Compare commits

..

4 Commits

Author SHA1 Message Date
Dai Ha e81944cef6 CB-567: compose charters per spawn
CI / build (pull_request) Successful in 57s
CI / contract (pull_request) Successful in 1m3s
2026-08-15 05:00:48 +02:00
Dai Ha 7bdd39ab9a CB-566 follow-up: say why the five-arg Fleet constructor is kept
CI / contract (push) Successful in 43s
CI / build (push) Successful in 1m8s
Reviewing the merge I read the five-arg constructor as dead code and
removed it. That was wrong: the tests call it as BridgedConfig.Fleet,
which my grep for 'new Fleet(' did not match, and the build failed on
eight call sites. It is restored with a javadoc that says why keeping it
is safe here even though an overload that drops a new field is normally
the shape to avoid — nothing reads a charter through a constructor, and
Jackson binds the canonical one, so it cannot swallow an operator's YAML.

Also drop a redundant java.util.Arrays qualifier (the class is already
imported) and rewrap a javadoc line the change had left over-long.
2026-08-15 04:54:09 +02:00
Dai Ha f4b38f040e Merge CB-566: the charter config surface (PR #37)
fleet.charters is a validated Map<String,String>, not a record: Fleet is
@JsonIgnoreProperties(ignoreUnknown = true), so a record field named
architetc would be dropped in silence and the operator would never learn
of the typo. A map lets validateCharters see the bad key and refuse it.

The key sits under fleet: because ConfigRef already treats that block as
hot and changedDeferredKeys does not list it. A new top-level key would
inherit nothing, and forgetting to classify it means a reload prints
'config reloaded' and does nothing.

A blank value is refused while an absent one is fine: an absent key means
the operator configured no charter, a blank one means they tried and
failed. Refusing at both startup and reload is the point — wiring only
one of the two paths is the whole bug.
2026-08-15 04:52:48 +02:00
Dai Ha 890190263e CB-560/562/563: the shipped block tells the truth again
The canonical CLAUDE.md block is the instruction surface this daemon ships to
every agent that mounts it, so a code change that silently invalidates it is an
incomplete change. CB-548 added a third principal kind and the block was never
revisited. Four statements in it were simply false:

  * whoami was documented as returning only primary or worker;
  * "spawn/stop/send/drain are lead-only" — Authz permits SEND to an architect;
  * delivery was documented as idle/blocked — injectable() is IDLE|BLOCKED|DONE,
    and a spawned member must also have mounted the bridge MCP, which is the
    exact condition that made every architect undeliverable for a day;
  * the tool table said bridge_list returns `workers` — the JSON key is
    `members`.

Also corrected: the fallback ladder claimed each one-way signal identifies a
"worker", but an architect gets the same charter, the same mount and the same
env, so those signals identify a spawned member and only bridge_whoami
separates the two. The safe default stays "act as a worker" — it is the most
restricted member role.

The turn contract now covers both member kinds, and says why the completion
fallback is not a substitute for bridge_reply: it returns at most the last 4000
characters, so a long report reaches the lead with its end cut off. That is not
hypothetical — it happened twice today.

