Aitask Contribution Review

Workflow

Step 0: Resolve Issue Number

If <issue_number> argument is provided: Use it directly and proceed to Step 1.

If no argument is provided: List open contribution issues using the platform-encapsulated script:

./.aitask-scripts/aitask_contribution_review.sh list-issues

Parse the structured output blocks:

• @@@ISSUE:<num>@@@ — Issue separator
• TITLE:<title> — Issue title
• HAS_METADATA:true|false — Whether issue has aitask-contribute-metadata

If output is NO_ISSUES: Inform user "No open contribution issues found." and end the workflow.

Filter to only issues with HAS_METADATA:true (these are actual contribution issues).

If no issues have metadata: Inform user "No contribution issues with metadata found among open issues." and end the workflow.

Present the filtered issues via AskUserQuestion:

• Question: "Which contribution issue should I review?"
• Header: "Issue"
• Options: Each contribution issue (label: #<num> — <short_title>, description: full title)

Use the selected issue number as <issue_number> for Step 1.

IMPORTANT: Do NOT call gh, glab, curl, or any platform-specific CLI commands directly. Always use the aitask_contribution_review.sh script subcommands, which encapsulate platform detection and work on GitHub, GitLab, and Bitbucket.

Step 1: Validate and Fetch Target Issue

Fetch the target issue using the helper script (encapsulates platform detection and API access):

./.aitask-scripts/aitask_contribution_review.sh fetch <issue_number>

Parse the structured output lines:

• ISSUE_JSON:<json> — Full issue JSON including comments
• HAS_METADATA:true|false — Whether the issue has aitask-contribute-metadata
• CONTRIBUTOR:<name> — Contributor username
• EMAIL:<email> — Contributor email
• AREAS:<csv> — Affected code areas
• FILE_PATHS:<csv> — Changed file paths
• FILE_DIRS:<csv> — Changed directories
• CHANGE_TYPE:<type> — Type of change

If HAS_METADATA:false: Inform user "Issue #N is not a contribution issue (no aitask-contribute-metadata found)" and abort.

Display summary:

## Contribution Issue #<N>
- **Title:** <from JSON>
- **Contributor:** <name> (<email>)
- **Areas:** <areas>
- **Change type:** <type>
- **Files:** <file_paths>

Step 1b: Check for Duplicate Import

Check if this issue has already been imported as a task:

./.aitask-scripts/aitask_contribution_review.sh check-imported <issue_number>

Parse the output:

• IMPORTED:<task_file_path> — Issue was already imported
• NOT_IMPORTED — Issue has not been imported yet

If IMPORTED:<path>: Extract the task name from the path (e.g., t387_fix_nodejs_deprecation from aitasks/t387_fix_nodejs_deprecation.md). Use AskUserQuestion:

• Question: "Issue # has already been imported as task (). How to proceed?"
• Header: "Duplicate"
• Options:
  • "Proceed anyway" (description: "Continue with the review despite the existing import")
  • "Abort" (description: "Stop — this issue is already tracked")

If "Abort": Inform user "Issue # is already imported as . No action taken." and end the workflow.

Step 2: Gather Related Issues

Run the find-related subcommand:

./.aitask-scripts/aitask_contribution_review.sh find-related <issue_number>

Parse output lines:

• OVERLAP:<num>:<score> — Issue found via fingerprint overlap (score >= 4)
• LINKED:<num>:<title> — Issue found via #N references in body/comments
• BOTH:<num>:<score>:<title> — Issue found in both sources
• NO_BOT_COMMENT — No overlap analysis bot comment found
• TOTAL_CANDIDATES:<count> — Total unique candidate issues

If NO_BOT_COMMENT appears in output: Use AskUserQuestion:

• Question: "No overlap analysis comment found on issue #. Run a local overlap check?"
• Header: "Overlap"
• Options:
  • "Run local check" (description: "Execute aitask_contribution_check.sh --dry-run to compute overlaps locally")
  • "Skip overlap check" (description: "Proceed with only linked issue references")

If "Run local check":

./.aitask-scripts/aitask_contribution_check.sh <issue_number> --dry-run --silent

Parse the output for <!-- overlap-results top_overlaps: ... --> line. Extract N:S pairs where score >= 4 and add them to the candidate list.

