identify_potential_duplicate_issues.yml

  1name: Identify potential duplicates among new bug/crash reports
  2
  3on:
  4  issues:
  5    types: [opened]
  6  workflow_dispatch:
  7    inputs:
  8      issue_number:
  9        description: "Issue number to analyze (for testing)"
 10        required: true
 11        type: number
 12
 13concurrency:
 14  group: potential-duplicate-check-${{ github.event.issue.number || inputs.issue_number }}
 15  # let's not overspend tokens on multiple parallel checks of the same issue
 16  cancel-in-progress: true
 17
 18jobs:
 19  identify-duplicates:
 20    if: github.repository == 'zed-industries/zed'
 21    runs-on: ubuntu-latest
 22    # let's not overspend tokens on checks that went too deep into the rabbit hole
 23    timeout-minutes: 5
 24    permissions:
 25      contents: read
 26      issues: read
 27      id-token: write
 28
 29    steps:
 30      - name: Get github app token
 31        id: get-app-token
 32        uses: actions/create-github-app-token@bef1eaf1c0ac2b148ee2a0a74c65fbe6db0631f1 # v2.1.4
 33        with:
 34          app-id: ${{ secrets.ZED_COMMUNITY_BOT_APP_ID }}
 35          private-key: ${{ secrets.ZED_COMMUNITY_BOT_PRIVATE_KEY }}
 36          owner: zed-industries
 37
 38      - name: Check issue type
 39        id: check-type
 40        uses: actions/github-script@60a0d83039c74a4aee543508d2ffcb1c3799cdea # v7.0.1
 41        with:
 42          github-token: ${{ steps.get-app-token.outputs.token }}
 43          script: |
 44            const issueNumber = context.payload.issue?.number || ${{ inputs.issue_number || 0 }};
 45            if (!issueNumber) {
 46              core.setFailed('No issue number provided');
 47              return;
 48            }
 49
 50            const { data: issue } = await github.rest.issues.get({
 51              owner: context.repo.owner,
 52              repo: context.repo.repo,
 53              issue_number: issueNumber
 54            });
 55
 56            const typeName = issue.type?.name;
 57            const isTargetType = typeName === 'Bug' || typeName === 'Crash';
 58
 59            console.log(`Issue #${issueNumber}: "${issue.title}"`);
 60            console.log(`Issue type: ${typeName || '(none)'}`);
 61            console.log(`Is target type (Bug/Crash): ${isTargetType}`);
 62
 63            core.setOutput('issue_number', issueNumber);
 64            core.setOutput('issue_author', issue.user?.login || '');
 65            core.setOutput('is_target_type', isTargetType);
 66
 67            if (!isTargetType) {
 68              console.log('::notice::Skipping - issue type is not Bug or Crash');
 69            }
 70
 71      - name: Check if author is staff
 72        if: steps.check-type.outputs.is_target_type == 'true'
 73        id: check-staff
 74        uses: actions/github-script@60a0d83039c74a4aee543508d2ffcb1c3799cdea # v7.0.1
 75        with:
 76          github-token: ${{ steps.get-app-token.outputs.token }}
 77          script: |
 78            const author = process.env.ISSUE_AUTHOR || '';
 79            if (!author) {
 80              console.log('Could not determine issue author, proceeding with check');
 81              core.setOutput('is_staff', 'false');
 82              return;
 83            }
 84
 85            try {
 86              const response = await github.rest.teams.getMembershipForUserInOrg({
 87                org: 'zed-industries',
 88                team_slug: 'staff',
 89                username: author
 90              });
 91              const isStaff = response.data.state === 'active';
 92              core.setOutput('is_staff', String(isStaff));
 93              if (isStaff) {
 94                console.log(`::notice::Skipping - author @${author} is a staff member`);
 95              }
 96            } catch (error) {
 97              if (error.status === 404) {
 98                core.setOutput('is_staff', 'false');
 99              } else {
100                throw error;
101              }
102            }
103        env:
104          ISSUE_AUTHOR: ${{ steps.check-type.outputs.issue_author }}
105
106      - name: Checkout repository
107        if: |
108          steps.check-type.outputs.is_target_type == 'true' &&
109          steps.check-staff.outputs.is_staff == 'false'
110        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
111        with:
112          fetch-depth: 1
113
114      - name: Analyze for potential duplicates (DRY RUN)
115        if: |
116          steps.check-type.outputs.is_target_type == 'true' &&
117          steps.check-staff.outputs.is_staff == 'false'
118        id: analyze
119        uses: anthropics/claude-code-action@231bd75b7196d48291c1498f1c6d277c2810d9a3 # v1
120        with:
121          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY_ISSUE_DEDUP }}
122          github_token: ${{ steps.get-app-token.outputs.token }}
123
124          prompt: |
125            You are analyzing issue #${{ steps.check-type.outputs.issue_number }} in the zed-industries/zed repository to determine if it might be a duplicate of an existing issue.
