Critique/docs
Reference

Chat Tools

Complete reference for every tool available in Critique Chat, including parameters, side effects, and availability conditions.

Critique Chat exposes a set of tools that ground conversational output in real data. The agent calls tools automatically based on the question, but understanding what each tool does helps you get better answers.

Tool categories

Tools are grouped by what they interact with:

CategoryTools
RepositorysearchRepository, inspectPullRequest, inspectWorkflowRun
External docssearchLibraryDocs, searchDocs, fetchDocPage
Workspace stategetWorkspaceStatus, getActiveContext, searchSharedMemory
Review runslistRecentReviewRuns, inspectReviewRun
RemedyinspectRemedyRuntime, startRemedy, checkRemedyStatus
Agent handoffprepareClaudeCodeHandoff
GitHub actionscreateGitHubIssue, commentOnPullRequest

searchRepository

Explore the selected GitHub repository using warm snapshot retrieval when available, with GitHub code search as fallback.

PropertyValue
Requires repositoryYes
Side effectsRead
RendererCard

Parameters

NameTypeRequiredDescription
querystringYesSearch query (min 1 char)

When to use: Before making repository-specific claims, tracing implementation details, or locating files and code paths.

When not to use: When no repository is selected or the answer is purely about product or runtime state.

Result shape: summary, relevant_files[], code_context[], search_notes, retrieval, execution


inspectPullRequest

Fetch a GitHub pull request: metadata, changed files, merge and check summary, and review thread context for the head commit.

PropertyValue
Requires repositoryYes
Side effectsRead
RendererCard
AvailabilityRequires GitHub App installation access

Parameters

NameTypeRequiredDefaultDescription
pullNumberintegerYesPR number (min 1)
includePatchbooleanNofalseInclude full patch content
maxFilesintegerNo30Max files to return (1–100)

When to use: When the user references a PR number, PR URL, or asks about changes, blockers, open comments, or CI status for a specific PR.

When not to use: For generic repo exploration without a PR — use searchRepository instead.

Result shape: pull, files[], issueCommentsSummary[], reviewCommentsSummary[], checksSummary, links[]


inspectWorkflowRun

Load a GitHub Actions workflow run or job: status, conclusion, associated commit, steps, and capped log excerpts.

PropertyValue
Requires repositoryYes
Side effectsRead
RendererCard
AvailabilityRequires GitHub App installation access

Parameters

NameTypeRequiredDefaultDescription
mode'run' or 'job'YesWhether to fetch a full run or a specific job
runIdintegerNoWorkflow run ID (min 1)
jobIdintegerNoJob ID (min 1)
logCharsPerJobintegerNo12000Max log chars per job (0–50000)
logMaxCharsintegerNo24000Global log char cap (0–100000)

When to use: When debugging CI failures, flaky tests, or deployment workflows tied to a known run or job ID.

When not to use: For source-code-only questions or when no run/job identifier is known.

Result shape: run, jobs[], logExcerpts[], artifactsSummary[]


searchLibraryDocs

Retrieve version-aware documentation snippets for a named library or framework.

PropertyValue
Requires repositoryNo
Side effectsRead
RendererCard
AvailabilityRequires Context7 integration

Parameters

NameTypeRequiredDefaultDescription
libraryNamestringYesLibrary or framework name (max 200 chars)
topicstringYesWhat to look up (max 2000 chars)
versionstringNoSpecific version to target (max 80 chars)

When to use: For framework or library API questions where generic web search is noisy or version-specific behavior matters.

When not to use: For arbitrary doc URLs or when the library docs integration is unavailable.

Result shape: libraryId, version, snippets[], sourceUrls[]


searchDocs

Search docs and web pages through Firecrawl and return relevant URLs for current documentation or vendor facts.

PropertyValue
Requires repositoryNo
Side effectsRead
RendererCard
AvailabilityRequires Firecrawl integration

Parameters

NameTypeRequiredDefaultDescription
querystringYesSearch query (min 1 char)
domainsstring[]NoRestrict search to these domains (max 6)
limitintegerNo5Max results (1–8)

When to use: For current external docs, vendor APIs, pricing, or product facts not guaranteed by the local repository.

When not to use: For repository-local answers or when Firecrawl is unavailable.

Result shape: hits[], creditsUsed


fetchDocPage

Fetch and scrape a specific documentation page through Firecrawl and return markdown content plus metadata.

PropertyValue
Requires repositoryNo
Side effectsRead
RendererCard
AvailabilityRequires Firecrawl integration

Parameters

NameTypeRequiredDescription
urlstring (URL)YesThe page URL to fetch

When to use: After a docs search when you need the full contents of a specific page before answering.

When not to use: Unless a specific allowed URL is already known.

Result shape: url, title, description, markdown


getWorkspaceStatus

Read GitHub connection, installation, repository, and automation readiness state for this user.

PropertyValue
Requires repositoryNo
Side effectsRead
RendererCard

Parameters: None.

When to use: For questions about setup, GitHub connectivity, installations, or review readiness.

When not to use: For repo code questions or external docs.

Result shape: githubLinked, installations, activeRepositories, reviewRuns, automationReady, nextStep


getActiveContext

Read the current durable context bundle for workspace, runtime, and repository state.

PropertyValue
Requires repositoryNo
Side effectsRead
RendererCard

Parameters

NameTypeRequiredDescription
reviewRunIdstringNoScope context to a specific review run

When to use: Before answering questions about current product truth, runtime state, or accumulated known context.

When not to use: As a replacement for repo search or docs search.

Result shape: summary, documents[]


searchSharedMemory

Search persisted board entries and memory chunks across chat, review runs, and repository memory.

