CB-530..536, CB-538 groundwork: leads as peers, and a fleet that can find itself

Two leads now work as peers rather than one primary plus workers. The arc: CB-530/531 lead identity: `leaders:` names panes, `leadScan:` discovers them by tab label (LeadTabScanner, TTL-cached, worker spaces excluded). CB-532 leads can message each other AND be answered. Principal.leader now carries its terminal, so ownsSession() can be true for a lead; the "and you must be a worker" conjunct beside it protected nothing. Retires `primary:` — reply nudges follow the delegating lead, a binding recorded at bridge_send where both halves are known. CB-533 ClaudeCodeLauncher passes --model. argv is usually a wrapper (`ccs <profile>`) that re-exports its own model family, so ANTHROPIC_MODEL alone was silently overruled. CB-534 a lead is deliverable. The CB-113 readiness gate only opened for terminals in WorkerPresence, which only workers ever enter, so every lead->lead send waited out the ~60s grace and failed having never been typed. The gate guards a *spawned* peer's boot window; a lead is never spawned. CB-535 bridge_list returns `leads` alongside `workers`, with `self` on the caller's row. An empty worker roster no longer reads as "no peers". CB-536 CLAUDE.md: lead<->lead is coordinate-only, never sideways delegation. Propagated byte-identically to wiki/7-Use-Cases.md. MIXED PROVENANCE — recorded deliberately rather than hidden. This tree also carries in-progress CB-537 (context separation) authored by the peer lead gpt-sol-5.6 and its worker: Capability.CONTEXT_RESET, SessionManager.clearAfterTurn, and the Injector/TurnListener/CompletionResolver/launcher changes around it. That work was done in this shared working tree rather than a worktree, and is entangled with the above in BridgedConfig.java, Bridged.java and ClaudeCodeLauncher.java, so neither lead could stage its own half without sweeping in the other's. Committing the whole green state is the honest resolution; the peer branches from here. Note for whoever picks CB-537 up: the design in this commit is SUPERSEDED. Both leads agreed to replace the global `clearAfterTurn` boolean with per-delivery policy (inherit|fresh|thread) applied PRE-delivery, because a post-turn reset races by construction — Injector.onStatus clears awaitingCompletion and dequeues the next message in the same tick. `fresh` is also a correctness guarantee, so an adapter without a reset capability must refuse it rather than log a no-op. mvn clean install: Tests run: 464, Failures: 0, Errors: 0, Skipped: 0. BUILD SUCCESS.
2026-08-13 09:38:24 +02:00
parent ef1e014b41
commit bc13b8e92c
34 changed files with 2350 additions and 115 deletions
@@ -0,0 +1,172 @@
+---
+name: port-to-opencode
+description: Make an OpenCode session a first-class participant in a Claude Code workspace — instructions, MCP servers, and secrets — without duplicating config. Use when onboarding opencode to a project that already has CLAUDE.md and .mcp.json, or when an opencode peer needs the same tools and rules as the Claude session.
+---
+
+# Porting a Claude Code workspace to OpenCode
+
+**The headline: there is almost nothing to port.** OpenCode reads `CLAUDE.md` natively. The only
+artifact you create is one `opencode.json` mapping MCP servers. Do not translate instructions, do
+not generate a second rules file, and do not install a sync tool — every one of those makes the
+workspace worse.
+
+Everything below was verified against `opencode 1.18.16` and the OpenCode docs.
+
+## 1. Know what you get for free
+
+OpenCode's instruction search order:
+
+```
+1. walking up from cwd:   AGENTS.md , then CLAUDE.md
+2. global:                ~/.config/opencode/AGENTS.md
+3. Claude Code global:    ~/.claude/CLAUDE.md          (unless disabled)
+```
+
+*"The first matching file wins in each category."*
+
+Consequences that decide the whole procedure:
+
+- **A project `CLAUDE.md` is already read.** No port needed.
+- **Your user-level `~/.claude/CLAUDE.md` is already read too.** Global preferences carry over.
+- **An `AGENTS.md` in the repo SHADOWS `CLAUDE.md`.** If one exists from a previous Codex port,
+  **delete it** — otherwise opencode reads the stale translated copy instead of the real rules.
+  This is the single most likely way to get this wrong.
+
+## 2. Create `opencode.json` for MCP servers only
+
+Project config lives at `opencode.json` in the repo root; the global one is
+`~/.config/opencode/opencode.json`. **Configs merge, they do not replace** — so machine-local
+servers belong in the global file and shared ones in the project file.
+
+Map each entry from `.mcp.json`:
+
+| `.mcp.json` | `opencode.json` |
+|---|---|
+| `"type": "http"` / `"sse"` | `"type": "remote"`, `"url"` |
+| `"type": "stdio"` | `"type": "local"`, `"command": ["bin", "arg"]` |
+| `"command"` + `"args"` | single `"command"` array |
+| `"env"` | `"environment"` |
+| `"headers"` | `"headers"` |
+
+```json
+{
+  "$schema": "https://opencode.ai/config.json",
+  "instructions": ["CLAUDE.md"],
+  "mcp": {
+    "bridged":  { "type": "remote", "url": "http://127.0.0.1:8765/mcp", "enabled": true },
+    "context7": { "type": "remote", "url": "https://example.dev/mcp", "enabled": true,
+                  "headers": { "Authorization": "Bearer {env:CONTEXT7_TOKEN}" } },
+    "gitea":    { "type": "local", "command": ["gitea-mcp", "-t", "stdio"], "enabled": true,
+                  "environment": { "GITEA_ACCESS_TOKEN": "{env:GITEA_ACCESS_TOKEN}" } }
+  }
+}
+```
+
+Set `instructions` explicitly even though `CLAUDE.md` is found anyway — the fallback only applies
+while no `AGENTS.md` exists, and being explicit survives someone adding one later.
