) | 0 |
| Special | 4 | 0 (1 effect-based) | 3 existing (TOOLS-012, TOOLS-027, TOOLS-028) |
| **Total** | **42** | **16** | **All CC tools mapped to existing gaps** |
New Gaps to Create
- **GAP-TOOLS-007**: JS/TS REPL tool (CC's REPLTool equivalent)
- **GAP-TOOLS-029**: Structured output tool (CC's SyntheticOutputTool equivalent)
- **GAP-TOOLS-030**: Effect cancellation tool (CC's TaskStopTool equivalent)
- **GAP-TOOLS-031**: MCP resource browsing/reading tools
- **GAP-TOOLS-032**: MCP authentication tool
- **GAP-TOOLS-033**: Runtime configuration tool
- **GAP-TOOLS-034**: Dynamic tool discovery/search tool
- **GAP-TOOLS-035**: Grep output modes and context params (CC GrepTool parity)
- **GAP-TOOLS-036**: Bash background execution and description params (CC BashTool parity)
- **GAP-TOOLS-037**: Fetch content processing/prompt param (CC WebFetchTool parity)
- **GAP-TOOLS-038**: Ask tool interaction model alignment (CC AskUserQuestionTool parity)
Feature Gap Details for Partial-Parity Tools
This section details exactly what each partial-parity tool needs to reach full CC parity.
`grep` (GAP-TOOLS-035)
Current babysitter params: pattern, path, glob, type, i, context, limit, offset, multiline
Missing CC params:
- **
output_mode**: CC supportscontent(matching lines),files_with_matches(file paths only, default), andcount(match counts). Babysitter always returns content. - **
-B(before context)**: CC allows separate before-match context lines. Babysitter only has-Cequivalent (context). - **
-A(after context)**: CC allows separate after-match context lines. - **
-n(line numbers)**: CC has a toggle for line numbers in output (default true). Babysitter has no toggle. - **
head_limit**: CC defaults to 250 lines max output. Babysitter useslimit/offsetdifferently.
`bash` (GAP-TOOLS-036)
Current babysitter params: command, timeout, env, cwd
Missing CC params:
- **
run_in_background**: CC can run commands in the background and be notified on completion. Babysitter always runs synchronously. - **
description**: CC requires a human-readable description of what the command does. Babysitter has no equivalent.
Babysitter-only params (CC lacks):
- **
env**: Babysitter can pass environment variables to the command. - **
cwd**: Babysitter can set the working directory. CC infers from context.
`fetch` (GAP-TOOLS-037)
Current babysitter params: url, method, headers, body, timeout, raw
Missing CC params:
- **
prompt**: CC requires a prompt that describes how to process/summarize the fetched content before returning it. This is a fundamentally different approach -- CC returns AI-processed content, babysitter returns raw HTTP responses.
`read` (Near parity -- minor gap)
Current babysitter params: path, offset, limit
Missing CC params:
- **
pages**: CC supports apagesparam for reading specific page ranges from PDF files (e.g., "1-5", "10-20"). Babysitter has no PDF page selection.
CC uses absolute paths; babysitter uses workspace-relative paths (design choice, not a gap).
`ask` (GAP-TOOLS-038)
CC model: Single question, free-text response. Simple prompt-and-answer interaction.
Babysitter model: Structured multi-question with options (array of choices), multi_select (boolean), and recommended (index). Supports complex approval workflows.
These are fundamentally different interaction models. Babysitter's is richer but cannot replicate CC's simple free-text flow without ignoring its structured fields. Alignment may require supporting both modes or adding a mode parameter.