PropertyValue
Requires repositoryNo
Side effectsRead
RendererCard

Parameters

NameTypeRequiredDefaultDescription
querystringYesSearch query (min 1 char)
reviewRunIdstringNoScope to a specific review run
limitintegerNo6Max results (1–10)

When to use: When the user asks about earlier findings, persisted learnings, or prior run context.

When not to use: For current repo implementation details when fresh repository search is needed.

Result shape: results[]


listRecentReviewRuns

List recent review runs visible to the current user.

PropertyValue
Requires repositoryNo
Side effectsRead
RendererCard

Parameters

NameTypeRequiredDefaultDescription
limitintegerNo5Max runs to return (1–8)

When to use: When the user asks about recent PR reviews or wants a run to inspect.

Result shape: reviewRun[]


inspectReviewRun

Inspect a specific review run, including findings and latest Remedy state.

PropertyValue
Requires repositoryNo
Side effectsRead
RendererCard

Parameters

NameTypeRequiredDescription
reviewRunIdstringYesThe review run ID to inspect

When to use: When the user names or selects a review run and needs its details.

When not to use: Without a run identifier.

Result shape: Review run snapshot with findings and Remedy state.


inspectRemedyRuntime

Explain how Remedy executes today and which runtime surface it uses.

PropertyValue
Requires repositoryNo
Side effectsRead
RendererCard

Parameters: None.

When to use: When the user asks about Remedy architecture or current execution behavior.

When not to use: To start Remedy — use startRemedy instead.

Result shape: Remedy runtime summary.


startRemedy

Queue a Remedy run for a specific review run. Shows a confirmation card before executing — it does not auto-start.

PropertyValue
Requires repositoryNo
Side effectsWrite
RendererCard

Parameters

NameTypeRequiredDescription
reviewRunIdstringNoReview run to remediate (uses selected repo if omitted)
suggestedInstructionsstringNoPre-fill the edit prompt in the confirmation dialog (max 1000 chars)

When to use: Only when the user explicitly asks to start Remedy.

When not to use: Speculatively or as part of ordinary information gathering.

Result shape: remedyRunId, status, executionSurface

Confirmation required

startRemedy emits a confirmation card. The user must approve before execution begins. No code is modified without explicit consent.


checkRemedyStatus

Check the current status of a running or completed Remedy run, including recent progress events.

PropertyValue
Requires repositoryNo
Side effectsRead
RendererCard

Parameters

NameTypeRequiredDescription
remedyRunIdstringYesThe Remedy run ID to check

When to use: When the user asks about the status of a Remedy run, or to check if a queued or running Remedy has completed.

When not to use: To start Remedy — use startRemedy instead.

Result shape: remedyRunId, status, recentEvents, commitUrl, streamUrl


prepareClaudeCodeHandoff

Compile the current Critique Chat thread into a detailed Markdown handoff and send it through the Anthropic Messages API using the operator's own API key.

PropertyValue
Requires repositoryNo
Side effectsRead
RendererCard
AvailabilityRequires user Anthropic API key in Settings

Parameters

NameTypeRequiredDefaultDescription
taskFocusstringNoExtra instructions or scope emphasis for the handoff (max 8000 chars)
maxTranscriptMessagesintegerNo48How many recent messages to include (4–120)

When to use: When the user explicitly wants Claude Code, a coding agent, or an autonomous fixer to take over.

When not to use: For ordinary Q&A, repo search, or when the user has not asked for an agent handoff.

Result shape: handoffMarkdown, anthropicModel, transcriptStats


createGitHubIssue

Draft a GitHub issue from the current chat context and show a confirmation card before posting.

PropertyValue
Requires repositoryYes
Side effectsWrite
RendererCard
AvailabilityRequires GitHub App installation access

Parameters

NameTypeRequiredDefaultDescription
titlestringYesIssue title (max 240 chars)
bodystringYesIssue body in Markdown (max 12000 chars)
labelsstring[]No[]Labels to apply (max 12, each max 60 chars)

When to use: When the user explicitly asks to open an issue, file a bug, or create a GitHub issue from this conversation.

When not to use: For ordinary summaries — do not auto-post without explicit user intent.

Confirmation required

Like startRemedy, this tool shows a confirmation card. The issue is only created after the user approves.


commentOnPullRequest

Draft a PR comment from the current chat context and show a confirmation card before posting.

PropertyValue
Requires repositoryYes
Side effectsWrite
RendererCard
AvailabilityRequires GitHub App installation access

Parameters

NameTypeRequiredDefaultDescription
pullNumberintegerYesPR number
bodystringYesComment body in Markdown (max 12000 chars)
pathstringNoFile path for line-specific comments (max 500 chars)
lineintegerNoLine number for line-specific comments
side'LEFT' or 'RIGHT'NoDiff side for line-specific comments
commitShastringNoCommit SHA (required for line-specific comments, min 7 chars)

Supports both general PR comments and file-line-specific review comments.

When to use: When the user explicitly asks to post findings or feedback onto a pull request from chat.

When not to use: For private notes, ordinary summaries, or speculative posting without explicit user intent.


Hidden tools

Two additional tools are always registered but do not appear in the tool manifest or UI:

  • promoteReviewRunLearning — Promote a review run into repository memory after a human explicitly decides the run contains reusable learning. Parameters: reviewRunId (required), rationale (required, min 8 chars).

  • promoteRemedyOutcome — Promote a completed Remedy outcome into repository memory when the result should inform future runs. Parameters: remedyRunId (required), rationale (optional, min 8 chars).

These tools write to repository memory and are not directly invocable through the chat UI.