chore: Update CHANGELOG.md

* Unify issue lifecycle labeling and sweep into a single system

Consolidate issue triage, stale detection, and lifecycle enforcement into
two components: a Claude-powered triage workflow and a unified sweep script.

Triage workflow changes:
- Add issue_comment trigger so Claude can re-evaluate lifecycle labels when
  someone responds to a needs-repro/needs-info issue
- Add concurrency group per issue with cancel-in-progress to avoid pile-up
- Filter out bot comments to prevent sweep/dedupe triggering re-triage
- Hardcode allowed label whitelist to prevent label sprawl (was discovering
  labels via gh label list, leading to junk variants like 'needs repro' vs
  'needs-repro')
- Replace MCP GitHub server with gh CLI — simpler, no Docker dependency,
  chaining is caught by the action so permissions are equivalent
- Add lifecycle labels (needs-repro, needs-info) for bugs missing info
- Add invalid label for off-topic issues (Claude API, billing, etc.)
- Add anti-patterns to prevent false positives (don't require specific
  format, model behavior issues don't need traditional repro, etc.)

Sweep script changes:
- Absorb stale issue detection (was separate stale-issue-manager workflow)
- Mark issues as stale after 14 days of inactivity
- Skip assigned issues (team is working on it internally)
- Skip enhancements with 10+ thumbs up (community wants it)
- Add invalid label with 3-day timeout
- Add autoclose label support to drain 200+ legacy issues
- Drop needs-votes (stale handles inactive enhancements)
- Unify close messages into a single template with per-label reasons
- Run 2x daily instead of once

Delete stale-issue-manager.yml — its logic is now in sweep.ts.

## Test plan

Dry-run sweep locally:
GITHUB_TOKEN=$(gh auth token) GITHUB_REPOSITORY_OWNER=anthropics   GITHUB_REPOSITORY_NAME=claude-code bun run scripts/sweep.ts --dry-run

Triage workflow will be tested by opening a test issue after merge.

* Update .github/workflows/claude-issue-triage.yml

Co-authored-by: Ashwin Bhat <ashwin@anthropic.com>

---------

Co-authored-by: Ashwin Bhat <ashwin@anthropic.com>

.github/workflows/claude-dedupe-issues.yml

@@ -17,16 +17,19 @@ jobs:
     permissions:
       contents: read
       issues: write
+      id-token: write
 
     steps:
       - name: Checkout repository
         uses: actions/checkout@v4
 
       - name: Run Claude Code slash command
-        uses: anthropics/claude-code-base-action@v1
+        uses: anthropics/claude-code-action@v1
         env:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         with:
+          github_token: ${{ secrets.GITHUB_TOKEN }}
+          allowed_non_write_users: "*"
           prompt: "/dedupe ${{ github.repository }}/issues/${{ github.event.issue.number || inputs.issue_number }}"
           anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
           claude_args: "--model claude-sonnet-4-5-20250929"

.github/workflows/claude-issue-triage.yml

@@ -1,108 +1,104 @@
 name: Claude Issue Triage
-description: Automatically triage GitHub issues using Claude Code
 on:
   issues:
     types: [opened]
+  issue_comment:
+    types: [created]
 
 jobs:
   triage-issue:
     runs-on: ubuntu-latest
     timeout-minutes: 10
+    if: >-
+      github.event_name == 'issues' ||
+      (github.event_name == 'issue_comment' && !github.event.issue.pull_request && github.event.comment.user.type != 'Bot')
+    concurrency:
+      group: issue-triage-${{ github.event.issue.number }}
+      cancel-in-progress: true
     permissions:
       contents: read
       issues: write
+      id-token: write
 
