You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In the official harness vote (【Vote / 投票】Which Harness should codexhost support next? / codexhost 下一个支持哪个 Harness？ #59), ZCode was the top-voted candidate with 17 👍 (ahead of Kimi Code CLI at 10 and Google Antigravity at 8). When the vote was closed on 9-16, the closing note said further harness demand can be collected via new issues — this is that issue.
A complete ZCode integration previously existed and was removed from main (791f7252), preserved as the Draft PR feat: preserve ZCode integration for future redesign #337 pending redesign. The removal reasons (per feat: preserve ZCode integration for future redesign #337) were: the experimental Desktop backend required native remote-control pairing and a single workspace, and real official-account pairing / GLM conversation / CAPTCHA acceptance were not yet verified.
Demand signals from the vote thread
"增加ZCode就够用了" (+ suggest a harness-visibility config so the picker only shows selected agents)
"ZCode 带的 150% 额度，要是适配也能用上这多的额度就好了" — GLM Coding Plan users specifically want to use their ZCode-side quota bonus
Multiple additional "zcode" votes in comments
For CN users on GLM Coding Plan, ZCode is the official GLM harness, and CodexHost currently has no GLM-native option.
Proposal: re-introduce ZCode in stages
Stage 1 — baseline stdio harness: restore the stdio adapter from feat: preserve ZCode integration for future redesign #337 as a regular external harness (API key / GLM Coding Plan endpoint), explicitly scoped: no native account pairing yet; unsupported capabilities (Fork / Rollback / Subagent transcript) declared per the existing capability-boundaries mechanism instead of being silently absent. This already covers the main vote-thread ask (quota-backed GLM sessions inside Codex Desktop).
Stage 2 — native account pairing: official account pairing / CAPTCHA acceptance and the Desktop-backend experience, iterated after the baseline is stable.
This is complementary to docs/proposals/zcode-independent-client-feasibility.md: that track is about a future standalone client (UI layer), while harness-layer access serves current Codex Desktop users now. #337 already holds the preserved implementation to build from.
Environment for testing
Windows 11, CodexHost 0.9.1 (installer + npm CLI 0.9.1), Codex Desktop 26.915.4065.0, GLM Coding Plan available — happy to test any reintroduction build on Windows.
Context
791f7252), preserved as the Draft PR feat: preserve ZCode integration for future redesign #337 pending redesign. The removal reasons (per feat: preserve ZCode integration for future redesign #337) were: the experimental Desktop backend required native remote-control pairing and a single workspace, and real official-account pairing / GLM conversation / CAPTCHA acceptance were not yet verified.Demand signals from the vote thread
For CN users on GLM Coding Plan, ZCode is the official GLM harness, and CodexHost currently has no GLM-native option.
Proposal: re-introduce ZCode in stages
This is complementary to
docs/proposals/zcode-independent-client-feasibility.md: that track is about a future standalone client (UI layer), while harness-layer access serves current Codex Desktop users now. #337 already holds the preserved implementation to build from.Environment for testing
Windows 11, CodexHost 0.9.1 (installer + npm CLI 0.9.1), Codex Desktop 26.915.4065.0, GLM Coding Plan available — happy to test any reintroduction build on Windows.