If TOTAL_CANDIDATES:0 (and no local check produced results): Skip to Step 5 with recommendation "single import, no related issues".

Step 3: Fetch Related Issue Details

Collect all candidate issue numbers from Step 2. Fetch their full content:

./.aitask-scripts/aitask_contribution_review.sh fetch-multi <N1>,<N2>,<N3>

Parse the structured output. Each issue is delimited by:

• @@@ISSUE:<num>@@@ — Issue separator
• TITLE:<title> — Issue title
• CONTRIBUTOR:<name> — Contributor username
• >>>BODY_START / <<<BODY_END — Body content boundaries

Extract code diffs from each issue body:

• Inline diff blocks (```diff sections)
• Hidden full diffs in <!-- full-diff:filename ... --> HTML comment blocks

Present summary table to user:

## Related Contribution Issues

| Issue | Title | Contributor | Source | Score | Changed Files |
|-------|-------|-------------|--------|-------|---------------|
| #42   | Fix X | @user1      | Both   | 7     | foo.sh, bar.sh |
| #38   | Add Y | @user2      | Overlap| 5     | foo.sh         |

Step 4: AI Analysis of Code Modifications

Read the actual diffs from all candidate issues plus the target issue. Analyze:

• Same files/functions touched? (strongest merge signal)
• Same bug fixed in different ways? (merge: pick best approach)
• Complementary changes? (merge: combine both)
• Unrelated despite fingerprint similarity? (don't merge)

Score thresholds for reference:

• Score >= 7: "high" overlap — very likely should be merged
• Score >= 4: "likely" overlap — worth investigating
• Score < 4: "low" overlap — probably unrelated

Generate a structured recommendation with rationale:

## Recommendation

**Action:** Merge issues #42, #38, #15 into one task
**Rationale:** All three issues modify the same functions in foo.sh...

Step 5: Present Proposal to User (AskUserQuestion)

If merge is recommended:

Use AskUserQuestion:

• Question: "Group these issues into one task: #42, #38, #15 — [brief rationale]"
• Header: "Import"
• Options:
  • "Import as merged task" (description: "Merge issues into a single task with combined diffs and multi-contributor attribution")
  • "Import only #" (description: "Import only the target issue as a single task")
  • "Skip" (description: "Don't import yet — cancel without creating a task")

If no merge is recommended:

Use AskUserQuestion:

• Question: "Issue # appears independent — no related contributions found worth merging"
• Header: "Import"
• Options:
  • "Import as single task" (description: "Import issue # as an aitask")
  • "Skip" (description: "Don't import yet — cancel without creating a task")

Step 5b: Check for Overlapping Existing Tasks

Execute the Related Task Discovery Procedure (see .claude/skills/task-workflow/related-task-discovery.md) with:

• Matching context: The contribution's title, description, areas (<areas> from Step 1), file paths (<file_paths>), and change type (<change_type>)
• Purpose text: "already cover this contribution's scope (they can be folded into the imported task or updated directly)"
• Min eligible: 1
• Selection mode: ai_filtered

If no overlapping tasks found: Proceed to Step 6 as normal.

If overlapping tasks found and user selected task(s): Use AskUserQuestion:

• Question: "How should the overlap with existing task(s) be handled?"
• Header: "Overlap"
• Options:
  • "Fold into new imported task" (description: "Import the contribution as new task and fold the overlapping existing task(s) into it")
  • "Update existing task instead" (description: "Add contribution content to the existing task — no new task created")
  • "Ignore overlap" (description: "Proceed with normal import, leave existing tasks unchanged")

Store the user's choice and the selected task IDs for use in Steps 6/6b.

Step 6: Execute Import

If "Update existing task instead" was selected in Step 5b: Skip this step and proceed to Step 6b.

