diff --git a/.claude/skills/e2e/implement.md b/.claude/skills/e2e/implement.md
index 342c87213..db9ba8e93 100644
--- a/.claude/skills/e2e/implement.md
+++ b/.claude/skills/e2e/implement.md
@@ -2,6 +2,10 @@
 
 Apply fixes for E2E test failures, verify with scoped E2E tests.
 
+> **Before implementing any fixes, enter plan mode by invoking /plan.**
+> Analyze the findings (Steps 1-2 below), produce a complete fix plan with
+> specific file paths and code changes, and get user approval before executing.
+
 > **IMPORTANT: Running real E2E tests is a HARD REQUIREMENT of this procedure.**
 > Every fix MUST be verified with real E2E tests before the summary step.
 > Canary tests use the Vogon fake agent and cannot catch agent-specific issues.
diff --git a/.github/workflows/e2e-fix.yml b/.github/workflows/e2e-fix.yml
new file mode 100644
index 000000000..c110c948b
--- /dev/null
+++ b/.github/workflows/e2e-fix.yml
@@ -0,0 +1,140 @@
+name: E2E Fix
+
+on:
+  workflow_dispatch:
+    inputs:
+      triage_run_id:
+        description: Run ID of the triage workflow (for downloading plan artifacts)
+        required: true
+        type: string
+      run_url:
+        description: Original failed E2E run URL
+        required: true
+        type: string
+      failed_agents:
+        description: Comma-separated list of agents to fix
+        required: true
+        type: string
+      slack_channel:
+        description: Slack channel ID for thread replies
+        required: false
+        type: string
+      slack_thread_ts:
+        description: Slack thread timestamp for replies
+        required: false
+        type: string
+
+permissions:
+  actions: read
+  contents: write
+  pull-requests: write
+  id-token: write
+
+concurrency:
+  group: e2e-fix-${{ inputs.run_url || github.run_id }}
+  cancel-in-progress: true
+
+jobs:
+  fix:
+    runs-on: ubuntu-latest
+    timeout-minutes: 30
+    env:
+      SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}
+      SLACK_CHANNEL: ${{ inputs.slack_channel }}
+      SLACK_THREAD_TS: ${{ inputs.slack_thread_ts }}
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+        with:
+          fetch-depth: 0
+
+      - name: Post fix started
+        if: ${{ env.SLACK_BOT_TOKEN != '' && env.SLACK_CHANNEL != '' && env.SLACK_THREAD_TS != '' }}
+        shell: bash
+        env:
+          FAILED_AGENTS: ${{ inputs.failed_agents }}
+        run: |
+          set -euo pipefail
+
+          scripts/post-slack-message.sh "Starting E2E fix for \`${FAILED_AGENTS}\`."
+
+      - name: Setup mise
+        uses: jdx/mise-action@v4
+
+      - name: Download plan artifacts
+        env:
+          GH_TOKEN: ${{ github.token }}
+          TRIAGE_RUN_ID: ${{ inputs.triage_run_id }}
+          FAILED_AGENTS: ${{ inputs.failed_agents }}
+        shell: bash
+        run: |
+          set -euo pipefail
+
+          mkdir -p triage-plans
+
+          IFS=',' read -ra agents <<< "$FAILED_AGENTS"
+          for agent in "${agents[@]}"; do
+            agent="$(echo "$agent" | xargs)"  # trim whitespace
+            echo "Downloading plan for $agent..."
+            gh run download "$TRIAGE_RUN_ID" \
+              --name "e2e-plan-${agent}" \
+              --dir "triage-plans/${agent}" || {
+              echo "warning: no plan artifact found for $agent" >&2
+              continue
+            }
+          done
+
+          echo "Downloaded plans:"
+          find triage-plans -name '*.md' -type f
+
+      - name: Apply fixes
+        id: fix
+        uses: anthropics/claude-code-action@v1
+        with:
+          prompt: |
+            Read the fix plans in the triage-plans/ directory. Each subdirectory contains a plan.md for one agent.
+
+            Execute all fixes exactly as specified in the plans. After applying fixes, run:
+            1. mise run fmt
+            2. mise run lint
+            3. mise run test:e2e:canary
+
+            If verification passes, create a git branch fix/e2e-${{ github.run_id }}, commit all changes,
+            push, and create a draft PR with a summary of what was fixed.
+
+            If verification fails, fix the issues and retry. Do not give up without attempting to fix lint/format errors.
