fix(kimi): adapt to newer Kimi Code CLI (0.28.1) flag and stream changes

[OctopusWen]

The newer Kimi Code CLI removed the standalone --work-dir, --quiet, --resume and --print flags, and changed the stream-json format:

• Probe kimi --help for --work-dir/--quiet/--resume in addition to --print, and only emit flags the installed binary advertises. Passing --work-dir to 0.28.1 errors with 'unknown option'.
• Fall back to cmd.Dir (already set to the session workDir) when --work-dir is unsupported; degrade --quiet to a debug log; use the hidden -r alias when --resume is gone.
• Accept assistant/tool content as a plain string, not just block arrays, so text is no longer silently dropped.
• Handle the meta session.resume_hint event the new CLI emits instead of the legacy plain-text resume line.

Summary

Type of change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (fix or feature that would cause existing behavior to change)
• Documentation only
• Internal refactor / chore (no user-visible change)

Testing

Automated tests added in this PR

• TestX_Y_Z in path/to/file_test.go — what it asserts

For bug fixes only — regression test

• Regression test name: Test...
• Manual verification this test catches the regression:
  • Reverted the fix locally; the regression test failed as expected.

Critical User Journeys (CUJ) impact

• No CUJ touched (small refactor, doc change, etc.)
• A — basic conversation
• B — session lifecycle (/new /switch /list /history etc.)
• C — agent execution control (/mode /cancel /stop permissions)
• D — security & permissions (allow_from admin_from banned_words rate limits)
• E — scheduled tasks (/cron /timer)
• F — config switching (/lang /provider /model reload)
• G — error handling & robustness (LLM failure, ws reconnect, agent crash)
• H — multi-platform / multi-project isolation
• I — UI rendering correctness (cards, streaming, display modes)

If any CUJ group is touched, confirm:

• go test ./core/ -run TestCUJ passes locally.
• If the change alters an existing user-visible flow, the corresponding
  CUJ test was updated (or a new CUJ added) to cover the new behavior.

Manual / user-visible behavior change

Checklist (reviewer will verify)

• go build ./... passes
• go test ./... passes (with -race if touching concurrency)
• AGENTS.md Pre-Commit Checklist items are satisfied
• No new hardcoded platform/agent names in core/
• i18n strings have all-language translations (if any new user-facing text)
• No secrets / credentials in source

Related

• Issue:
• Related PR:

[chenhg5]

Conclusion: Request changes

Overall assessment:

• The code is technically solid and well-scoped: probe-driven flag gating for --quiet / --resume / --work-dir, a meta role case for session.resume_hint, format-tolerant content parsers in both handleAssistant and handleTool, and six new black-box tests via buildArgs / handleEvent / flushPendingAsText. CI is 5/5 green, go test -race passes, go vet clean. However, this PR is functionally subsumed by #1564 — every change here is also in #1564 (with a CHANGELOG entry, a live_e2e_test.go, dual session-storage roots, state.json shape tolerance, and /list / /delete parity that this PR doesn't touch). Merging #1586 before #1564 would create duplicate code paths on the kimiFlagSupport struct and the -r fallback branch, requiring future reconciliation.

Review scope:

• Reviewed agent/kimi/probe.go, agent/kimi/probe_test.go, agent/kimi/session.go, agent/kimi/session_test.go (+272/-58 across 4 files).
• Focused on correctness, probe-based design consistency, and coverage vs the 4 concurrent PRs.

✅ What looks good:

• Pattern consistency with #1461 / #1483. All four gates (Print, WorkDir, Resume, Quiet) follow the same probe-driven shape. Reviewers who already approved those can verify in 30 seconds.
• -r fallback matches the kimi team's own resume_hint. The CLI advertises kimi -r <id> in the modern meta line; this PR picks -r over the more verbose --session (which #1566 picked instead — #1566 is the outlier here).
• Two-realistic-surfaces test fixtures (LegacyCLISurface + ModernCLISurface) pin the all-true and all-false probe surfaces symmetrically. The drainEvents helper at the bottom is clean.
• Meta resume-hint handler strictly checks type == "session.resume_hint"; TestHandleEvent_MetaResumeHint explicitly asserts unrelated meta doesn't clobber the session ID — exactly the right regression test.
• Surgical diff. No scope creep into other agents, no platform changes, no core changes.

🚨/🔴 Must fix:

• P1 — WorkDir half of this diff is a duplicate of #1483 (already merged). The struct field, the probe wiring, and the gate are all identical to what #1483 just shipped (merge commit 6c860798). The only delta is the expanded doc-comment in probe.go. Merging this half of #1586 would re-introduce the same struct field, which is harmless but pollutes git history.
  Impact: code duplication; future maintainers must reconcile which kimiFlagSupport shape is canonical.
  Suggested fix: if the maintainer chooses Strategy A (merge #1564, close #1586), no fix needed — close and credit the author. If Strategy B (merge #1586 as minimal fix ahead of #1564), coordinate with #1483 / #1601 / #1564 so the struct shape is consistent.
  Verification: gh pr diff 1586 -- agent/kimi/probe.go shows only the doc-comment delta vs main's #1483 result.

🟠 Should improve:

• P2 — CHANGELOG entry missing. Per cc-connect convention (#1461, #1483, #1564 all add a ### Fixed bullet). Flagging, not blocking — maintainer may prefer to consolidate into the eventual #1564 entry.
• P2 — No live_e2e_test.go. #1564 ships a KIMI_LIVE_E2E=1-gated live test that exercises the multi-turn resume path this PR is meant to fix. PR description doesn't claim live validation either. Without it, the CI-evidenced surface for the meta-hint path is only the unit test.
• P2 — Coordinate with highland0971 (#1601) on kimiFlagSupport.Quiet. Both this PR and #1601 add fields to kimiFlagSupport; the struct shapes must match for either to merge cleanly with #1564's broader struct. If both land independently, #1564's rebase will see conflicts.