     steps:
       - name: Checkout repository
         uses: actions/checkout@v4
 
-      - name: Create triage prompt
-        run: |
-          mkdir -p /tmp/claude-prompts
-          cat > /tmp/claude-prompts/triage-prompt.txt << 'EOF'
-          You're an issue triage assistant for GitHub issues. Your task is to analyze the issue and select appropriate labels from the provided list.
-
-          IMPORTANT: Don't post any comments or messages to the issue. Your only action should be to apply labels.
-
-          Issue Information:
-          - REPO: ${{ github.repository }}
-          - ISSUE_NUMBER: ${{ github.event.issue.number }}
-
-          TASK OVERVIEW:
-
-          1. First, fetch the list of labels available in this repository by running: `gh label list`. Run exactly this command with nothing else.
-
-          2. Next, use the GitHub tools to get context about the issue:
-             - You have access to these tools:
-               - mcp__github__get_issue: Use this to retrieve the current issue's details including title, description, and existing labels
-               - mcp__github__get_issue_comments: Use this to read any discussion or additional context provided in the comments
-               - mcp__github__update_issue: Use this to apply labels to the issue (do not use this for commenting)
-               - mcp__github__search_issues: Use this to find similar issues that might provide context for proper categorization and to identify potential duplicate issues
-               - mcp__github__list_issues: Use this to understand patterns in how other issues are labeled
-             - Start by using mcp__github__get_issue to get the issue details
-
-          3. Analyze the issue content, considering:
-             - The issue title and description
-             - The type of issue (bug report, feature request, question, etc.)
-             - Technical areas mentioned
-             - Severity or priority indicators
-             - User impact
-             - Components affected
-
-          4. Select appropriate labels from the available labels list provided above:
-             - Choose labels that accurately reflect the issue's nature
-             - Be specific but comprehensive
-             - Select priority labels if you can determine urgency (high-priority, med-priority, or low-priority)
-             - Consider platform labels (android, ios) if applicable
-             - If you find similar issues using mcp__github__search_issues, consider using a "duplicate" label if appropriate. Only do so if the issue is a duplicate of another OPEN issue.
-
-          5. Apply the selected labels:
-             - Use mcp__github__update_issue to apply your selected labels
-             - DO NOT post any comments explaining your decision
-             - DO NOT communicate directly with users
-             - If no labels are clearly applicable, do not apply any labels
-
-          IMPORTANT GUIDELINES:
-          - Be thorough in your analysis
-          - Only select labels from the provided list above
-          - DO NOT post any comments to the issue
-          - Your ONLY action should be to apply labels using mcp__github__update_issue
-          - It's okay to not add any labels if none are clearly applicable
-          EOF
-
-      - name: Setup GitHub MCP Server
-        run: |
-          mkdir -p /tmp/mcp-config
-          cat > /tmp/mcp-config/mcp-servers.json << 'EOF'
-          {
-            "mcpServers": {
-              "github": {
-                "command": "docker",
-                "args": [
-                  "run",
-                  "-i",
-                  "--rm",
-                  "-e",
-                  "GITHUB_PERSONAL_ACCESS_TOKEN",
-                  "ghcr.io/github/github-mcp-server:sha-7aced2b"
-                ],
-                "env": {
-                  "GITHUB_PERSONAL_ACCESS_TOKEN": "${{ secrets.GITHUB_TOKEN }}"
-                }
-              }
-            }
-          }
-          EOF
-
       - name: Run Claude Code for Issue Triage
         timeout-minutes: 5
-        uses: anthropics/claude-code-base-action@v1
+        uses: anthropics/claude-code-action@v1
         env:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         with:
-          prompt_file: /tmp/claude-prompts/triage-prompt.txt
+          github_token: ${{ secrets.GITHUB_TOKEN }}
+          allowed_non_write_users: "*"
+          prompt: |
+            You're an issue triage assistant. Analyze the issue and manage labels.
+
+            IMPORTANT: Don't post any comments or messages to the issue. Your only actions are adding or removing labels.
+
+            Context:
+            - REPO: ${{ github.repository }}
+            - ISSUE_NUMBER: ${{ github.event.issue.number }}
+            - EVENT: ${{ github.event_name }}
+
+            ALLOWED LABELS — you may ONLY use labels from this list. Never invent new labels.
+
+            Type: bug, enhancement, question, documentation, duplicate, invalid
+            Lifecycle: needs-repro, needs-info
+            Platform: platform:linux, platform:macos, platform:windows, platform:wsl, platform:ios, platform:android, platform:vscode, platform:intellij, platform:web, platform:aws-bedrock
+            API: api:bedrock, api:vertex
+
+            TOOLS:
+            - `gh issue view NUMBER`: Read the issue title, body, and labels
+            - `gh issue view NUMBER --comments`: Read the conversation
+            - `gh search issues QUERY`: Find similar or duplicate issues
+            - `gh issue edit NUMBER --add-label` / `--remove-label`: Add or remove labels
+
+            TASK:
+
+            1. Run `gh issue view ${{ github.event.issue.number }}` to read the issue details.
+            2. Run `gh issue view ${{ github.event.issue.number }} --comments` to read the conversation.
+
+            **If EVENT is "issues" (new issue):**
+
+            3. First, check if this issue is actually about Claude Code (the CLI/IDE tool). Issues about the Claude API, claude.ai, the Claude app, Anthropic billing, or other Anthropic products should be labeled `invalid`. If invalid, apply only that label and stop.
+
+            4. Analyze and apply category labels:
+               - Type (bug, enhancement, question, etc.)
+               - Technical areas and platform
+               - Check for duplicates with `gh search issues`. Only mark as duplicate of OPEN issues.
+
+            5. Evaluate lifecycle labels:
+               - `needs-repro` (bugs only, 7 days): Bug reports without clear steps to reproduce. A good repro has specific, followable steps that someone else could use to see the same issue.
+                 Do NOT apply if the user already provided error messages, logs, file paths, or a description of what they did. Don't require a specific format — narrative descriptions count.
+                 For model behavior issues (e.g. "Claude does X when it should do Y"), don't require traditional repro steps — examples and patterns are sufficient.
+               - `needs-info` (bugs only, 7 days): The issue needs something from the community before it can progress — e.g. error messages, versions, environment details, or answers to follow-up questions. Don't apply to questions or enhancements.
+                 Do NOT apply if the user already provided version, environment, and error details. If the issue just needs engineering investigation, that's not `needs-info`.
+
+               Issues with these labels are automatically closed after the timeout if there's no response.
+               The goal is to avoid issues lingering without a clear next step.
+
+            6. Apply all selected labels:
+               `gh issue edit ${{ github.event.issue.number }} --add-label "label1" --add-label "label2"`
+
+            **If EVENT is "issue_comment" (comment on existing issue):**
+
+            3. Evaluate lifecycle labels based on the full conversation:
+               - If the issue has `needs-repro` or `needs-info` and the missing information has now been provided, remove the label:
+                 `gh issue edit ${{ github.event.issue.number }} --remove-label "needs-repro"`
+               - If the issue doesn't have lifecycle labels but clearly needs them (e.g., a maintainer asked for repro steps or more details), add the appropriate label.
+               - Comments like "+1", "me too", "same here", or emoji reactions are NOT the missing information. Only remove labels when substantive details are actually provided.
+               - Do NOT add or remove category labels (bug, enhancement, etc.) on comment events.
+
+            GUIDELINES:
+            - ONLY use labels from the ALLOWED LABELS list above — never create or guess label names
+            - DO NOT post any comments to the issue
+            - Be conservative with lifecycle labels — only apply when clearly warranted
+            - Only apply lifecycle labels (`needs-repro`, `needs-info`) to bugs — never to questions or enhancements
+            - When in doubt, don't apply a lifecycle label — false positives are worse than missing labels
+            - It's okay to not add any labels if none are clearly applicable
           anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
           claude_args: |
-            --model claude-sonnet-4-5-20250929
-            --mcp-config /tmp/mcp-config/mcp-servers.json
-            --allowedTools "Bash(gh label list),mcp__github__get_issue,mcp__github__get_issue_comments,mcp__github__update_issue,mcp__github__search_issues,mcp__github__list_issues"
+            --model claude-opus-4-6
+            --allowedTools "Bash(gh issue view:*),Bash(gh issue edit:*),Bash(gh search issues:*)"

.github/workflows/oncall-triage.yml

@@ -16,74 +16,12 @@ jobs:
     permissions:
       contents: read
       issues: write
+      id-token: write
 
     steps:
       - name: Checkout repository
         uses: actions/checkout@v4
 
-      - name: Create oncall triage prompt
-        run: |
-          mkdir -p /tmp/claude-prompts
-          cat > /tmp/claude-prompts/oncall-triage-prompt.txt << 'EOF'
-          You're an oncall triage assistant for GitHub issues. Your task is to identify critical issues that require immediate oncall attention.
-
-          Important: Don't post any comments or messages to the issues. Your only action should be to apply the "oncall" label to qualifying issues.
-
-          Repository: ${{ github.repository }}
-
-          Task overview:
-          1. Fetch all open issues updated in the last 3 days:
-             - Use mcp__github__list_issues with:
-               - state="open"
-               - first=5 (fetch only 5 issues per page)
-               - orderBy="UPDATED_AT"
-               - direction="DESC"
-             - This will give you the most recently updated issues first
-             - For each page of results, check the updatedAt timestamp of each issue
-             - Add issues updated within the last 3 days (72 hours) to your TODO list as you go
-             - Keep paginating using the 'after' parameter until you encounter issues older than 3 days
-             - Once you hit issues older than 3 days, you can stop fetching (no need to fetch all open issues)
-
-          2. Build your TODO list incrementally as you fetch:
-             - As you fetch each page, immediately add qualifying issues to your TODO list
-             - One TODO item per issue number (e.g., "Evaluate issue #123")
-             - This allows you to start processing while still fetching more pages
-
-          3. For each issue in your TODO list:
-             - Use mcp__github__get_issue to read the issue details (title, body, labels)
-             - Use mcp__github__get_issue_comments to read all comments
-             - Evaluate whether this issue needs the oncall label:
-               a) Is it a bug? (has "bug" label or describes bug behavior)
-               b) Does it have at least 50 engagements? (count comments + reactions)
-               c) Is it truly blocking? Read and understand the full content to determine:
-                  - Does this prevent core functionality from working?
-                  - Can users work around it?
-                  - Consider severity indicators: "crash", "stuck", "frozen", "hang", "unresponsive", "cannot use", "blocked", "broken"
-                  - Be conservative - only flag issues that truly prevent users from getting work done
-
-          4. For issues that meet all criteria and do not already have the "oncall" label:
-             - Use mcp__github__update_issue to add the "oncall" label
-             - Do not post any comments
-             - Do not remove any existing labels
-             - Do not remove the "oncall" label from issues that already have it
-
-          Important guidelines:
-          - Use the TODO list to track your progress through ALL candidate issues
-          - Process issues efficiently - don't read every single issue upfront, work through your TODO list systematically
-          - Be conservative in your assessment - only flag truly critical blocking issues
-          - Do not post any comments to issues
-          - Your only action should be to add the "oncall" label using mcp__github__update_issue
-          - Mark each issue as complete in your TODO list as you process it
-
-          7. After processing all issues in your TODO list, provide a summary of your actions:
-             - Total number of issues processed (candidate issues evaluated)
-             - Number of issues that received the "oncall" label
-             - For each issue that got the label: list issue number, title, and brief reason why it qualified
-             - Close calls: List any issues that almost qualified but didn't quite meet the criteria (e.g., borderline blocking, had workarounds)
-             - If no issues qualified, state that clearly
-             - Format the summary clearly for easy reading
-          EOF
-
       - name: Setup GitHub MCP Server
         run: |
           mkdir -p /tmp/mcp-config
@@ -110,11 +48,70 @@ jobs:
 
