Software engineer. This README documents how this account works, so that maintainers and reviewers know exactly what to expect from my pull requests.
I work with AI coding agents (agent-assisted workflow). The division of labor:
- 🧭 I choose the problem, the repository, and the scope.
- 🤖 The agent investigates the codebase, reads issues and docs, implements a candidate fix, and runs the test suite.
- 👀 I review every line of the resulting diff, reproduce the verification myself, and only then open the PR.
- ✍️ I take full responsibility for everything submitted under this account.
The agent never opens, updates, or merges a PR on its own. Every submission passes through my review first.
|
👤 Human accountable I can explain what every change does and how it interacts with the rest of the system, without asking the AI. |
📣 Disclosure Every PR states that it was prepared with AI assistance, naming the tool. |
🧩 Small & reviewable Small diffs, one problem per PR, tests included, low cadence. |
|
📜 Respect upstream rules Contribution guides, issue-first workflows, CLA/DCO — checked before code. |
🚫 No benchmark gaming Evaluation harnesses are never modified; results only via official channels. |
🤐 No marketing PRs carry no product or brand promotion. |
34 merged pull requests in other people's repositories, across 2 open-source projects. From 2026-07-27 to 2026-11-06.
Every one of them was reviewed line by line before submission, discloses its AI assistance in the body, and is my responsibility.
| PR | What | Size | Merged |
|---|---|---|---|
| #1762 | fix(doctor): say the exclude list already keeps an MCP server of your own | +47/-12 | 2026-11-06 |
| #1759 | fix(push): give every push branch a name nothing has yet | +26/-8 | 2026-11-06 |
| #1756 | fix(dashboard): give the benchmark Recall column its zh-CN label | +1/-0 | 2026-11-06 |
| #1740 | fix(benchmark): keep an arm's recall when a record is read back | +79/-5 | 2026-11-06 |
| #1729 | test(relations): cover a learning that lives in the learnings worktree | +78/-1 | 2026-11-05 |
| #1728 | feat(benchmark): give an arm a clone that deploys the team's recall | +175/-17 | 2026-11-05 |
| #1727 | fix(tool-launch): give a confined Codex judge the member's CODEX_API_KEY | +204/-1 | 2026-11-05 |
| #1726 | fix(update): report an interrupted install as interrupted, not as a failure | +92/-6 | 2026-11-05 |
| #1695 | feat(shared-instructions): accept agentsmd/ as a shared-instruction source directory |
+473/-49 | 2026-11-04 |
| #1679 | test(sessions): cover reading Qoder and Qoder CN transcripts (#1418) | +144/-0 | 2026-11-03 |
| #1637 | feat(lint,viz,stats): see asset relations — lint counts, health view, opens via relation (#1562, stacked) | +1613/-116 | 2026-11-02 |
| #1636 | feat(relations): derived rule-path edges and reconcile candidates (#1563) | +676/-19 | 2026-11-02 |
| #1629 | feat(recall): declared asset relations on recall hits, checked by review --branch (#1552 first slice) | +1612/-210 | 2026-10-31 |
| #1628 | feat(recall): record query text in the session recall-quality cache | +60/-5 | 2026-10-31 |
| #1555 | rank(search): ignore question words and normalize the title field | +268/-36 | 2026-10-30 |
| #1549 | rank(recall): prefer the codebase page the query names, and keep the top 3 source-diverse | +299/-3 | 2026-10-29 |
| #1547 | fix(hooks): record what a CodeBuddy session does, and what it loads | +289/-42 | 2026-10-29 |
| #1542 | fix(workflow): a stop of an attempt reaches the setup's process group | +89/-17 | 2026-10-29 |
| #1524 | fix(recall): rank an explicit docs match above a saturated codebase hit | +248/-17 | 2026-10-29 |
| #1522 | test(workflow): give the eight flaky workflow cases a budget that fits their real work (#1504) | +8/-4 | 2026-10-28 |
| #1521 | test(ci): forget a pid the runner cannot signal, and run the CodeBuddy gate on Windows (#1504) | +34/-3 | 2026-10-28 |
| #1497 | feat(hooks): run project team hooks on OpenCode and OMP (#1492) | +729/-28 | 2026-10-28 |
| #1465 | feat(worktree): one side-branch checkout per self-mode repo (#1443) | +610/-46 | 2026-10-28 |
| #1445 | fix(worktree): evict a stale side-branch checkout from either scope (#1443) | +140/-11 | 2026-10-27 |
| #1421 | fix(ci): read changes through the default branch and accept skin-tone rejections | +53/-6 | 2026-10-26 |
| #1417 | fix(import): preserve curated docs and domain metadata across imports | +366/-2 | 2026-10-26 |
| #1411 | fix(ci): preserve the instance scheme, paginate the MR lookup, address the project by path | +288/-257 | 2026-10-26 |
| #1321 | feat(ci): support GitLab in ci extract-mr comments and rejection marks | +645/-24 | 2026-10-26 |
| #321 | test(providers/gitlab): add opt-in live smoke test against real GitLab | +22/-0 | 2026-08-25 |
| PR | What | Size | Merged |
|---|---|---|---|
| #2430 | test(server): stop the durable-ack fallback cases from racing their own window | +18/-6 | 2026-11-05 |
| #2429 | test(server): give the lost-wake provider-models test real timing headroom | +9/-4 | 2026-11-04 |
| #2428 | test(web): pin Intl default locale in schedules-section-dom assertions | +20/-0 | 2026-11-04 |
| #2427 | fix(server): send non-ASCII attachment names as RFC 5987 filename* | +31/-6 | 2026-11-04 |
| #1975 | fix(kimi-code): pass KIMI_CODE_HOME to harness for config/credential resolution | +237/-7 | 2026-07-27 |
AI coding agents (Claude Code and similar), mainstream standard-compatible model APIs, sandboxed execution for untrusted upstream code.
Open an issue or discussion in the relevant repository — I read everything.