🔵 Optional:

• P3 — drainEvents helper (last 10 lines of the diff) is added to session_test.go. If it's not reused elsewhere, consider moving to a testhelper_test.go package-private file to keep session_test.go focused.
• P3 — Doc comment in probe.go duplicates what #1483 already added. Most of the wording mirrors #1483's commit; cleanup if #1586 lands standalone.

❓ Questions:

• Q1 — Is the author aware of #1564 / #1566 / #1601? This is a first-time-contributor PR; the PR description doesn't acknowledge the overlap. A maintainer comment pointing at #1564 as the canonical fix (once rebased) would save the author follow-up effort and credit their work.
• Q2 — If the maintainer closes this in favor of #1564, can the author contribute the missing pieces (e.g. a live_e2e_test.go, or the --quiet gate in #1564's stack)? That's a way to credit the work without leaving duplicate code in main.

Testing / Risk:

• Verified locally: go test -race ./agent/kimi/... -count=1 -timeout 60s → PASS. go vet ./agent/kimi/... → clean. CI 5/5 SUCCESS at b635b8a8.
• Backward compat: legacy kimi-cli path unchanged when probe advertises all flags. The -r fallback is safe because legacy kimi-cli also accepts -r per the legacyKimiHelp fixture (which the diff extends).
• Remaining risk if merged as-is: code duplication with #1564 on the struct shape and the -r fallback.

Next step:

• Maintainer: decide between Strategy A (merge #1564 once rebased, close #1586 + #1566 + #1601) or Strategy B (merge #1586 / #1601 as minimal fix, then trimmed #1564). I'd recommend Strategy A — #1564's live e2e test is the strongest signal that the full #1561 surface is covered end-to-end.
• Author: if Strategy A, I'll close #1586 with a thank-you pointing at #1564. If Strategy B, please address the CHANGELOG + live e2e gaps and coordinate with highland0971 on kimiFlagSupport.Quiet.

[chenhg5]

@OctopusWen 谢谢 PR! 整体方向、probe-gating 模式和测试都做对了 — 跟 #1461 / #1483 风格一致, 也是用 -r 作为 resume fallback 这个细节也对 (跟 kimi 团队自己的 resume_hint 一致)。

不过有两个 maintenance 层面的考量想跟你说明:

核心问题: 这个 PR 的所有改动已经被 #1564 完整覆盖了 — #1564 多带了 4 件事: dual session-storage roots、state.json shape tolerance、/list / /delete parity、还有 live_e2e_test.go (KIMI_LIVE_E2E=1 gated)。如果两个 PR 都合, kimiFlagSupport struct 和 resume fallback 会有重复, 后面维护者要回头 reconcile。

#1586 跟 #1483 的重叠: 你新加的 WorkDir 字段和 gate 跟 #1483 (今天刚合并) 完全一样 — 那一半是 duplicate, 没新增价值。

两个方向供你选:

1. 关 PR 让 feat(kimi): native Kimi Code CLI dialect support (Fixes #1561) #1564 兜底 — 我会在 close comment 里注明你的贡献, owner 那边会看到。或者
2. 如果你愿意, 可以把缺失的几块 (live_e2e_test.go、CHANGELOG entry) 补到 feat(kimi): native Kimi Code CLI dialect support (Fixes #1561) #1564 的 PR 里 — 这样你的工作并入 canonical 修复而不是被覆盖。

倾向选哪个? 你说一声。

— cc-connect/qa-claudecode

[OctopusWen]

@chenhg5
谢谢详细 review！也感谢你开源并一直维护这么棒的工具。

方案 2：已把 #1564 rebase 到最新 main（去掉与已合并 #1483 的重复），补了 CHANGELOG #1561 条目和一个 #1586 的回归测试（TestHandleAssistantStringContentWithToolCalls），并验证在 kimi-code 0.36.1 上跑通了 live e2e（meta resume_hint + -r resume 锚点召回成功）。

已PR到 whyihaveyou 的分支：whyihaveyou#1。#1564 合并后我会关闭本 PR。

[chenhg5]

Closing in favor of #1564 — credit preserved

@OctopusWen 谢谢你的 PR! 整体方向、probe-gating 模式、测试都做对了 — 跟 #1461 / #1483 风格一致。

8-15 你主动选了方案 2:在 #1564 上加了 TestHandleAssistantStringContentWithToolCalls 回归测试 + CHANGELOG #1561 条目,kimi-code 0.36.1 上 live e2e 跑通。

结果 #1564 (现在 +795/-120, 14 commits, CI 5/5 ✓) 已经吸收了你的核心贡献:

• -r resume fallback (你的 main insight, 跟 kimi 团队 resume_hint 一致)
• flushPendingAsThinking 在 tool-use event 之前的顺序保护 (现在有你的 regression test 锁住)

这个 PR 作为独立 commit 历史保留在这里 — git log 里 b635b8a8 (#1586) 还是能查到你最初的工作。

#1564 已经过 maintainer fresh review 通过,可以合并。后续 issue / 新适配如果有 Kimi Code CLI 0.28+ 兼容需求,欢迎继续 push。

感谢对 cc-connect 的贡献! 🙏

— cc-connect/qa-claudecode