Credits
The people behind the features you are using, and what each of them built.
Made and maintained by @khalilgharbaoui (Khalil Gharbaoui). This plugin absorbs work from its forks directly, cherry-picked with the original authorship preserved or reimplemented with the author named in the commit, rather than waiting on pull requests. The people behind the features you are using:
| Who | What | Where |
|---|---|---|
| @khalilgharbaoui (Khalil Gharbaoui) | Maintainer. Everything after the first version: the opencode-side tool proxy and its broker, accounts and failover, the model registry, the background subagents, the skill bridge, the doctor and its bundle, the two-major support, the tests and the measurement discipline in AGENTS.md, this site. Reviews, merges and credits every contribution. |
git log --author on this repository |
| @dkraemerwork (Dennis Krämer) | Wrote the proof of concept this plugin descends from: a provider built into opencode itself that drives the local claude CLI, January 2026. It closed upstream, and the first version of this plugin was built from its provider files, which is why getClaudeUserMessage, compactConversationHistory, mapTool and mapToolInput still carry its names. |
anomalyco/opencode#10097, the patch |
| @unixfox (Émilien) | Wrote the first version of this plugin in March 2026, and with it the idea it still runs on: an opencode provider that drives the official claude CLI as a subprocess instead of calling the API. This repository began as a fork of that work, and its tool mapping, session manager and message builder still carry his code. |
b03fa8e (the initial commit) and four more, to 2026-04-06 |
| @galvani (Jan Kozak) | Per-session working directory for opencode serve, so one server spawns each project’s claude in the right place. Also found the stale toolCallMap re-emission three months before it was fixed here. |
9e02ce4, 2238ed0 |
| @HeikoAtGitHub | Stopped sending AGENTS.md to the model twice (opencode already forwards it). Independently diagnosed the 5-minute proxy wall. |
25260a4, 42f426d |
| @bernardofortes (Bernardo Fortes) | idleProcessTimeoutMs, idle eviction of retained claude workers. |
a5f723a |
| @broskees (Joseph Roberts) | Task proxy default-on (PR #18), the abort interrupt so Esc really stops the CLI, the skill bridge, task_batch for concurrent subagents (and the measurement that the CLI serialises MCP calls), the undici 300 s diagnosis of the proxy wall, the lifecycle release of proxied calls that made the task deadline unnecessary (PR #36), Claude Opus 5.5 with its fast-mode entry (PR #43), the fix for turn-summed usage that made opencode auto-compact far below the window (PR #63), accountFailover: "auto", the switch the form makes, made without the form inside the same turn (PR #115), and Claude Fable 5.2 and Mythos 5.2 registered the day they shipped (PR #173). |
PR #18, 68ed142, PR #36, PR #43, PR #63, PR #115, PR #173 |
| @valioox | Traced opencode 2’s ENOENT ... posix_spawn to the session shim appending the project-relative subpath to a directory that already was the session’s own, with the exact fields and the fix (issue #113). |
issue #113 |
| @stech-mostafa | Reported the context loss when switching between the plan and build agents in one conversation, which turned out to be a reused claude process never being told what the other agent’s turns said (issue #114). |
issue #114 |
| @clementrahula | Found that the text replay of a long session had no total size and failed the request as prompt-too-long, with the fix shape (issue #116), and that changing an agent’s cacheTtl or effort mid-session left two claude processes serving one chat (issue #140). |
issue #116, #140 |
| @jknlsn (Jake Nelson) | Per-tool proxy timeouts, subagent dispatch steering, the question proxy, the start watchdog respawn. | 84f3db9, 94980a6, 47501d0, ffefc24 |
| @CollieIsCute (Collie Tsai) | The plan-mode approval bridge. | 8c5b583 |
| @flupkede | The compress proxy tool design and the AI-SDK v4 image-part fix. | 4ac319f, 60a6e9a |
| @CNQQC | Cost units corrected to dollars per million tokens (PR #25). | PR #25 |
| @willmcginnis | The proxy endpoint authentication (PR #28, GHSA-3mxm-w7gf-3c5x). | PR #28 |
| @nic-lan (Nicolas Languille) | The issue #29 diagnosis of subagent output lost across the CLI resume boundary, the fix for unattended output replaying as one text block per delta (PR #35), and bridgeMcpOauthTokens: giving a bridged remote MCP server the OAuth token opencode already holds, so a server the operator authenticated in opencode stops reporting needs-auth to the Claude CLI for the whole session. |
#29, PR #35, 5001cff |
| @hmjBill | Two precise reports from real long sessions: a usage-limited compaction stored the CLI’s limit sentence as the conversation summary (issue #90), and an effort or model change, a restart or a crash dropped the Claude session into a lossy transcript replay (issue #91). | #90, #91 |
| @acastro2 (Alexandre Castro) | Found and fixed CLI tool results being emitted under a different name than their call, which made opencode 2 abort every turn that used a Claude-side MCP server (PR #46). | PR #46 |
| @bangnh1 | Independently found and diagnosed the turn-summed usage that tripped auto-compaction after a single prompt, measured on opencode 2 (PR #62; the fix landed as PR #63), and fixed opencode 2’s MCP config layout (mcp.servers, disabled, providers.<id>.settings) with opt-in Code Mode execute proxying (PR #67). Found that a replayed thread kept each message’s tool output and dropped the answer behind it, and fixed it by clipping the middle instead (ae82650, taken in part). Solved the agent-switch catch-up independently, and checked the accountGroups group on every catch-up where this plugin checked it only on the fresh-start turn: that one rule was a real leak here, in two places, and is where the guard now lives (c614345, the idea taken). |
PR #62, PR #67, ae82650, c614345 |
| @JWebCoder (joao moura) | Diagnosed that auto-continue never fires on current CLIs (PR #15). | PR #15 |
| @internetisalie (Michael Crawford) | Found and fixed, each measured over days of real transcripts, two ways a proxied tool call went wrong: a session in another workspace read as idle, so opencode’s routine abort at a tool boundary rejected the call and told the model the user had interrupted it (PR #89), and user messages opencode promoted beside a tool result (background-PTY notices, steered prompts) never reached the CLI (PR #88). | PR #88, PR #89 |
| @taylormitchell (Taylor Mitchell) | Forwarded opencode 2’s per-call subagent model through the proxied task and task_batch, so a model you ask for in the conversation reaches the subagent, and kept it off opencode 1, whose task tool has none (PR #123). |
PR #123 |
| @army113 (David Sluga) | Found that the provider factory dropped the Windows account flag, so on Windows every configured account ran on the default login, and fixed it with a test that builds a model through the factory (PR #158). Answered opencode 2’s detached generate calls, which OpenChamber’s rename and walkthroughs use, with a lean claude -p instead of a full turn (leanGenerate, PR #171), and traced every opencode 2 session title reading “Here First Message Conversation” to the title request’s lead-in, with an opt-in model that names sessions properly (titleModel, PR #172). |
PR #158, PR #171, PR #172 |
| @effectustasi | Found that on Windows a switch off the default account, by a failover or by a subagent dispatched to another account, kept the default login and reused its claude process, and fixed both halves (PR #168). |
PR #168 |
Commit hashes are on the contributors’ forks where the work was cherry-picked; git log --author on this repo shows the preserved authorship.
Star History
Free and MIT-licensed, made by @khalilgharbaoui. If the plugin saves you time, you can buy its maintainer a coffee.