       - name: Run Claude Code for Oncall Triage
         timeout-minutes: 10
-        uses: anthropics/claude-code-base-action@v1
+        uses: anthropics/claude-code-action@v1
         env:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         with:
-          prompt_file: /tmp/claude-prompts/oncall-triage-prompt.txt
+          github_token: ${{ secrets.GITHUB_TOKEN }}
+          allowed_non_write_users: "*"
+          prompt: |
+            You're an oncall triage assistant for GitHub issues. Your task is to identify critical issues that require immediate oncall attention.
+
+            Important: Don't post any comments or messages to the issues. Your only action should be to apply the "oncall" label to qualifying issues.
+
+            Repository: ${{ github.repository }}
+
+            Task overview:
+            1. Fetch all open issues updated in the last 3 days:
+               - Use mcp__github__list_issues with:
+                 - state="open"
+                 - first=5 (fetch only 5 issues per page)
+                 - orderBy="UPDATED_AT"
+                 - direction="DESC"
+               - This will give you the most recently updated issues first
+               - For each page of results, check the updatedAt timestamp of each issue
+               - Add issues updated within the last 3 days (72 hours) to your TODO list as you go
+               - Keep paginating using the 'after' parameter until you encounter issues older than 3 days
+               - Once you hit issues older than 3 days, you can stop fetching (no need to fetch all open issues)
+
+            2. Build your TODO list incrementally as you fetch:
+               - As you fetch each page, immediately add qualifying issues to your TODO list
+               - One TODO item per issue number (e.g., "Evaluate issue #123")
+               - This allows you to start processing while still fetching more pages
+
+            3. For each issue in your TODO list:
+               - Use mcp__github__get_issue to read the issue details (title, body, labels)
+               - Use mcp__github__get_issue_comments to read all comments
+               - Evaluate whether this issue needs the oncall label:
+                 a) Is it a bug? (has "bug" label or describes bug behavior)
+                 b) Does it have at least 50 engagements? (count comments + reactions)
+                 c) Is it truly blocking? Read and understand the full content to determine:
+                    - Does this prevent core functionality from working?
+                    - Can users work around it?
+                    - Consider severity indicators: "crash", "stuck", "frozen", "hang", "unresponsive", "cannot use", "blocked", "broken"
+                    - Be conservative - only flag issues that truly prevent users from getting work done
+
+            4. For issues that meet all criteria and do not already have the "oncall" label:
+               - Use mcp__github__update_issue to add the "oncall" label
+               - Do not post any comments
+               - Do not remove any existing labels
+               - Do not remove the "oncall" label from issues that already have it
+
+            Important guidelines:
+            - Use the TODO list to track your progress through ALL candidate issues
+            - Process issues efficiently - don't read every single issue upfront, work through your TODO list systematically
+            - Be conservative in your assessment - only flag truly critical blocking issues
+            - Do not post any comments to issues
+            - Your only action should be to add the "oncall" label using mcp__github__update_issue
+            - Mark each issue as complete in your TODO list as you process it
+
+            7. After processing all issues in your TODO list, provide a summary of your actions:
+               - Total number of issues processed (candidate issues evaluated)
+               - Number of issues that received the "oncall" label
+               - For each issue that got the label: list issue number, title, and brief reason why it qualified
+               - Close calls: List any issues that almost qualified but didn't quite meet the criteria (e.g., borderline blocking, had workarounds)
+               - If no issues qualified, state that clearly
+               - Format the summary clearly for easy reading
           anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
           claude_args: |
             --mcp-config /tmp/mcp-config/mcp-servers.json

.github/workflows/stale-issue-manager.yml

@@ -1,157 +0,0 @@
-name: "Manage Stale Issues"
-
-on:
-  schedule:
-    # 2am Pacific = 9am UTC (10am UTC during DST)
-    - cron: "0 10 * * *"
-  workflow_dispatch:
-
-permissions:
-  issues: write
-
-concurrency:
-  group: stale-issue-manager
-
-jobs:
-  manage-stale-issues:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Manage stale issues
-        uses: actions/github-script@v7
-        with:
-          script: |
-            const oneMonthAgo = new Date();
-            oneMonthAgo.setDate(oneMonthAgo.getDate() - 30);
-
-            const twoMonthsAgo = new Date();
-            twoMonthsAgo.setDate(twoMonthsAgo.getDate() - 60);
-
-            const warningComment = `This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.`;
-
-            const closingComment = `This issue has been automatically closed due to 60 days of inactivity. If you're still experiencing this issue, please open a new issue with updated information.`;
-
-            let page = 1;
-            let hasMore = true;
-            let totalWarned = 0;
-            let totalClosed = 0;
-            let totalLabeled = 0;
-
-            while (hasMore) {
-              // Get open issues sorted by last updated (oldest first)
-              const { data: issues } = await github.rest.issues.listForRepo({
-                owner: context.repo.owner,
-                repo: context.repo.repo,
-                state: 'open',
-                sort: 'updated',
-                direction: 'asc',
-                per_page: 100,
-                page: page
-              });
-              
-              if (issues.length === 0) {
-                hasMore = false;
-                break;
-              }
-              
-              for (const issue of issues) {
-                // Skip if already locked
-                if (issue.locked) continue;
-                
-                // Skip pull requests
-                if (issue.pull_request) continue;
-                
-                // Check if updated more recently than 30 days ago
-                const updatedAt = new Date(issue.updated_at);
-                if (updatedAt > oneMonthAgo) {
-                  // Since issues are sorted by updated_at ascending, 
-                  // once we hit a recent issue, all remaining will be recent too
-                  hasMore = false;
-                  break;
-                }
-                
-                // Check if issue has autoclose label
-                const hasAutocloseLabel = issue.labels.some(label => 
-                  typeof label === 'object' && label.name === 'autoclose'
-                );
-                
-                try {
-                  // Get comments to check for existing warning
-                  const { data: comments } = await github.rest.issues.listComments({
-                    owner: context.repo.owner,
-                    repo: context.repo.repo,
-                    issue_number: issue.number,
-                    per_page: 100
-                  });
-                  
-                  // Find the last comment from github-actions bot
-                  const botComments = comments.filter(comment => 
-                    comment.user && comment.user.login === 'github-actions[bot]' &&
-                    comment.body && comment.body.includes('inactive for 30 days')
-                  );
-                  
-                  const lastBotComment = botComments[botComments.length - 1];
-                  
-                  if (lastBotComment) {
-                    // Check if the bot comment is older than 30 days (total 60 days of inactivity)
-                    const botCommentDate = new Date(lastBotComment.created_at);
-                    if (botCommentDate < oneMonthAgo) {
-                      // Close the issue - it's been stale for 60+ days
-                      console.log(`Closing issue #${issue.number} (stale for 60+ days): ${issue.title}`);
-                      
-                      // Post closing comment
-                      await github.rest.issues.createComment({
-                        owner: context.repo.owner,
-                        repo: context.repo.repo,
-                        issue_number: issue.number,
-                        body: closingComment
-                      });
-                      
-                      // Close the issue
-                      await github.rest.issues.update({
-                        owner: context.repo.owner,
-                        repo: context.repo.repo,
-                        issue_number: issue.number,
-                        state: 'closed',
-                        state_reason: 'not_planned'
-                      });
-                      
-                      totalClosed++;
-                    }
-                    // If bot comment exists but is recent, issue already has warning
-                  } else if (updatedAt < oneMonthAgo) {
-                    // No bot warning yet, issue is 30+ days old
-                    console.log(`Warning issue #${issue.number} (stale for 30+ days): ${issue.title}`);
-                    
-                    // Post warning comment
-                    await github.rest.issues.createComment({
-                      owner: context.repo.owner,
-                      repo: context.repo.repo,
-                      issue_number: issue.number,
-                      body: warningComment
-                    });
-                    
-                    totalWarned++;
-                    
-                    // Add autoclose label if not present
-                    if (!hasAutocloseLabel) {
-                      await github.rest.issues.addLabels({
-                        owner: context.repo.owner,
-                        repo: context.repo.repo,
-                        issue_number: issue.number,
-                        labels: ['autoclose']
-                      });
-                      totalLabeled++;
-                    }
-                  }
-                } catch (error) {
-                  console.error(`Failed to process issue #${issue.number}: ${error.message}`);
-                }
-              }
-              
-              page++;
-            }
-
-            console.log(`Summary:`);
-            console.log(`- Issues warned (30 days stale): ${totalWarned}`);
-            console.log(`- Issues labeled with autoclose: ${totalLabeled}`);
-            console.log(`- Issues closed (60 days stale): ${totalClosed}`);

