clawq-code-maintenance-2

code maintenace | have a family of subagents (5+) crawl over each component of clawq and look for bugs and issues. They should compile a list incrementally as they go (saving it outside of git as a temporary markdown file) and report back to you. you should then compile a master list which references (by filename and line number) the issues in the reports from subagents. You are then to investigate, research, and plan out (write the plan to a temp markdown file) fixes for ALL legitimate issues found and raised by the subagents.

1. write general prompt (see skills like review-and-fix, ultra-code-quality, and any bug/review related skills to learn what to write) about doing code review in clawq. Then, start each subagent and give it the scope of the review, the general prompt file path, and any other relevant info. specify the output file path for the subagent to write their findings to incrementally (append only).
2. start a tail -f monitor on each agent's output file. you can contact subagents via c2c if you have input or need to steer or whatever.
3. compile a master list incrementally; organize it by conceptual group/area so that the areas are (mostly) disjoint. Thus, issues are grouped in a way that is more efficient to batch fix.
4. when all subagents are done, start a subagent to run the review-and-fix skill over the master list and the documents created by the subagents. goal: ensure the master list is complete.
5. for each conceptual group in the master list, start a subagent to run the pirfl skill and fix all of the issues in that group. each subagent should create a new worktree under .worktrees for their work. They should commit at regular convenient intervals. As part of pirfl, they should run their own review-and-fix loop when they are finishing up with their work. Good. Tell them at this point they should contact you and ask if it's okay to merge to master and clean up their worktree. When they do this, they should run sleep 20 via bash tool repeatedly (up to 45 times) until you message them back. This is the 'sleep-20-x-45 loop'. it's crucial that they run these sleeps one at a time so that incoming messages have a chance to be appended to the chat log after a tool call. Before you approve them to merge to master, at this time (but not before), instruct them to run the review-and-fix skill in a loop for a second time but crucially to use openai-codex/gpt-5.5 subagents for the reviewer subagent (and the fixer subagnet should have the model left blank). they must provide the reviewer subagent with the path to the master list + the group they're working on. When this final review-and-fix loop is done (returns PASS), they should contact you and enter the sleep 20 x 45 loop again waiting for a response. Provided it's not an issue, you should approve them and they should automatically merge their changes back to master and clean up the worktree.
6. when all that is done, start a review-and-fix loop yourself, use openai-codex/gpt-5.5 as the reviewer model and ask it to review the whole master list for completeness. Fix all issues that are found by starting a fixer agent and instructing them to use the pirfl skill to fix the issue (and provide a detailed description of the issue); leave model blank for the fixer model. Continue loop until review PASSes.

subagent model: leave blank unless explicitly specified for that subagent.