126
127            THIS IS A DRY RUN - do not post any comments or modify anything. Only analyze and return your findings.
128
129            ## Instructions
130
131            1. Use mcp__github__get_issue to fetch the full details of issue #${{ steps.check-type.outputs.issue_number }}
132
133            2. Extract key identifying information:
134               - Error messages (exact text)
135               - Stack traces or panic messages
136               - Affected features/components
137               - Steps to reproduce
138               - Platform/OS information
139
140            3. Search for potential duplicates using mcp__github__search_issues with:
141               - Key error messages or panic text (most reliable signal)
142               - Specific feature names or components mentioned
143               - Limit search to repo:zed-industries/zed and recent issues (last 90 days)
144               - Search both open AND closed issues (duplicates may have been closed)
145
146            4. For each potential match, evaluate similarity:
147               - SAME error message or stack trace = high confidence
148               - SAME steps to reproduce with same outcome = high confidence
149               - Similar description but different error/context = low confidence
150               - Vaguely related topic = NOT a duplicate
151
152            ## Critical Guidelines
153
154            - Be VERY conservative. When in doubt, conclude it is NOT a duplicate.
155            - Only flag as potential duplicate if you have HIGH confidence (same error, same repro steps, same root cause).
156            - "Similar topic" or "related feature" is NOT sufficient - the issues must describe the SAME bug.
157            - False positives are worse than false negatives. Users finding their legitimate issue incorrectly flagged as duplicate is a poor experience.
158
159            ## Output
160
161            Return your analysis as JSON with this exact structure. Do not include any other text outside the JSON.
162
163          claude_args: |
164            --max-turns 3
165            --allowedTools mcp__github__get_issue,mcp__github__search_issues,mcp__github__list_issues
166            --json-schema {"type":"object","properties":{"issue_number":{"type":"integer"},"issue_title":{"type":"string"},"is_potential_duplicate":{"type":"boolean"},"confidence":{"type":"string","enum":["high","medium","low","none"]},"potential_duplicates":{"type":"array","items":{"type":"object","properties":{"number":{"type":"integer"},"title":{"type":"string"},"similarity_reason":{"type":"string"}},"required":["number","title","similarity_reason"]}},"analysis_summary":{"type":"string"},"recommendation":{"type":"string","enum":["flag_as_duplicate","needs_human_review","not_a_duplicate"]}},"required":["issue_number","is_potential_duplicate","confidence","potential_duplicates","analysis_summary","recommendation"]}
167
168      - name: Log analysis results
169        if: |
170          steps.check-type.outputs.is_target_type == 'true' &&
171          steps.check-staff.outputs.is_staff == 'false' &&
172          !cancelled()
173        uses: actions/github-script@60a0d83039c74a4aee543508d2ffcb1c3799cdea # v7.0.1
174        with:
175          script: |
176            const output = process.env.ANALYSIS_OUTPUT || '';
177
178            console.log('='.repeat(60));
179            console.log('DRY RUN ANALYSIS RESULTS');
180            console.log('='.repeat(60));
181
182            if (!output || output === '') {
183              console.log('No structured output received from analysis');
184              core.summary.addHeading('⚠️ Analysis did not produce output', 2);
185              core.summary.addRaw('The duplicate detection analysis did not return structured output. Check the workflow logs for details.');
186              await core.summary.write();
187              return;
188            }
189
190            try {
191              const analysis = JSON.parse(output);
192
193              console.log(`\nIssue: #${analysis.issue_number} - ${analysis.issue_title || 'N/A'}`);
194              console.log(`Is Potential Duplicate: ${analysis.is_potential_duplicate}`);
195              console.log(`Confidence: ${analysis.confidence}`);
196              console.log(`Recommendation: ${analysis.recommendation}`);
197              console.log(`\nAnalysis Summary:\n${analysis.analysis_summary}`);
198
199              if (analysis.potential_duplicates.length > 0) {
200                console.log(`\nPotential Duplicates Found: ${analysis.potential_duplicates.length}`);
201                for (const dup of analysis.potential_duplicates) {
202                  console.log(`  - #${dup.number}: ${dup.title}`);
203                  console.log(`    Reason: ${dup.similarity_reason}`);
204                }
205              } else {
206                console.log('\nNo potential duplicates identified.');
207              }
208
209              console.log('\n' + '='.repeat(60));
210
211              // set summary for workflow run
212              const summaryIcon = analysis.is_potential_duplicate ? '⚠️' : '✅';
213              const summaryText = analysis.is_potential_duplicate
214                ? `Potential duplicate detected (${analysis.confidence} confidence)`
215                : 'No duplicate detected';
216
217              core.summary.addHeading(`${summaryIcon} Issue #${analysis.issue_number}: ${summaryText}`, 2);
218              core.summary.addRaw(`\n**Recommendation:** ${analysis.recommendation}\n\n`);
219              core.summary.addRaw(`**Summary:** ${analysis.analysis_summary}\n\n`);
220
221              if (analysis.potential_duplicates.length > 0) {
222                core.summary.addHeading('Potential Duplicates', 3);
223                const rows = analysis.potential_duplicates.map(d => [
224                  `#${d.number}`,
225                  d.title,
226                  d.similarity_reason
227                ]);
228                core.summary.addTable([
229                  [{data: 'Issue', header: true}, {data: 'Title', header: true}, {data: 'Similarity Reason', header: true}],
230                  ...rows
231                ]);
232              }
233
234              await core.summary.write();
235
236            } catch (e) {
237              console.log('Failed to parse analysis output:', e.message);
238              console.log('Raw output:', output);
239              core.summary.addHeading('⚠️ Failed to parse analysis output', 2);
240              core.summary.addRaw(`Error: ${e.message}\n\nRaw output:\n\`\`\`\n${output}\n\`\`\``);
241              await core.summary.write();
242            }
243        env:
244          ANALYSIS_OUTPUT: ${{ steps.analyze.outputs.structured_output }}