.github/workflows/sweep.yml

@@ -0,0 +1,31 @@
+name: "Issue Sweep"
+
+on:
+  schedule:
+    - cron: "0 10,22 * * *"
+  workflow_dispatch:
+
+permissions:
+  issues: write
+
+concurrency:
+  group: daily-issue-sweep
+
+jobs:
+  sweep:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Setup Bun
+        uses: oven-sh/setup-bun@v2
+        with:
+          bun-version: latest
+
+      - name: Enforce lifecycle timeouts
+        run: bun run scripts/sweep.ts
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          GITHUB_REPOSITORY_OWNER: ${{ github.repository_owner }}
+          GITHUB_REPOSITORY_NAME: ${{ github.event.repository.name }}

CHANGELOG.md

@@ -1,5 +1,310 @@
 # Changelog
 
+## 2.1.41
+
+- Fixed AWS auth refresh hanging indefinitely by adding a 3-minute timeout
+- Added `claude auth login`, `claude auth status`, and `claude auth logout` CLI subcommands
+- Added Windows ARM64 (win32-arm64) native binary support
+- Improved `/rename` to auto-generate session name from conversation context when called without arguments
+- Improved narrow terminal layout for prompt footer
+- Fixed file resolution failing for @-mentions with anchor fragments (e.g., `@README.md#installation`)
+- Fixed FileReadTool blocking the process on FIFOs, `/dev/stdin`, and large files
+- Fixed background task notifications not being delivered in streaming Agent SDK mode
+- Fixed cursor jumping to end on each keystroke in classifier rule input
+- Fixed markdown link display text being dropped for raw URL
+- Fixed auto-compact failure error notifications being shown to users
+- Fixed permission wait time being included in subagent elapsed time display
+- Fixed proactive ticks firing while in plan mode
+- Fixed clear stale permission rules when settings change on disk
+- Fixed hook blocking errors showing stderr content in UI
+
+## 2.1.39
+
+- Added guard against launching Claude Code inside another Claude Code session
+- Fixed Agent Teams using wrong model identifier for Bedrock, Vertex, and Foundry customers
+- Fixed a crash when MCP tools return image content during streaming
+- Fixed /resume session previews showing raw XML tags instead of readable command names
+- Improved model error messages for Bedrock/Vertex/Foundry users with fallback suggestions
+- Fixed plugin browse showing misleading "Space to Toggle" hint for already-installed plugins
+- Fixed hook blocking errors (exit code 2) not showing stderr to the user
+- Added `speed` attribute to OTel events and trace spans for fast mode visibility
+- Fixed /resume showing interrupt messages as session titles
+- Fixed Opus 4.6 launch announcement showing for Bedrock/Vertex/Foundry users
+- Improved error message for many-image dimension limit errors with /compact suggestion
+- Fixed structured-outputs beta header being sent unconditionally on Vertex/Bedrock
+- Fixed spurious warnings for non-agent markdown files in `.claude/agents/` directory
+- Improved terminal rendering performance
+- Fixed fatal errors being swallowed instead of displayed
+- Fixed process hanging after session close
+- Fixed character loss at terminal screen boundary
+- Fixed blank lines in verbose transcript view
+
+## 2.1.38
+
+- Fixed VS Code terminal scroll-to-top regression introduced in 2.1.37
+- Fixed Tab key queueing slash commands instead of autocompleting
+- Fixed bash permission matching for commands using environment variable wrappers
+- Fixed text between tool uses disappearing when not using streaming
+- Fixed duplicate sessions when resuming in VS Code extension
+- Improved heredoc delimiter parsing to prevent command smuggling
+- Blocked writes to `.claude/skills` directory in sandbox mode
+
+## 2.1.37
+
+- Fixed an issue where /fast was not immediately available after enabling /extra-usage
+
+## 2.1.36
+
+- Fast mode is now available for Opus 4.6. Learn more at https://code.claude.com/docs/en/fast-mode
+
+## 2.1.34
+
+- Fixed a crash when agent teams setting changed between renders
+- Fixed a bug where commands excluded from sandboxing (via `sandbox.excludedCommands` or `dangerouslyDisableSandbox`) could bypass the Bash ask permission rule when `autoAllowBashIfSandboxed` was enabled
+
+## 2.1.33
+
+- Fixed agent teammate sessions in tmux to send and receive messages
+- Fixed warnings about agent teams not being available on your current plan
+- Added `TeammateIdle` and `TaskCompleted` hook events for multi-agent workflows
+- Added support for restricting which sub-agents can be spawned via `Task(agent_type)` syntax in agent "tools" frontmatter
+- Added `memory` frontmatter field support for agents, enabling persistent memory with `user`, `project`, or `local` scope
+- Added plugin name to skill descriptions and `/skills` menu for better discoverability
+- Fixed an issue where submitting a new message while the model was in extended thinking would interrupt the thinking phase
+- Fixed an API error that could occur when aborting mid-stream, where whitespace text combined with a thinking block would bypass normalization and produce an invalid request
+- Fixed API proxy compatibility issue where 404 errors on streaming endpoints no longer triggered non-streaming fallback
+- Fixed an issue where proxy settings configured via `settings.json` environment variables were not applied to WebFetch and other HTTP requests on the Node.js build
+- Fixed `/resume` session picker showing raw XML markup instead of clean titles for sessions started with slash commands
+- Improved error messages for API connection failures — now shows specific cause (e.g., ECONNREFUSED, SSL errors) instead of generic "Connection error"
+- Errors from invalid managed settings are now surfaced
+- VSCode: Added support for remote sessions, allowing OAuth users to browse and resume sessions from claude.ai
+- VSCode: Added git branch and message count to the session picker, with support for searching by branch name
+- VSCode: Fixed scroll-to-bottom under-scrolling on initial session load and session switch
+
+## 2.1.32
+
+- Claude Opus 4.6 is now available!
+- Added research preview agent teams feature for multi-agent collaboration (token-intensive feature, requires setting CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1)
+- Claude now automatically records and recalls memories as it works
+- Added "Summarize from here" to the message selector, allowing partial conversation summarization.
+- Skills defined in `.claude/skills/` within additional directories (`--add-dir`) are now loaded automatically.
+- Fixed `@` file completion showing incorrect relative paths when running from a subdirectory
+- Updated --resume to re-use --agent value specified in previous conversation by default.
+- Fixed: Bash tool no longer throws "Bad substitution" errors when heredocs contain JavaScript template literals like `${index + 1}`, which previously interrupted tool execution
+- Skill character budget now scales with context window (2% of context), so users with larger context windows can see more skill descriptions without truncation
+- Fixed Thai/Lao spacing vowels (สระ า, ำ) not rendering correctly in the input field
+- VSCode: Fixed slash commands incorrectly being executed when pressing Enter with preceding text in the input field
+- VSCode: Added spinner when loading past conversations list
+
+## 2.1.31
+
+- Added session resume hint on exit, showing how to continue your conversation later
+- Added support for full-width (zenkaku) space input from Japanese IME in checkbox selection
+- Fixed PDF too large errors permanently locking up sessions, requiring users to start a new conversation