Found by a reviewer asked whether the block still matches the code. Verified
against injectable(), Authz, BridgeMcp.listFleet and LeadLauncher before
applying. The wiki template is updated in the same shape and re-checked
byte-identical.
2026-08-15 04:37:03 +02:00
10 changed files with 200 additions and 186 deletions
+39 -31
View File
@@ -11,41 +11,45 @@
If no `bridge_*` MCP tools are mounted in this session, this section does not apply — skip it.
`bridged` is the **sole communication gateway** between agents here. The orchestrating session (the
**primary**) and every delegated peer (a **worker**) mount the *same* MCP server and talk only
**primary**) and every delegated peer (a **member**) mount the *same* MCP server and talk only
through its `bridge_*` tools. No session addresses a peer, a broker, or the network directly.
### Which role am I? — settle this before acting
**Both roles read this file.** A worker runs in a git worktree of this same repo, so it inherits
**Every role reads this file.** A member runs in a git worktree of this same repo, so it inherits
this `CLAUDE.md` verbatim, and every rule below is role-conditional.
**Call `bridge_whoami`.** It returns `{"role":"primary"}` or `{"role":"worker","sessionId":…,
"profile":…,"worktree":…,"branch":…}`, resolved by the daemon from your connection — unforgeable,
and the same resolution its authorization gate uses. Don't infer what you can ask.
**Call `bridge_whoami`.** It returns `primary`, `worker`, or `architect`, resolved by the daemon from
your connection — unforgeable, and the same resolution its authorization gate uses. A worker also
carries its `sessionId`, `profile`, `worktree` and `branch`; an architect carries the slot name it
was bound to. Don't infer what you can ask.
Only if that call is unavailable, fall back to these — each is one-way, so keep reading until one
fires: the reply charter in your system prompt (*"You are an off-subscription worker in the
claude-bridge fleet"*) ⇒ **worker**; bridge tools prefixed `mcp__bridge__*` ⇒ **worker** (the
launcher fixes that mount name; a primary's mount is named by whoever wrote its `.mcp.json`, so it
varies); `ANTHROPIC_BASE_URL` set ⇒ **worker** (Claude-model workers run on a clean env, so its
*absence* proves nothing). **Still unsure ⇒ act as a worker.** The two mistakes are not symmetric: a
primary acting as a worker is refused by the authorization gate — loud and self-correcting — while a
worker acting as the primary ends its turn with no `bridge_reply`, and the sender silently receives
nothing. Fail toward the recoverable error.
claude-bridge fleet"*) ⇒ **spawned member**; bridge tools prefixed `mcp__bridge__*` ⇒ **spawned
member** (the launcher fixes that mount name; a primary's mount is named by whoever wrote its
`.mcp.json`, so it varies); `ANTHROPIC_BASE_URL` set ⇒ **spawned member** (Claude-model members run
on a clean env, so its *absence* proves nothing). None of these separate a worker from an architect —
only `bridge_whoami` does. **Still unsure ⇒ act as a worker**, the most restricted member role. The
two mistakes are not symmetric: a primary acting as a worker is refused by the authorization gate —
loud and self-correcting — while a member acting as the primary ends its turn with no `bridge_reply`,
and the sender silently receives nothing. Fail toward the recoverable error.
### Invariants — both roles, no exceptions
1. **Never set, export, or forward `ANTHROPIC_BASE_URL`** (or `ANTHROPIC_AUTH_TOKEN`). The primary
stays on subscription; only the bridge puts a worker off it, at spawn. Mounting the bridge must
stays on subscription; only the bridge puts a member off it, at spawn. Mounting the bridge must
never move a session across that boundary.
2. **The bridge is the only channel.** Text you print in your terminal reaches nobody — the other
side cannot see your screen. An answer that isn't in a `bridge_*` call is silently discarded.
3. **Identity comes from the connection, never an argument.** Workers never pass a target; you
cannot act as another session. Spawn/stop/send/drain are lead-only; reply/ask are