+
+## 3. Reference secrets, never embed them
+
+OpenCode substitutes at load time, in both `headers` and `environment`:
+
+```
+{env:VARIABLE_NAME}      value from the environment
+{file:~/.secrets/token}  value read from a file
+```
+
+**No credential ever belongs in `opencode.json`.** With `{env:…}` there is no reason to write one,
+which is what makes this file safe to commit — and it must be committable, because a peer running
+in a git worktree receives tracked files only.
+
+**But `{env:…}` reads OpenCode's *process* environment — and OpenCode has no env store of its
+own.** A Claude Code `env` block in `~/.claude/settings.json` or `.claude/settings.local.json` does
+**not** reach it: those files are Claude Code's, and the variables exist only inside processes
+Claude Code spawned. Verify with a clean login shell, not the shell your agent hands you:
+
+```bash
+env -u MY_TOKEN zsh -lc 'echo "${MY_TOKEN:-NOT IN PROFILE}"'
+```
+
+So `{env:…}` only works if something puts the variable in the environment first. Pick the supply
+route by who launches opencode:
+
+| Launcher | Route |
+|---|---|
+| a human, from a terminal | `{file:…}` — see below |
+| a spawner (bridge, CI, IDE) | `{env:…}`, with the spawner injecting the variable |
+
+For the human case prefer **`{file:…}` with a workspace-relative path**, kept in a gitignored
+`.secrets/` directory beside `opencode.json`:
+
+```json
+"headers": { "Authorization": "Bearer {file:.secrets/api-token}" }
+```
+
+Verified: opencode resolves relative `{file:}` paths against the project root, so this needs no
+shell setup at all — no rc export leaking the secret to every process, no direnv dependency.
+
+**The catch, and state it out loud:** `.secrets/` is gitignored, so a peer running in a git worktree
+does **not** get it — worktrees receive tracked files only, the same rule that makes `opencode.json`
+itself worth committing. Spawned peers must therefore be fed through `{env:…}` by whatever launches
+them. Check the variable *names* match: a spawner often injects under a different name than your
+shell uses, and the config has no fallback.
+
+## 4. Do not port machine-local MCP servers
+
+IDE indexes, language servers, editor bridges — anything bound to *your* checkout — stay out of the
+project file. Put them in `~/.config/opencode/opencode.json` if you want them personally.
+
+A committed project config reaches every worktree. A worker that mounts servers whose paths point
+into the primary's checkout will edit the primary's files while building in its own — every build
+passes, every change lands in the wrong tree.
+
+Port the servers the work needs. Leave the rest.
+
+## 5. Verify against the running agent, not the file
+
+A file on disk proves nothing about what the agent loaded.
+
+```bash
+opencode run "In one line: state a rule from this project's instructions."
+```
+
+The answer must reflect the actual `CLAUDE.md`. If it answers generically, the instructions did not
+reach the model and everything after this is built on sand.
+
+Then confirm the tools are mounted:
+
+```bash
+opencode mcp list
+```
+
+**"connected" does not mean "working".** A stdio server with a missing credential still completes
+the MCP handshake and reports green; only a real tool call reveals it. Verified: `gitea` showed
+`✓ connected` with no token, then failed the first call with `token is required`. A remote server
+is more honest (`⚠ needs authentication`), but do not rely on that difference — **exercise one
+authenticated tool per server**:
+
+```bash
+opencode run "Call <server>'s <tool>. Report the result or the exact error. One line."
+```
+
+Check too that no server you deliberately withheld is present.
+
+## 6. Report
+
+State what you changed, which servers crossed and which you withheld and why, and quote the
+verification answer verbatim. If any server failed to connect, say so plainly — a partially mounted
+peer is worse than a missing one, because it looks configured.
+
+## Gotchas
+
+- **A leftover `AGENTS.md` silently wins over `CLAUDE.md`.** Check for one before anything else.
+- **`opencode.json` is merged, not overridden** — a global entry and a project entry with the same
+  server name both matter; keep names distinct unless you intend to layer them.
+- **`enabled: false`** turns a server off without deleting its config — prefer it over removal when
+  you may want the server back.
+- **OAuth-based servers** store tokens in `~/.local/share/opencode/mcp-auth.json` after
+  `opencode mcp auth <server>`; that is machine state, never config to commit.
+- **Skills and subagents do not port.** OpenCode uses its own agent markdown under
+  `.opencode/agents/`; `.claude/skills/**` is not read. If a delegation brief tells a peer to load a
+  skill by name, that instruction has no effect on an opencode peer — spell the procedure out in the
+  brief, or author the equivalent agent file.