+- Fixed bash commands incorrectly reporting failure with "Read-only file system" errors when sandbox mode was enabled
+- Fixed a crash that made sessions unusable after entering plan mode when project config in `~/.claude.json` was missing default fields
+- Fixed `temperatureOverride` being silently ignored in the streaming API path, causing all streaming requests to use the default temperature (1) regardless of the configured override
+- Fixed LSP shutdown/exit compatibility with strict language servers that reject null params
+- Improved system prompts to more clearly guide the model toward using dedicated tools (Read, Edit, Glob, Grep) instead of bash equivalents (`cat`, `sed`, `grep`, `find`), reducing unnecessary bash command usage
+- Improved PDF and request size error messages to show actual limits (100 pages, 20MB)
+- Reduced layout jitter in the terminal when the spinner appears and disappears during streaming
+- Removed misleading Anthropic API pricing from model selector for third-party provider (Bedrock, Vertex, Foundry) users
+
+## 2.1.30
+
+- Added `pages` parameter to the Read tool for PDFs, allowing specific page ranges to be read (e.g., `pages: "1-5"`). Large PDFs (>10 pages) now return a lightweight reference when `@` mentioned instead of being inlined into context.
+- Added pre-configured OAuth client credentials for MCP servers that don't support Dynamic Client Registration (e.g., Slack). Use `--client-id` and `--client-secret` with `claude mcp add`.
+- Added `/debug` for Claude to help troubleshoot the current session
+- Added support for additional `git log` and `git show` flags in read-only mode (e.g., `--topo-order`, `--cherry-pick`, `--format`, `--raw`)
+- Added token count, tool uses, and duration metrics to Task tool results
+- Added reduced motion mode to the config
+- Fixed phantom "(no content)" text blocks appearing in API conversation history, reducing token waste and potential model confusion
+- Fixed prompt cache not correctly invalidating when tool descriptions or input schemas changed, only when tool names changed
+- Fixed 400 errors that could occur after running `/login` when the conversation contained thinking blocks
+- Fixed a hang when resuming sessions with corrupted transcript files containing `parentUuid` cycles
+- Fixed rate limit message showing incorrect "/upgrade" suggestion for Max 20x users when extra-usage is unavailable
+- Fixed permission dialogs stealing focus while actively typing
+- Fixed subagents not being able to access SDK-provided MCP tools because they were not synced to the shared application state
+- Fixed a regression where Windows users with a `.bashrc` file could not run bash commands
+- Improved memory usage for `--resume` (68% reduction for users with many sessions) by replacing the session index with lightweight stat-based loading and progressive enrichment
+- Improved `TaskStop` tool to display the stopped command/task description in the result line instead of a generic "Task stopped" message
+- Changed `/model` to execute immediately instead of being queued
+- [VSCode] Added multiline input support to the "Other" text input in question dialogs (use Shift+Enter for new lines)
+- [VSCode] Fixed duplicate sessions appearing in the session list when starting a new conversation
+
+## 2.1.29
+
+- Fixed startup performance issues when resuming sessions that have `saved_hook_context`
+
+## 2.1.27
+
+- Added tool call failures and denials to debug logs
+- Fixed context management validation error for gateway users, ensuring `CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1` avoids the error
+- Added `--from-pr` flag to resume sessions linked to a specific GitHub PR number or URL
+- Sessions are now automatically linked to PRs when created via `gh pr create`
+- Fixed /context command not displaying colored output
+- Fixed status bar duplicating background task indicator when PR status was shown
+- Windows: Fixed bash command execution failing for users with `.bashrc` files
+- Windows: Fixed console windows flashing when spawning child processes
+- VSCode: Fixed OAuth token expiration causing 401 errors after extended sessions
+
+## 2.1.25
+
+- Fixed beta header validation error for gateway users on Bedrock and Vertex, ensuring `CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1` avoids the error
+
+## 2.1.23
+
+- Added customizable spinner verbs setting (`spinnerVerbs`)
+- Fixed mTLS and proxy connectivity for users behind corporate proxies or using client certificates
+- Fixed per-user temp directory isolation to prevent permission conflicts on shared systems
+- Fixed a race condition that could cause 400 errors when prompt caching scope was enabled
+- Fixed pending async hooks not being cancelled when headless streaming sessions ended
+- Fixed tab completion not updating the input field when accepting a suggestion
+- Fixed ripgrep search timeouts silently returning empty results instead of reporting errors
+- Improved terminal rendering performance with optimized screen data layout
+- Changed Bash commands to show timeout duration alongside elapsed time
+- Changed merged pull requests to show a purple status indicator in the prompt footer
+- [IDE] Fixed model options displaying incorrect region strings for Bedrock users in headless mode
+
+## 2.1.22
+
+- Fixed structured outputs for non-interactive (-p) mode
+
+## 2.1.21
+
+- Added support for full-width (zenkaku) number input from Japanese IME in option selection prompts
+- Fixed shell completion cache files being truncated on exit
+- Fixed API errors when resuming sessions that were interrupted during tool execution
+- Fixed auto-compact triggering too early on models with large output token limits
+- Fixed task IDs potentially being reused after deletion
+- Fixed file search not working in VS Code extension on Windows
+- Improved read/search progress indicators to show "Reading…" while in progress and "Read" when complete
+- Improved Claude to prefer file operation tools (Read, Edit, Write) over bash equivalents (cat, sed, awk)
+- [VSCode] Added automatic Python virtual environment activation, ensuring `python` and `pip` commands use the correct interpreter (configurable via `claudeCode.usePythonEnvironment` setting)
+- [VSCode] Fixed message action buttons having incorrect background colors
+
+## 2.1.20
+
+- Added arrow key history navigation in vim normal mode when cursor cannot move further
+- Added external editor shortcut (Ctrl+G) to the help menu for better discoverability
+- Added PR review status indicator to the prompt footer, showing the current branch's PR state (approved, changes requested, pending, or draft) as a colored dot with a clickable link
+- Added support for loading `CLAUDE.md` files from additional directories specified via `--add-dir` flag (requires setting `CLAUDE_CODE_ADDITIONAL_DIRECTORIES_CLAUDE_MD=1`)
+- Added ability to delete tasks via the `TaskUpdate` tool
+- Fixed session compaction issues that could cause resume to load full history instead of the compact summary
+- Fixed agents sometimes ignoring user messages sent while actively working on a task
+- Fixed wide character (emoji, CJK) rendering artifacts where trailing columns were not cleared when replaced by narrower characters
+- Fixed JSON parsing errors when MCP tool responses contain special Unicode characters
+- Fixed up/down arrow keys in multi-line and wrapped text input to prioritize cursor movement over history navigation
+- Fixed draft prompt being lost when pressing UP arrow to navigate command history
+- Fixed ghost text flickering when typing slash commands mid-input