+          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
+          claude_args: "--allowedTools 'Edit,Write,Read,Glob,Grep,Bash(git:*),Bash(mise:*),Bash(gh:*)'"
+
+      - name: Post success to Slack
+        if: success() && env.SLACK_BOT_TOKEN != '' && env.SLACK_CHANNEL != '' && env.SLACK_THREAD_TS != ''
+        shell: bash
+        env:
+          GH_TOKEN: ${{ github.token }}
+          FIX_BRANCH: fix/e2e-${{ github.run_id }}
+          RUN_URL: https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }}
+        run: |
+          set -euo pipefail
+
+          # Find the draft PR URL from the fix step output
+          pr_url="$(gh pr list --head "$FIX_BRANCH" --json url -q '.[0].url' 2>/dev/null || true)"
+
+          if [ -n "$pr_url" ]; then
+            message="E2E fix complete — draft PR ready: <${pr_url}|Review PR>"
+          else
+            message="E2E fix complete — changes applied but no PR was created. Check the <${RUN_URL}|workflow run> for details."
+          fi
+
+          scripts/post-slack-message.sh "$message"
+
+      - name: Post failure to Slack
+        if: failure() && env.SLACK_BOT_TOKEN != '' && env.SLACK_CHANNEL != '' && env.SLACK_THREAD_TS != ''
+        shell: bash
+        env:
+          RUN_URL: https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }}
+        run: |
+          set -euo pipefail
+
+          message="E2E fix failed. Check the <${RUN_URL}|workflow run> for details."
+
+          scripts/post-slack-message.sh "$message"
diff --git a/.github/workflows/e2e-triage.yml b/.github/workflows/e2e-triage.yml
new file mode 100644
index 000000000..e26a2bb23
--- /dev/null
+++ b/.github/workflows/e2e-triage.yml
@@ -0,0 +1,327 @@
+name: E2E Triage
+
+on:
+  workflow_dispatch:
+    inputs:
+      run_url:
+        description: GitHub Actions run URL to triage
+        required: true
+        type: string
+      sha:
+        description: Commit SHA (auto-detected from run if omitted)
+        required: false
+        type: string
+      failed_agents:
+        description: Comma-separated list of failed agents (auto-detected from run if omitted)
+        required: false
+        type: string
+      slack_channel:
+        description: Slack channel ID for the originating thread
+        required: false
+        type: string
+      slack_thread_ts:
+        description: Slack thread timestamp for replies
+        required: false
+        type: string
+      rerun:
+        description: 'Re-run failing tests locally to detect flaky (costs API tokens)'
+        required: false
+        type: boolean
+        default: false
+
+permissions:
+  actions: read
+  contents: read
+  id-token: write
+
+concurrency:
+  group: e2e-triage-${{ inputs.run_url || github.run_id }}
+  cancel-in-progress: true
+
+jobs:
+  matrix-setup:
+    runs-on: ubuntu-latest
+    outputs:
+      agents: ${{ steps.set.outputs.agents }}
+      run_url: ${{ steps.set.outputs.run_url }}
+      sha: ${{ steps.set.outputs.sha }}
+      slack_channel: ${{ steps.set.outputs.slack_channel }}
+      slack_thread_ts: ${{ steps.set.outputs.slack_thread_ts }}
+    steps:
+      - name: Validate payload and build matrix
+        id: set
+        shell: bash
+        env:
+          GH_TOKEN: ${{ github.token }}
+          REPO_NAME: ${{ github.repository }}
+          RUN_URL_INPUT: ${{ inputs.run_url }}