Based on user selection:

Merged import:

./.aitask-scripts/aitask_issue_import.sh --batch --merge-issues <N1>,<N2>,<N3> --commit

Single import:

./.aitask-scripts/aitask_issue_import.sh --batch --issue <N> --commit

Skip: Display "No task created." and end workflow.

After successful import, display the created task file path from the script output.

If "Fold into new imported task" was selected in Step 5b:

After import completes:

1. Parse the import output to get the created task file path:
   • Single import: output contains Created: <filepath>
   • Merge import: output contains Merged N issues into: <filepath>
2. Extract the task number from the filename (e.g., t42 from aitasks/t42_foo.md)
3. Read the created task file's description body
4. Execute the Task Fold Content Procedure (see .claude/skills/task-workflow/task-fold-content.md) with:
   • primary_description: The imported task's description body
   • folded_task_files: File paths of each selected overlapping task from Step 5b
5. Update the imported task's description with the returned merged content:

   ./.aitask-scripts/aitask_update.sh --batch <new_task_num> --desc-file - <<'TASK_DESC'
   <merged description>
   TASK_DESC
   

6. Execute the Task Fold Marking Procedure (see .claude/skills/task-workflow/task-fold-marking.md) with:
   • primary_task_num: <new_task_num>
   • folded_task_ids: Selected overlapping task IDs from Step 5b
   • handle_transitive: true
   • commit_mode: "fresh"

Key constraint: The skill produces at most ONE task per invocation. To process multiple unrelated contribution issues, run the skill multiple times.

Step 6b: Update Existing Task with Contribution (Alternative)

This step is reached when the user chose "Update existing task instead" in Step 5b.

Select target task: If multiple overlapping tasks were selected in Step 5b, use AskUserQuestion:

• Question: "Which existing task should be updated with this contribution?"
• Header: "Target"
• Options: Each selected overlapping task (label: filename, description: brief summary)

If only one task was selected, use it directly.

Append contribution content to existing task:

1. Read the existing task file
2. Append a new section at the end of the body:

   ## Contribution from <contributor_name> (Issue #<N>)
   
   **Areas:** <areas from Step 1>
   **Files:** <file_paths from Step 1>
   **Change type:** <change_type from Step 1>
   
   <contribution description/body from Step 1>
   

3. Write the updated content back to the task file

Update task frontmatter:

./.aitask-scripts/aitask_update.sh --batch <task_num> --contributor "<contributor_name>" --contributor-email "<contributor_email>" --issue "<issue_url>"

Where <contributor_name>, <contributor_email> are from Step 1 metadata, <issue_url> is the contribution issue URL, and <task_num> is the target task number.

Post notification on the contribution issue:

./.aitask-scripts/aitask_contribution_review.sh post-comment <issue_number> "This contribution has been incorporated into existing task **t<task_num>** (<task_title>). The contributor will be credited via Co-authored-by when the task is implemented."

Commit changes:

./ait git add aitasks/<task_file>
./ait git commit -m "ait: Update t<task_num> with contribution from issue #<N>"

End workflow: Display "Contribution from issue #<N> incorporated into existing task t<task_num>. No new task created."

Notes

• This skill produces at most one task per invocation — ONE task group, never more
• No execution profiles are used (simple standalone skill)
• No handoff to task-workflow (import-only, no implementation step)
• Platform detection is handled by the helper script (aitask_contribution_review.sh), which sources aitask_contribution_check.sh for platform backends (GitHub gh CLI, GitLab glab+curl, Bitbucket curl)
• The --dry-run output from aitask_contribution_check.sh includes the full overlap comment with <!-- overlap-results --> machine-readable block
• For Bitbucket: no label-based issue filtering; overlap detection relies on metadata presence scanning
• Score thresholds: >= 7 "high", >= 4 "likely", < 4 "low"
• The --merge-issues flag in aitask_issue_import.sh requires at least 2 issue numbers
• Multi-contributor attribution: primary contributor (largest diff) gets Co-Authored-By trailer; others listed in commit body