+- Fixed marketplace source removal not properly deleting settings
+- Fixed duplicate output in some commands like `/context`
+- Fixed task list sometimes showing outside the main conversation view
+- Fixed syntax highlighting for diffs occurring within multiline constructs like Python docstrings
+- Fixed crashes when cancelling tool use
+- Improved `/sandbox` command UI to show dependency status with installation instructions when dependencies are missing
+- Improved thinking status text with a subtle shimmer animation
+- Improved task list to dynamically adjust visible items based on terminal height
+- Improved fork conversation hint to show how to resume the original session
+- Changed collapsed read/search groups to show present tense ("Reading", "Searching for") while in progress, and past tense ("Read", "Searched for") when complete
+- Changed `ToolSearch` results to appear as a brief notification instead of inline in the conversation
+- Changed the `/commit-push-pr` skill to automatically post PR URLs to Slack channels when configured via MCP tools
+- Changed the `/copy` command to be available to all users
+- Changed background agents to prompt for tool permissions before launching
+- Changed permission rules like `Bash(*)` to be accepted and treated as equivalent to `Bash`
+- Changed config backups to be timestamped and rotated (keeping 5 most recent) to prevent data loss
+
+## 2.1.19
+
+- Added env var `CLAUDE_CODE_ENABLE_TASKS`, set to `false` to keep the old system temporarily
+- Added shorthand `$0`, `$1`, etc. for accessing individual arguments in custom commands
+- Fixed crashes on processors without AVX instruction support
+- Fixed dangling Claude Code processes when terminal is closed by catching EIO errors from `process.exit()` and using SIGKILL as fallback
+- Fixed `/rename` and `/tag` not updating the correct session when resuming from a different directory (e.g., git worktrees)
+- Fixed resuming sessions by custom title not working when run from a different directory
+- Fixed pasted text content being lost when using prompt stash (Ctrl+S) and restore
+- Fixed agent list displaying "Sonnet (default)" instead of "Inherit (default)" for agents without an explicit model setting
+- Fixed backgrounded hook commands not returning early, potentially causing the session to wait on a process that was intentionally backgrounded
+- Fixed file write preview omitting empty lines
+- Changed skills without additional permissions or hooks to be allowed without requiring approval
+- Changed indexed argument syntax from `$ARGUMENTS.0` to `$ARGUMENTS[0]` (bracket syntax)
+- [SDK] Added replay of `queued_command` attachment messages as `SDKUserMessageReplay` events when `replayUserMessages` is enabled
+- [VSCode] Enabled session forking and rewind functionality for all users
+
+## 2.1.18
+
+- Added customizable keyboard shortcuts. Configure keybindings per context, create chord sequences, and personalize your workflow. Run `/keybindings` to get started. Learn more at https://code.claude.com/docs/en/keybindings
+
+## 2.1.17
+
+- Fixed crashes on processors without AVX instruction support
+
+## 2.1.16
+
+- Added new task management system, including new capabilities like dependency tracking
+- [VSCode] Added native plugin management support
+- [VSCode] Added ability for OAuth users to browse and resume remote Claude sessions from the Sessions dialog
+- Fixed out-of-memory crashes when resuming sessions with heavy subagent usage
+- Fixed an issue where the "context remaining" warning was not hidden after running `/compact`
+- Fixed session titles on the resume screen not respecting the user's language setting
+- [IDE] Fixed a race condition on Windows where the Claude Code sidebar view container would not appear on start
+
+## 2.1.15
+
+- Added deprecation notification for npm installations - run `claude install` or see https://docs.anthropic.com/en/docs/claude-code/getting-started for more options
+- Improved UI rendering performance with React Compiler
+- Fixed the "Context left until auto-compact" warning not disappearing after running `/compact`
+- Fixed MCP stdio server timeout not killing child process, which could cause UI freezes
+
+## 2.1.14
+
+- Added history-based autocomplete in bash mode (`!`) - type a partial command and press Tab to complete from your bash command history
+- Added search to installed plugins list - type to filter by name or description
+- Added support for pinning plugins to specific git commit SHAs, allowing marketplace entries to install exact versions
+- Fixed a regression where the context window blocking limit was calculated too aggressively, blocking users at ~65% context usage instead of the intended ~98%
+- Fixed memory issues that could cause crashes when running parallel subagents
+- Fixed memory leak in long-running sessions where stream resources were not cleaned up after shell commands completed
+- Fixed `@` symbol incorrectly triggering file autocomplete suggestions in bash mode
+- Fixed `@`-mention menu folder click behavior to navigate into directories instead of selecting them
+- Fixed `/feedback` command generating invalid GitHub issue URLs when description is very long
+- Fixed `/context` command to show the same token count and percentage as the status line in verbose mode
+- Fixed an issue where `/config`, `/context`, `/model`, and `/todos` command overlays could close unexpectedly
+- Fixed slash command autocomplete selecting wrong command when typing similar commands (e.g., `/context` vs `/compact`)
+- Fixed inconsistent back navigation in plugin marketplace when only one marketplace is configured
+- Fixed iTerm2 progress bar not clearing properly on exit, preventing lingering indicators and bell sounds
+- Improved backspace to delete pasted text as a single token instead of one character at a time
+- [VSCode] Added `/usage` command to display current plan usage
+
+## 2.1.12
+
+- Fixed message rendering bug
+
+## 2.1.11
+
+- Fixed excessive MCP connection requests for HTTP/SSE transports
+
+## 2.1.10
+
+- Added new `Setup` hook event that can be triggered via `--init`, `--init-only`, or `--maintenance` CLI flags for repository setup and maintenance operations
+- Added keyboard shortcut 'c' to copy OAuth URL when browser doesn't open automatically during login
+- Fixed a crash when running bash commands containing heredocs with JavaScript template literals like `${index + 1}`
+- Improved startup to capture keystrokes typed before the REPL is fully ready
+- Improved file suggestions to show as removable attachments instead of inserting text when accepted
+- [VSCode] Added install count display to plugin listings
+- [VSCode] Added trust warning when installing plugins
+
+## 2.1.9
+
+- Added `auto:N` syntax for configuring the MCP tool search auto-enable threshold, where N is the context window percentage (0-100)
+- Added `plansDirectory` setting to customize where plan files are stored
+- Added external editor support (Ctrl+G) in AskUserQuestion "Other" input field
+- Added session URL attribution to commits and PRs created from web sessions
+- Added support for `PreToolUse` hooks to return `additionalContext` to the model
+- Added `${CLAUDE_SESSION_ID}` string substitution for skills to access the current session ID
+- Fixed long sessions with parallel tool calls failing with an API error about orphan tool_result blocks
+- Fixed MCP server reconnection hanging when cached connection promise never resolves
+- Fixed Ctrl+Z suspend not working in terminals using Kitty keyboard protocol (Ghostty, iTerm2, kitty, WezTerm)
+
 ## 2.1.7
 