+          SHA_INPUT: ${{ inputs.sha }}
+          FAILED_AGENTS_INPUT: ${{ inputs.failed_agents }}
+          SLACK_CHANNEL_INPUT: ${{ inputs.slack_channel }}
+          SLACK_THREAD_TS_INPUT: ${{ inputs.slack_thread_ts }}
+        run: |
+          set -euo pipefail
+
+          csv_to_json() {
+            printf '%s' "$1" | jq -R -s -c 'split(",") | map(gsub("^\\s+|\\s+$"; "")) | map(select(length > 0))'
+          }
+
+          run_url="$RUN_URL_INPUT"
+          sha="${SHA_INPUT:-}"
+          slack_channel="${SLACK_CHANNEL_INPUT:-}"
+          slack_thread_ts="${SLACK_THREAD_TS_INPUT:-}"
+
+          # Derive missing values from run URL via GitHub API
+          if [ -z "$sha" ] || [ -z "$FAILED_AGENTS_INPUT" ]; then
+            run_id=$(echo "$run_url" | grep -oE '/runs/[0-9]+' | grep -oE '[0-9]+')
+            if [ -z "$run_id" ]; then
+              echo "Could not extract run ID from run_url: $run_url" >&2
+              exit 1
+            fi
+            run_data=$(gh run view "$run_id" --repo "$REPO_NAME" --json headSha,jobs)
+            if [ -z "$sha" ]; then
+              sha=$(echo "$run_data" | jq -r '.headSha')
+            fi
+            if [ -z "$FAILED_AGENTS_INPUT" ]; then
+              agents_json=$(echo "$run_data" | jq -c '[.jobs[]
+                | select(.conclusion == "failure")
+                | (.name | (try capture("\\((?<agent>[^)]+)\\)").agent catch null))
+                | select(. != null)
+              ]')
+            fi
+          fi
+
+          if [ -n "$FAILED_AGENTS_INPUT" ]; then
+            agents_json="$(csv_to_json "$FAILED_AGENTS_INPUT")"
+          fi
+
+          if [ -z "$run_url" ]; then
+            echo "run_url is required" >&2
+            exit 1
+          fi
+          if [ -z "$sha" ]; then
+            echo "sha is required" >&2
+            exit 1
+          fi
+          if [ -z "$agents_json" ] || [ "$agents_json" = "[]" ] || [ "$agents_json" = "null" ]; then
+            echo "agents is required (provide failed_agents input or ensure failed job names contain '(agent-name)')" >&2
+            exit 1
+          fi
+
+          echo "run_url=$run_url" >> "$GITHUB_OUTPUT"
+          echo "sha=$sha" >> "$GITHUB_OUTPUT"
+          echo "slack_channel=$slack_channel" >> "$GITHUB_OUTPUT"
+          echo "slack_thread_ts=$slack_thread_ts" >> "$GITHUB_OUTPUT"
+          echo "agents=$agents_json" >> "$GITHUB_OUTPUT"
+
+  triage:
+    needs: [matrix-setup]
+    runs-on: ubuntu-latest
+    timeout-minutes: ${{ inputs.rerun == true && 90 || 45 }}
+    env:
+      RUN_URL: ${{ needs.matrix-setup.outputs.run_url }}
+      E2E_AGENT: ${{ matrix.agent }}
+      SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}
+      SLACK_CHANNEL: ${{ needs.matrix-setup.outputs.slack_channel }}
+      SLACK_THREAD_TS: ${{ needs.matrix-setup.outputs.slack_thread_ts }}
+    strategy:
+      fail-fast: false
+      matrix:
+        agent: ${{ fromJson(needs.matrix-setup.outputs.agents) }}
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+        with:
+          fetch-depth: 1
+
+      - name: Post triage started
+        if: ${{ env.SLACK_BOT_TOKEN != '' && env.SLACK_CHANNEL != '' && env.SLACK_THREAD_TS != '' }}
+        shell: bash
+        run: |
+          set -euo pipefail
+
+          scripts/post-slack-message.sh "Starting E2E triage for \`$E2E_AGENT\` on <$RUN_URL|this run>."
