Execution queue: improve Micro by completing and verifying real app tasks
Nobody has claimed this yet.
- Dominant language
- Go
- Stars
- 434
- Forks
- 21
- Avg merge
- 18m
- Merged PRs (30d)
- 326
Description
Owner: Codex execution queue, at Asim's request. Continue autonomously in small iterations; involve Asim only for a genuine access blocker, consequential product decision or action outside authorization. This is an active work queue, not another speculative architecture proposal.
Related: #1644, #1645, #1565, #1539.
Ordered work
- Diagnose the incomplete Assistant Copy task (7d3988d5-1b94-4867-a9bc-b20111515e2e). Inspect actual run/tool arguments, results, source access, provider termination and worker completion policy. Preserve evidence; distinguish unavailable tooling, execution errors, model stopping and incorrect task status.
- Reproduce the first demonstrated harness/tool deficiency with a focused regression; fix the smallest responsible layer. Do not special-case Copy or force unnecessary writes for read-only tasks.
- Ship through a reviewed PR: address comments, run relevant checks, self-merge under existing authorization and verify the deployed version. Keep go-micro pinned unless independently justified.
- Ask Micro to retry the original task. Independently inspect the published source and exercise the real feature, including streaming/follow-up, mobile and clipboard failure. Do not implement the app change ourselves and attribute it to Micro.
- If unsuccessful, record the next specific failure and repeat a bounded iteration. Do not blindly retry the same prompt indefinitely or create concurrent duplicate tasks.
- Verify the improvement with a different small app modification, then a personal fork, then one narrow multi-service app. Preserve originals, private data and clear rollback.
- Record the useful generalized regression cases and update #1644/#1645 with findings. Close only on independently demonstrated outcomes.
Per-iteration record
Request and acceptance criteria; task/run identity and model where observable; actual steps and failure; smallest fix and PR; verification evidence; next step or blocker. Keep private conversation contents and credentials out of GitHub.
Completion standard
Micro inspects real source, edits, tests, publishes and reports a verified result or an honest blocker. Tool success, final prose and a static app check alone do not establish completion. Work should recover safely after interruption and remain attributable.
Continuation
Use this issue as the durable checkpoint for scheduled Codex iterations. Read it and any active PR before starting work; finish an existing iteration first. Pause continuation when this bounded queue is complete or a real blocker requires Asim. Routine progress belongs here, not in repeated user questions.
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by reading this issue and related issues #1644, #1645, #1565, and #1539, then inspect the recorded run and tool arguments for the Assistant Copy task. Reproduce the first harness or tool deficiency with a focused regression and review any active PR before changing code. Done means a reviewed, deployed fix is independently verified through a real Micro task and the iteration evidence is recorded without exposing private data or credentials.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- go
- Domain
- ai, devtools, testing-qa
- Issue type
- Bug
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Active
- Clarity
- Needs clarification
- Newbie friendliness
- 20/100