only-as-itself — any peer may answer for its own pane, and for no other. A call outside your
role is refused, not queued.
cannot act as another session. Spawn/stop/drain are lead-only; **send is lead or architect**;
reply/ask are only-as-itself — any peer may answer for its own pane, and for no other. A call
outside your role is refused, not queued.
4. **Delivery is status-gated: one message per turn.** Don't busy-poll a peer's terminal and don't
re-send because a call looks slow — the bridge delivers when the peer is `idle`/`blocked`.
re-send because a call looks slow — the bridge delivers when the peer is `idle`, `blocked` or
`done`. A spawned member must **also** have mounted the bridge MCP: until it has, it is not
deliverable, and a send waits on that gate for ~60s and then fails without ever reaching its pane.
5. **Never drive the terminal multiplexer directly** (no `herdr` CLI, no socket). The bridge owns
policy; the multiplexer owns PTYs. Going around the bridge bypasses every rule above.
@@ -101,26 +105,26 @@ the merge — and merging on a reviewer's word is delegating it by proxy.
|---|---|
| Confirm your own role | `bridge_whoami` |
| See backends available | `bridge_profiles` |
| Start a worker | `bridge_spawn{profile?, cwd?, worktree?, ticket?}` → `sessionId` + `paneId` |
| See the fleet | `bridge_list` → `leads` (your peers) + `workers` · one peer's state: `bridge_status{sessionId}` |
| Start a member | `bridge_spawn{role?, profile?, cwd?, worktree?, ticket?}` → `sessionId` + `paneId` |
| See the fleet | `bridge_list` → `leads` (your peers) + `members` · one peer's state: `bridge_status{sessionId}` |
| Delegate (blocking) | `bridge_send{sessionId, content}` |
| Delegate (long task) | `bridge_send{sessionId, content, wait:false}` → ticket → `bridge_poll{ticket}` |
| Answer a worker's `bridge_ask` | `bridge_send{turnId, content}` — **not** `sessionId` |
| Answer a member's `bridge_ask` | `bridge_send{turnId, content}` — **not** `sessionId` |
| Message a **peer lead** | `bridge_send{sessionId: <their terminal>, content}` — `bridge_list` → `leads` reports it. Coordination only, **never** a task |
| Answer a peer lead that messaged you | `bridge_reply{content}` — the one case a lead replies |
| Collect a held reply | `bridge_poll{target}` · then `bridge_ack{target, msgId}` |
| Tear down | `bridge_stop{paneId}` |
| Tear down a member | `bridge_stop{paneId}` |
### Lead ↔ lead — coordinate, never delegate
`bridge_list` returns `leads` alongside `workers`; your own row carries `self: true`. Every other row
is a peer — an orchestrator with its own context, its own workers, and its own judgment. An empty
`workers` array means no workers are spawned; it says nothing about peers.
`bridge_list` returns `leads` alongside `members`; your own row carries `self: true`. Every other row
is a peer — an orchestrator with its own context, its own members, and its own judgment. An empty
`members` array means no members are spawned; it says nothing about peers.
**A lead never assigns a task to another lead.** Work goes to workers — only ever downward, never
sideways. Sending a peer a brief with acceptance criteria is a category error: a brief is a worker's
**A lead never assigns a task to another lead.** Work goes to members — only ever downward, never
sideways. Sending a peer a brief with acceptance criteria is a category error: a brief is a member's
artefact, and a peer is not yours to task. If a unit needs doing and it falls in your area, spawn a
worker and delegate it yourself; if it falls in the peer's area, say so and let the peer assign it.
member and delegate it yourself; if it falls in the peer's area, say so and let the peer assign it.
The traffic between leads is coordination and nothing else:
1. **Divide the map, not the work.** Agree who owns which area, then each of you assigns inside your
@@ -138,7 +142,7 @@ Being messaged by a peer does not make you its worker: answer with `bridge_reply
the substance if it is wrong. A peer that simply complies has thrown away the reason there are two of
you.
### Worker — the turn contract
### Member (worker or architect) — the turn contract
1. **Load the playbook skill the lead named** before doing anything else.