+
+      - name: Setup mise
+        uses: jdx/mise-action@v4
+
+      - name: Download E2E artifacts
+        id: artifacts
+        if: inputs.rerun != true
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          artifact_path="$(scripts/download-e2e-artifacts.sh "$RUN_URL")"
+          echo "path=$artifact_path" >> "$GITHUB_OUTPUT"
+
+      - name: Install system dependencies
+        if: inputs.rerun == true
+        run: sudo apt-get update && sudo apt-get install -y tmux
+
+      - name: Install agent CLI
+        if: inputs.rerun == true
+        run: |
+          case "${{ matrix.agent }}" in
+            claude-code) curl -fsSL https://claude.ai/install.sh | bash ;;
+            opencode)    curl -fsSL https://opencode.ai/install | bash ;;
+            gemini-cli)  npm install -g @google/gemini-cli ;;
+            cursor-cli)  curl https://cursor.com/install -fsS | bash ;;
+            factoryai-droid) curl -fsSL https://app.factory.ai/cli | sh ;;
+            copilot-cli)  npm install -g @github/copilot  ;;
+            roger-roger) ;; # installed by mise (see mise.toml)
+          esac
+          echo "$HOME/.local/bin" >> $GITHUB_PATH
+
+      - name: Bootstrap agent
+        if: inputs.rerun == true && matrix.agent != 'roger-roger'
+        env:
+          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
+          GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }}
+          CURSOR_API_KEY: ${{ secrets.CURSOR_API_KEY }}
+          FACTORY_API_KEY: ${{ secrets.FACTORY_API_KEY }}
+          COPILOT_GITHUB_TOKEN: ${{ secrets.COPILOT_GITHUB_TOKEN }}
+        run: go run ./e2e/bootstrap
+
+      - name: Run triage (analysis only)
+        id: triage_analysis
+        if: inputs.rerun != true
+        uses: anthropics/claude-code-action@v1
+        with:
+          prompt: |
+            /e2e:triage-ci ${{ steps.artifacts.outputs.path }} --agent ${{ matrix.agent }} --sha ${{ needs.matrix-setup.outputs.sha }}
+          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
+          claude_args: "--allowedTools 'Read,Grep,Glob'"
+          display_report: true
+
+      - name: Run triage (with re-runs)
+        id: triage_rerun
+        if: inputs.rerun == true
+        uses: anthropics/claude-code-action@v1
+        env:
+          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
+          GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }}
+          CURSOR_API_KEY: ${{ secrets.CURSOR_API_KEY }}
+          FACTORY_API_KEY: ${{ secrets.FACTORY_API_KEY }}
+          COPILOT_GITHUB_TOKEN: ${{ secrets.COPILOT_GITHUB_TOKEN }}
+        with:
+          prompt: |
+            /e2e:triage-ci ${{ env.RUN_URL }} --agent ${{ matrix.agent }} --sha ${{ needs.matrix-setup.outputs.sha }}
+          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
+          claude_args: "--allowedTools 'Read,Grep,Glob,Bash(mise:*),Bash(scripts:*)'"
+          display_report: true
+
+      - name: Resolve triage outcome
+        id: triage
+        if: always()
+        shell: bash
+        env:
+          ANALYSIS_OUTCOME: ${{ steps.triage_analysis.outcome }}
+          ANALYSIS_EXEC: ${{ steps.triage_analysis.outputs.execution_file }}
+          RERUN_OUTCOME: ${{ steps.triage_rerun.outcome }}
+          RERUN_EXEC: ${{ steps.triage_rerun.outputs.execution_file }}
+        run: |
+          outcome="${ANALYSIS_OUTCOME:-skipped}"
+          exec_file="${ANALYSIS_EXEC}"
+          if [ "$outcome" = "skipped" ]; then
+            outcome="${RERUN_OUTCOME:-skipped}"
+            exec_file="${RERUN_EXEC}"
+          fi
+          echo "outcome=$outcome" >> "$GITHUB_OUTPUT"
+          echo "execution_file=$exec_file" >> "$GITHUB_OUTPUT"
+
+      - name: Extract triage output
+        id: triage_output
+        if: always() && steps.triage.outputs.execution_file != ''
+        shell: bash
+        env:
+          EXECUTION_FILE: ${{ steps.triage.outputs.execution_file }}
+          TRIAGE_OUTPUT_FILE: ${{ github.workspace }}/e2e-triage-artifacts/${{ matrix.agent }}/triage.md
+        run: |
+          set -euo pipefail
+
+          mkdir -p "$(dirname "$TRIAGE_OUTPUT_FILE")"
+          # Extract assistant text content from execution JSON
+          jq -r '[.[] | select(.type == "assistant") | .message.content[]
+            | select(.type == "text") | .text] | join("\n")' \
+            "$EXECUTION_FILE" > "$TRIAGE_OUTPUT_FILE"
+
+      - name: Generate fix plan
+        id: plan
+        if: steps.triage.outputs.outcome == 'success'
+        uses: anthropics/claude-code-action@v1
+        with:
+          prompt: |
+            /e2e:implement Read the triage findings at ${{ github.workspace }}/e2e-triage-artifacts/${{ matrix.agent }}/triage.md for agent ${{ matrix.agent }}.
