Chat Tools
What Critique Chat can look up, inspect, and do on your behalf — with confirmation for writes.
Critique Chat calls tools automatically based on your question. You do not pick tools manually; knowing what exists helps you ask better questions.
Write tools (start Remedy, create issue, post PR comment) always show a confirmation card before anything changes in GitHub or sandboxes.
Repository & GitHub
| Tool | Use when you want to… |
|---|---|
| searchRepository | Find files, symbols, or patterns in the connected repo |
| inspectPullRequest | See PR metadata, files, checks, and comment summaries |
| inspectWorkflowRun | Debug a CI run or job with log excerpts |
| commentOnPullRequest | Draft a PR comment (confirmed before post) |
| createGitHubIssue | Draft an issue from chat context (confirmed before post) |
Repository tools require a selected repo and GitHub App access.
Review & Remedy
| Tool | Use when you want to… |
|---|---|
| listRecentReviewRuns | See recent Critique reviews |
| inspectReviewRun | Deep-dive one run’s findings and Remedy state |
| inspectRemedyRuntime | Understand how Remedy executes in your environment |
| startRemedy | Queue Remedy for a run (confirmation required) |
| checkRemedyStatus | Poll an in-flight or finished Remedy run |
Workspace & memory
| Tool | Use when you want to… |
|---|---|
| getWorkspaceStatus | Check GitHub connection, installs, automation readiness |
| getActiveContext | Read current workspace / run context bundle |
| searchSharedMemory | Recall promoted learnings and board notes |
Memory search respects repository and run scope. See Memory & context.
Slack (Connections)
| Tool | Use when you want to… |
|---|---|
| postToSlack | Draft a channel message and confirm before posting to your connected workspace |
Requires Slack under Settings → Connections (bot token or incoming webhook). Write tool — always shows a confirmation card.
Linear & roadmap (Connections)
| Tool | Use when you want to… |
|---|---|
| searchIssuesAndRoadmap | Search Linear issues, assignees, and roadmap context tied to your question |
Requires an active Linear connection with chat enabled under Settings → Connections. If Linear is not connected, the tool is omitted from the agent registry.
External documentation
| Tool | Use when you want to… |
|---|---|
| searchLibraryDocs | Version-aware library docs (when Context7 is enabled) |
| searchDocs | Web/doc search via Firecrawl (when enabled) |
| fetchDocPage | Pull one doc page as markdown after a search hit |
Unavailable integrations return a clear error in chat — use repository search for code-local answers.
Agent handoff
| Tool | Use when you want to… |
|---|---|
| prepareClaudeCodeHandoff | Export the thread as Markdown for Claude Code (Anthropic key saved locally in this browser — Settings) |
Server-side BYOA vs chat handoff
Critique uses two different key models for external agents:
| Integration | Where the key lives | How you trigger it |
|---|---|---|
| Cursor Cloud Agent | Encrypted server-side — Settings → Cursor | Queue Cursor agent on a completed review run |
| Claude Managed Agent | Encrypted server-side — Settings → Anthropic | Queue Claude agent on a completed review run |
| OpenAI Codex | Encrypted server-side — Settings → OpenAI | Queue Codex run on a completed review run |
| Claude Code (chat) | Browser local storage only | prepareClaudeCodeHandoff in Critique Chat |
Server-side keys never return to the browser after save; they are used only when you queue a run. The chat handoff tool is for thread-scoped work — use review-run queueing for PR-scoped fixes.
For Copilot or agents without a queue integration, use Fix prompt on a review run. See BYOA. For OpenRouter-backed review, see BYOK. For MCP or REST access from an external agent, see Connections & Platform API.
Side effects
| Kind | Tools |
|---|---|
| Read-only | Search, inspect, status, memory, docs |
| Write (confirmed) | startRemedy, createGitHubIssue, commentOnPullRequest |
Promotion of review or Remedy outcomes into long-term memory uses internal tools invoked only after explicit operator intent — not from casual chat UI buttons.
Tips
- Name the PR number or review run when asking about a specific review.
- Ask for workspace status first if setup questions fail mysteriously.
- Prefer searchRepository before asking for file contents from memory — memory may be stale; code search is fresh.