 - Added `showTurnDuration` setting to hide turn duration messages (e.g., "Cooked for 1m 6s")
@@ -252,7 +557,7 @@
 - Added loading indicator when resuming conversations for better feedback
 - Fixed `/context` command not respecting custom system prompts in non-interactive mode
 - Fixed order of consecutive Ctrl+K lines when pasting with Ctrl+Y
-- Improved @ mention file suggestion speed (~3x faster in git repositories)
+- Improved @ mention file suggestion speed (~3× faster in git repositories)
 - Improved file suggestion performance in repos with `.ignore` or `.rgignore` files
 - Improved settings validation errors to be more prominent
 - Changed thinking toggle from Tab to Alt+T to avoid accidental triggers

examples/settings/README.md

@@ -0,0 +1,31 @@
+# Settings Examples
+
+Example Claude Code settings files, primarily intended for organization-wide deployments. Use these are starting points — adjust them to fit your needs.
+
+These may be applied at any level of the [settings hierarchy](https://code.claude.com/docs/en/settings#settings-files), though certain properties only take effect if specified in enterprise settings (e.g. `strictKnownMarketplaces`, `allowManagedHooksOnly`, `allowManagedPermissionRulesOnly`).
+
+
+## Configuration Examples
+
+> [!WARNING]
+> These examples are community-maintained snippets which may be unsupported or incorrect. You are responsible for the correctness of your own settings configuration.
+
+| Setting | [`settings-lax.json`](./settings-lax.json) | [`settings-strict.json`](./settings-strict.json) | [`settings-bash-sandbox.json`](./settings-bash-sandbox.json) |
+|---------|:---:|:---:|:---:|
+| Disable `--dangerously-skip-permissions` | ✅ | ✅ | |
+| Block plugin marketplaces | ✅ | ✅ | |
+| Block user and project-defined permission `allow` / `ask` / `deny` | | ✅ | ✅ |
+| Block user and project-defined hooks | | ✅ | |
+| Deny web fetch and search tools | | ✅ | |
+| Bash tool requires approval | | ✅ | |
+| Bash tool must run inside of sandbox | | | ✅ |
+
+## Tips
+- Consider merging snippets of the above examples to reach your desired configuration
+- Settings files must be valid JSON
+- Before deploying configuration files to your organization, test them locally by applying to `managed-settings.json`, `settings.json` or `settings.local.json`
+- The `sandbox` property only applies to the `Bash` tool; it does not apply to other tools (like Read, Write, WebSearch, WebFetch, MCPs), hooks, or internal commands
+
+## Full Documentation
+
+See https://code.claude.com/docs/en/settings for complete documentation on all available managed settings.

examples/settings/settings-bash-sandbox.json

@@ -0,0 +1,18 @@
+{
+  "allowManagedPermissionRulesOnly": true,
+  "sandbox": {
+    "enabled": true,
+    "autoAllowBashIfSandboxed": false,
+    "allowUnsandboxedCommands": false,
+    "excludedCommands": [],
+    "network": {
+      "allowUnixSockets": [],
+      "allowAllUnixSockets": false,
+      "allowLocalBinding": false,
+      "allowedDomains": [],
+      "httpProxyPort": null,
+      "socksProxyPort": null
+    },
+    "enableWeakerNestedSandbox": false
+  }
+}

examples/settings/settings-lax.json

@@ -0,0 +1,6 @@
+{
+  "permissions": {
+    "disableBypassPermissionsMode": "disable"
+  },
+  "strictKnownMarketplaces": []
+}

examples/settings/settings-strict.json

@@ -0,0 +1,28 @@
+{
+  "permissions": {
+    "disableBypassPermissionsMode": "disable",
+    "ask": [
+      "Bash"
+    ],
+    "deny": [
+      "WebSearch",
+      "WebFetch"
+    ]
+  },
+  "allowManagedPermissionRulesOnly": true,
+  "allowManagedHooksOnly": true,
+  "strictKnownMarketplaces": [],
+  "sandbox": {
+    "autoAllowBashIfSandboxed": false,
+    "excludedCommands": [],
+    "network": {
+      "allowUnixSockets": [],
+      "allowAllUnixSockets": false,
+      "allowLocalBinding": false,
+      "allowedDomains": [],
+      "httpProxyPort": null,
+      "socksProxyPort": null
+    },
+    "enableWeakerNestedSandbox": false
+  }
+}

plugins/code-review/commands/code-review.md

@@ -6,7 +6,7 @@ description: Code review a pull request
 Provide a code review for the given pull request.
 
 **Agent assumptions (applies to all agents and subagents):**
-- All tools are functional and will work without error. Do not test tools or make exploratory calls.
+- All tools are functional and will work without error. Do not test tools or make exploratory calls. Make sure this is clear to every subagent that is launched.
 - Only call a tool if it is required to complete the task. Every tool call should have a clear purpose.
 
 To do this, follow these steps precisely:
@@ -56,15 +56,23 @@ Note: Still review Claude generated PR's.
 
 6. Filter out any issues that were not validated in step 5. This step will give us our list of high signal issues for our review.
 
-7. If issues were found, skip to step 8 to post inline comments directly.
+7. Output a summary of the review findings to the terminal:
+   - If issues were found, list each issue with a brief description.
+   - If no issues were found, state: "No issues found. Checked for bugs and CLAUDE.md compliance."
 
-   If NO issues were found, post a summary comment using `gh pr comment` (if `--comment` argument is provided):
-   "No issues found. Checked for bugs and CLAUDE.md compliance."
+   If `--comment` argument was NOT provided, stop here. Do not post any GitHub comments.
 
