workflow-gerrit-fix

# Gerrit Fix & Review Workflow Use this skill when the user asks to "fix the errors on Gerrit issue X" or "address reviewer comments on CL X" or similar requests targeting a specific Gerrit Change List (CL). ## Core Principles 1. **Orchestration**: Do not do the work yourself. Break down the task and delegate to subagents. 2. **Parallelism**: Always identify at least two parallel tracks (e.g., fixing known errors vs. inspecting for other issues, or addressing different comments). 3. **Isolation**: Always operate in a dedicated worktree for the CL. 4. **Reproduction**: Always attempt to reproduce the failure locally before applying a fix, to ensure the fix is verified. ## Protocol ### 1. Information Gathering - **CQ Results**: Use `gerrit_cq` (from `v8-utils`) to check Commit Queue status. - **Presubmit Errors**: Use `gerrit_get_presubmit_errors` (from `pndMcp`) to fetch detailed presubmit results. - **Comments**: Use `gerrit_comments` or `gerrit_list_change_comments` to read reviewer feedback or robot comments. This is critical when the goal is to address reviewer comments. ### 2. Workspace Setup - **Worktree**: Create a new isolated worktree using `agents/scripts/create_worktree.sh <cl_id>` for the CL to avoid polluting the main workspace. - **Checkout**: Patch the CL in the new worktree (e.g., `git cl patch X`). - **Rebase**: Rebase the branch to `origin/main` to ensure it builds against the latest code. ### 3. Reproduction & Triage - **Reproduction**: Instruct the subagent to reproduce the failure locally (e.g., run the failing test or compile target) to confirm the issue and provide a baseline for verification. - **Triage & Delegation**: Break down the identified errors or comments and delegate them to subagents. - **Track A (Fix/Address)**: Delegate the fixing of reported errors or addressing specific reviewer comments to a `self` subagent with full tool access. - **Track B (Scan/Verify)**: Delegate a scan of other modified files or running local tests to ensure no other issues were introduced. ### 4. Verification - **Local Build/Test**: Instruct the subagent to verify that the fix compiles locally and the reproduced failure is now resolved. - **Upload**: Once verified locally, upload intermediate fixes as WIP using `agents/scripts/upload_cl.sh cur check "<what you changed>"`, and once ready for review, run `agents/scripts/upload_cl.sh cur check "<what you changed>" --ready` as a background task to run a CQ dry-run and switch the CL to ready only if the dry-run passes (or revert to WIP if the dry-run fails). ## Example: Fixing Errors If the user says "fix errors on CL 12345": 1. Call `gerrit_cq` for CL 12345. 2. Identify that `v8_linux64_rel` failed with failing tests. 3. Create worktree `worktrees/cl-12345` and checkout CL. 4. **Spawn Subagent A to reproduce the failing tests locally.** 5. Spawn Subagent A to fix the failing tests. 6. Spawn Subagent B to check other modified files for potential issues. 7. Once Subagent A reports success and verification (tests pass), report to user and ask if you should upload. ## Example: Addressing Comments If the user says "address comments on CL 12345": 1. Call `gerrit_comments` for CL 12345. 2. Identify a comment requesting a refactor in `src/objects/lookup.cc`. 3. Create worktree `worktrees/cl-12345` and checkout CL. 4. Spawn Subagent A to perform the refactor as requested. 5. Spawn Subagent B to verify it compiles and run relevant unit tests. 6. Once Subagent A reports success, report to user and ask if you should upload.