+          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
+          claude_args: "--allowedTools 'Read,Grep,Glob,Write,EnterPlanMode,ExitPlanMode,Bash(mise:*)'"
+          display_report: true
+
+      - name: Extract plan output
+        id: plan_output
+        if: steps.plan.outcome == 'success' && steps.plan.outputs.execution_file != ''
+        shell: bash
+        env:
+          EXECUTION_FILE: ${{ steps.plan.outputs.execution_file }}
+          PLAN_FILE: ${{ github.workspace }}/e2e-triage-artifacts/${{ matrix.agent }}/plan.md
+        run: |
+          set -euo pipefail
+
+          mkdir -p "$(dirname "$PLAN_FILE")"
+          jq -r '[.[] | select(.type == "assistant") | .message.content[]
+            | select(.type == "text") | .text] | join("\n")' \
+            "$EXECUTION_FILE" > "$PLAN_FILE"
+
+      - name: Post triage completion
+        if: ${{ always() && (steps.triage.outputs.outcome == 'success' || steps.triage.outputs.outcome == 'failure') && env.SLACK_BOT_TOKEN != '' && env.SLACK_CHANNEL != '' && env.SLACK_THREAD_TS != '' }}
+        shell: bash
+        env:
+          TRIAGE_OUTCOME: ${{ steps.triage.outputs.outcome }}
+        run: |
+          set -euo pipefail
+
+          if [ "$TRIAGE_OUTCOME" = "success" ]; then
+            message="E2E triage complete for \`$E2E_AGENT\`."
+          else
+            message="E2E triage failed for \`$E2E_AGENT\`."
+          fi
+
+          scripts/post-slack-message.sh "$message"
+
+      - name: Post fix plan to Slack
+        if: steps.plan_output.outcome == 'success' && env.SLACK_BOT_TOKEN != '' && env.SLACK_CHANNEL != '' && env.SLACK_THREAD_TS != ''
+        shell: bash
+        env:
+          PLAN_FILE: ${{ github.workspace }}/e2e-triage-artifacts/${{ matrix.agent }}/plan.md
+          TRIAGE_RUN_ID: ${{ github.run_id }}
+        run: |
+          set -euo pipefail
+
+          # Extract first few lines as summary
+          summary="$(head -20 "$PLAN_FILE" | sed '/^$/d' | head -5)"
+
+          # Construct Fix It URL
+          encoded_run_url="$(python3 -c "import urllib.parse, os; print(urllib.parse.quote(os.environ['RUN_URL'], safe=''))")"
+          fix_url="https://e2e-triage.entireio.workers.dev/fix?triage_run_id=${TRIAGE_RUN_ID}&run_url=${encoded_run_url}&failed_agents=${E2E_AGENT}&slack_channel=${SLACK_CHANNEL}&slack_thread_ts=${SLACK_THREAD_TS}"
+
+          message="Fix plan ready for \`$E2E_AGENT\`:
+          ${summary}
+
+          <${fix_url}|Fix It> — applies the plan and creates a draft PR"
+
+          scripts/post-slack-message.sh "$message"
+
+      - name: Upload triage output
+        if: always()
+        uses: actions/upload-artifact@v7
+        with:
+          name: e2e-triage-${{ matrix.agent }}
+          path: e2e-triage-artifacts/
+          retention-days: 7
+
+      - name: Upload plan artifact
+        if: always() && steps.plan_output.outcome == 'success'
+        uses: actions/upload-artifact@v7
+        with:
+          name: e2e-plan-${{ matrix.agent }}
+          path: e2e-triage-artifacts/${{ matrix.agent }}/plan.md
+          retention-days: 7
diff --git a/scripts/post-slack-message.sh b/scripts/post-slack-message.sh
new file mode 100755
index 000000000..76546dc0b
--- /dev/null
+++ b/scripts/post-slack-message.sh
@@ -0,0 +1,24 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+# Post a message to a Slack thread using the chat.postMessage API.
+# Requires SLACK_BOT_TOKEN, SLACK_CHANNEL, and SLACK_THREAD_TS env vars.
+
+text="${1:?message is required}"
+payload="$(jq -n \
+  --arg channel "$SLACK_CHANNEL" \
+  --arg thread_ts "$SLACK_THREAD_TS" \
+  --arg text "$text" \
+  '{channel: $channel, thread_ts: $thread_ts, text: $text}')"
+
+if ! response="$(curl -fsS https://slack.com/api/chat.postMessage \
+  -H "Authorization: Bearer ${SLACK_BOT_TOKEN}" \
+  -H 'Content-type: application/json; charset=utf-8' \
+  --data "$payload")"; then
+  echo "warning: slack notification failed" >&2
+  exit 0
+fi
+
+if ! jq -e '.ok == true' >/dev/null <<<"$response"; then
+  echo "warning: slack notification returned non-ok response" >&2
+fi