-8. Post inline comments for each issue using `mcp__github_inline_comment__create_inline_comment`. For each comment:
+   If `--comment` argument IS provided and NO issues were found, post a summary comment using `gh pr comment` and stop.
+
+   If `--comment` argument IS provided and issues were found, continue to step 8.
+
+8. Create a list of all comments that you plan on leaving. This is only for you to make sure you are comfortable with the comments. Do not post this list anywhere.
+
+9. Post inline comments for each issue using `mcp__github_inline_comment__create_inline_comment`. For each comment:
    - Provide a brief description of the issue
    - For small, self-contained fixes, include a committable suggestion block
    - For larger fixes (6+ lines, structural changes, or changes spanning multiple locations), describe the issue and suggested fix without a suggestion block
+   - Never post a committable suggestion UNLESS committing the suggestion fixes the issue entirely. If follow up steps are required, do not leave a committable suggestion.
 
    **IMPORTANT: Only post ONE comment per unique issue. Do not post duplicate comments.**
 
@@ -82,7 +90,7 @@ Notes:
 - Use gh CLI to interact with GitHub (e.g., fetch pull requests, create comments). Do not use web fetch.
 - Create a todo list before starting.
 - You must cite and link each issue in inline comments (e.g., if referring to a CLAUDE.md, include a link to it).
-- If no issues are found, post a comment with the following format:
+- If no issues are found and `--comment` argument is provided, post a comment with the following format:
 
 ---
 

scripts/sweep.ts

@@ -0,0 +1,163 @@
+#!/usr/bin/env bun
+
+// --
+
+const NEW_ISSUE = "https://github.com/anthropics/claude-code/issues/new/choose";
+const DRY_RUN = process.argv.includes("--dry-run");
+const STALE_DAYS = 14;
+const STALE_UPVOTE_THRESHOLD = 10;
+
+const CLOSE_MESSAGE = (reason: string) =>
+  `Closing for now — ${reason}. Please [open a new issue](${NEW_ISSUE}) if this is still relevant.`;
+
+const lifecycle = [
+  { label: "invalid",     days: 3,  reason: "this doesn't appear to be about Claude Code" },
+  { label: "needs-repro", days: 7,  reason: "we still need reproduction steps to investigate" },
+  { label: "needs-info",  days: 7,  reason: "we still need a bit more information to move forward" },
+  { label: "stale",       days: 14, reason: "inactive for too long" },
+  { label: "autoclose",   days: 14, reason: "inactive for too long" },
+];
+
+// --
+
+async function githubRequest<T>(
+  endpoint: string,
+  method = "GET",
+  body?: unknown
+): Promise<T> {
+  const token = process.env.GITHUB_TOKEN;
+  if (!token) throw new Error("GITHUB_TOKEN required");
+
+  const response = await fetch(`https://api.github.com${endpoint}`, {
+    method,
+    headers: {
+      Authorization: `Bearer ${token}`,
+      Accept: "application/vnd.github.v3+json",
+      "User-Agent": "sweep",
+      ...(body && { "Content-Type": "application/json" }),
+    },
+    ...(body && { body: JSON.stringify(body) }),
+  });
+
+  if (!response.ok) {
+    if (response.status === 404) return {} as T;
+    const text = await response.text();
+    throw new Error(`GitHub API ${response.status}: ${text}`);
+  }
+
+  return response.json();
+}
+
+// --
+
+async function markStale(owner: string, repo: string) {
+  const cutoff = new Date();
+  cutoff.setDate(cutoff.getDate() - STALE_DAYS);
+
+  let labeled = 0;
+
+  console.log(`\n=== marking stale (${STALE_DAYS}d inactive) ===`);
+
+  for (let page = 1; page <= 10; page++) {
+    const issues = await githubRequest<any[]>(
+      `/repos/${owner}/${repo}/issues?state=open&sort=updated&direction=asc&per_page=100&page=${page}`
+    );
+    if (issues.length === 0) break;
+
+    for (const issue of issues) {
+      if (issue.pull_request) continue;
+      if (issue.locked) continue;
+      if (issue.assignees?.length > 0) continue;
+
+      const updatedAt = new Date(issue.updated_at);
+      if (updatedAt > cutoff) return labeled;
+
+      const alreadyStale = issue.labels?.some(
+        (l: any) => l.name === "stale" || l.name === "autoclose"
+      );
+      if (alreadyStale) continue;
+
+      const isEnhancement = issue.labels?.some(
+        (l: any) => l.name === "enhancement"
+      );
+      const thumbsUp = issue.reactions?.["+1"] ?? 0;
+      if (isEnhancement && thumbsUp >= STALE_UPVOTE_THRESHOLD) continue;
+
+      const base = `/repos/${owner}/${repo}/issues/${issue.number}`;
+
+      if (DRY_RUN) {
+        const age = Math.floor((Date.now() - updatedAt.getTime()) / 86400000);
+        console.log(`#${issue.number}: would label stale (${age}d inactive) — ${issue.title}`);
+      } else {
+        await githubRequest(`${base}/labels`, "POST", { labels: ["stale"] });
+        console.log(`#${issue.number}: labeled stale — ${issue.title}`);
+      }
+      labeled++;
+    }
+  }
+
+  return labeled;
+}
+
+async function closeExpired(owner: string, repo: string) {
+  let closed = 0;
+
+  for (const { label, days, reason } of lifecycle) {
+    const cutoff = new Date();
+    cutoff.setDate(cutoff.getDate() - days);
+    console.log(`\n=== ${label} (${days}d timeout) ===`);
+
+    for (let page = 1; page <= 10; page++) {
+      const issues = await githubRequest<any[]>(
+        `/repos/${owner}/${repo}/issues?state=open&labels=${label}&sort=updated&direction=asc&per_page=100&page=${page}`
+      );
+      if (issues.length === 0) break;
+
+      for (const issue of issues) {
+        if (issue.pull_request) continue;
+        const base = `/repos/${owner}/${repo}/issues/${issue.number}`;
+
+        const events = await githubRequest<any[]>(`${base}/events?per_page=100`);
+
+        const labeledAt = events
+          .filter((e) => e.event === "labeled" && e.label?.name === label)
+          .map((e) => new Date(e.created_at))
+          .pop();
+
+        if (!labeledAt || labeledAt > cutoff) continue;
+
+        if (DRY_RUN) {
+          const age = Math.floor((Date.now() - labeledAt.getTime()) / 86400000);
+          console.log(`#${issue.number}: would close (${label}, ${age}d old) — ${issue.title}`);
+        } else {
+          await githubRequest(`${base}/comments`, "POST", { body: CLOSE_MESSAGE(reason) });
+          await githubRequest(base, "PATCH", { state: "closed", state_reason: "not_planned" });
+          console.log(`#${issue.number}: closed (${label})`);
+        }
+        closed++;
+      }
+    }
+  }
+
+  return closed;
+}
+
+// --
+
+async function main() {
+  const owner = process.env.GITHUB_REPOSITORY_OWNER;
+  const repo = process.env.GITHUB_REPOSITORY_NAME;
+  if (!owner || !repo)
+    throw new Error("GITHUB_REPOSITORY_OWNER and GITHUB_REPOSITORY_NAME required");
+
+  if (DRY_RUN) console.log("DRY RUN — no changes will be made\n");
+
+  const labeled = await markStale(owner, repo);
+  const closed = await closeExpired(owner, repo);
+
+  console.log(`\nDone: ${labeled} ${DRY_RUN ? "would be labeled" : "labeled"} stale, ${closed} ${DRY_RUN ? "would be closed" : "closed"}`);
+}
+
+main().catch(console.error);
+
+export {};