2. **Do the assigned scope only.** Note anything you spot outside it in one line; don't go hunt it.
@@ -147,6 +151,10 @@ you.
Don't ask what you could decide yourself.
4. **End the turn with exactly one `bridge_reply{content}`**, carrying your complete answer. This is
the whole handoff. No `bridge_reply` ⇒ the sender gets nothing and the exchange stalls.
Do **not** lean on the completion fallback to carry your answer for you: when you end a turn
without replying, the bridge scrapes your pane, and it can return only the last 4000 characters.
A clipped scrape is marked as partial, but the missing text is gone — your report reaches the
lead with its end cut off.
5. **Report honestly.** State only what you actually ran and its real output, including failures.
You mount **only** the bridge MCP — the primary's other servers (IDE, forge, docs) are not yours,
so never claim the result of a check you had no way to run.
@@ -157,9 +165,9 @@ you.
| Layer | Scope | Reaches |
|---|---|---|
| the launcher's reply charter | the one rule that must survive with no repo: *end every turn with `bridge_reply`* | every worker, at launch, every peer kind |
| **this section** | protocol + orchestration policy | primary **and** every Claude worker — tracked in git, so worktrees inherit it |
| role playbook skills | per-job procedure (commit/PR recipe, finding format) | a worker told to load one |
| the launcher's reply charter | the one rule that must survive with no repo: *end every turn with `bridge_reply`* | every spawned member, at launch, every peer kind — never a lead |
| **this section** | protocol + orchestration policy | primary **and** every member that reads the repo — tracked in git, so worktrees inherit it |
| role playbook skills | per-job procedure (commit/PR recipe, finding format) | a member told to load one |
| the bridge's own docs | design detail, flows, error model | on demand |
A rule belongs in **exactly one** layer — the outermost one that must obey it. Peers that don't read
@@ -129,13 +129,13 @@ public final class Bridged {
adapters.add(new ClaudeCodeLauncher(agents, spaces, guard,
claudeProfiles, cfg.effectiveDefaultProfile(), System::getenv,
cfg.spawnReadyTimeoutMs(), cfg.spawnReadyPollMs(),
() -> config.get().fleet().tabLabel()));
() -> config.get().fleet()));
}
if (!opencodeProfiles.isEmpty()) {
adapters.add(new OpenCodeLauncher(agents, spaces,
opencodeProfiles, cfg.effectiveDefaultProfile(), System::getenv,
cfg.spawnReadyTimeoutMs(), cfg.spawnReadyPollMs(),
() -> config.get().fleet().tabLabel()));
() -> config.get().fleet()));
}
AtomicReference<Function<String, Integer>> liveCountRef = new AtomicReference<>(_ -> 0);
PeerLauncher workers = new CompositePeerLauncher(
@@ -562,7 +562,15 @@ public record BridgedConfig(
tabLabel = (tabLabel == null || tabLabel.isBlank()) ? DEFAULT_TAB_LABEL : tabLabel;
}
/** Convenience constructor for code that does not configure launch charters. */
/**
* A fleet with no configured launch charters — the shape every deployment had before
* CB-566, and what most tests want.
*
* <p>Kept deliberately, even though an overload that drops a new field is normally the
* shape to avoid. It is safe here because nothing <em>reads</em> a charter through a
* constructor: the launcher reads {@code fleet.charters()} from the live config. Jackson
* binds the canonical constructor, so this one cannot swallow an operator's YAML.
*/
public Fleet(Map<String, Leader> leaders, Map<String, Slot> architects,
Map<String, Slot> developers, Map<String, Slot> reviewers, String tabLabel) {
this(leaders, architects, developers, reviewers, null, tabLabel);
@@ -1217,7 +1225,7 @@ public record BridgedConfig(
if (fleet == null || fleet.charters().isEmpty()) {
return;
}
List<String> valid = java.util.Arrays.stream(MemberRole.values())
List<String> valid = Arrays.stream(MemberRole.values())
.map(MemberRole::wireName)
.toList();
List<String> bad = new ArrayList<>();
@@ -28,9 +28,9 @@ import java.util.function.Supplier;
* <ul>
* <li><strong>Hot</strong> — re-read per use, so a reload takes effect on the next spawn:
* {@code fleet:} (every role pool, {@code charters}, and {@code tabLabel}),
* {@code placement:}, and an existing profile's {@code weight} / {@code maxLoad}. Those three are read through a supplier on
* {@code CompositePeerLauncher}, which is what makes them hot — not the fact that they are
* config.</li>
* {@code placement:}, and an existing profile's {@code weight} / {@code maxLoad}. Those
* three are read through a supplier on {@code CompositePeerLauncher}, which is what makes
* them hot — not the fact that they are config.</li>
* <li><strong>Deferred</strong> — accepted into the new snapshot, but the wiring built at startup
* keeps the old value until a restart: {@code lifecycle:}, {@code leadHeartbeat:},
* {@code spawnReadyTimeoutMs} / {@code spawnReadyPollMs}, {@code guard:},
@@ -44,23 +44,6 @@ public final class ClaudeCodeLauncher extends HerdrPeerLauncher {
private final SubscriptionGuard guard;
/**
* Standing instruction appended to the worker's system prompt so it returns its result via
* {@code bridge_reply}. Injected as a launch flag, so nothing is written to the worker's
* profile — it is guidance, and a worker that never replies is caught by the send's timeout.
*/
static final String REPLY_CHARTER =
"You are an off-subscription worker in the claude-bridge fleet. Every message you "
+ "receive arrives through the bridge, and the ONLY channel back to the sender is the "
+ "bridge_reply MCP tool. Text you write in your terminal is NOT sent anywhere — the "
+ "sender cannot see your screen, so an in-terminal answer is silently discarded. "
+ "Therefore you MUST end EVERY turn by calling bridge_reply with `content` set to your "
+ "complete response. This holds for every message without exception — tasks, questions, "
+ "clarifications, acknowledgements, and ordinary back-and-forth conversation. Call "
+ "bridge_reply exactly once, as the final action of your turn, with your full answer in "
+ "`content`; never wait for confirmation first. If you end a turn without calling "
+ "bridge_reply, the sender receives nothing and the exchange stalls.";
/**
* Production constructor — disables the spawn-ready gate ({@code spawnReadyTimeoutMs == 0}) so
* existing deployments and tests keep the legacy non-blocking spawn semantics.
@@ -93,11 +76,11 @@ public final class ClaudeCodeLauncher extends HerdrPeerLauncher {
Map<String, BridgedConfig.Profile> profiles, String defaultProfile,
Function<String, String> env,
long spawnReadyTimeoutMs, long spawnReadyPollMs,
Supplier<String> tabLabelTemplate) {
Supplier<BridgedConfig.Fleet> fleet) {
this(agents, spaces, guard, profiles, defaultProfile, env,
spawnReadyTimeoutMs,
System::currentTimeMillis, () -> sleepUninterruptibly(spawnReadyPollMs),
tabLabelTemplate);
fleet);
}
/**
@@ -129,31 +112,19 @@ public final class ClaudeCodeLauncher extends HerdrPeerLauncher {
/**
* Full testability constructor, plus the fleet-wide tab-label template (CB-557).
*
* @param tabLabelTemplate {@code fleet.tabLabel}; {@code null}/blank ⇒
* {@link BridgedConfig.Fleet#DEFAULT_TAB_LABEL}
* @param fleet live fleet config, read once for each spawn
*/
public ClaudeCodeLauncher(AgentControl agents, WorkspaceControl spaces, SubscriptionGuard guard,
Map<String, BridgedConfig.Profile> profiles, String defaultProfile,
Function<String, String> env,
long spawnReadyTimeoutMs,
LongSupplier nowMillis, Runnable sleeper,
Supplier<String> tabLabelTemplate) {
LongSupplier nowMillis, Runnable sleeper,
Supplier<BridgedConfig.Fleet> fleet) {
super(NAME_PREFIX, agents, spaces, profiles, defaultProfile, env,
spawnReadyTimeoutMs, nowMillis, sleeper, tabLabelTemplate);
spawnReadyTimeoutMs, nowMillis, sleeper, fleet);
this.guard = guard;
}
/**
* {@inheritDoc}
*
* <p>A legacy spawn with no session identity is a fresh, launcher-derived session — delegate to
* the session-aware form with no name and no resume id.
*/
@Override
protected Launch buildLaunch(BridgedConfig.Profile cfg) {
return buildLaunch(cfg, null, null);
}
/**
* {@inheritDoc}
*
@@ -164,7 +135,7 @@ public final class ClaudeCodeLauncher extends HerdrPeerLauncher {
* applied here — see {@link #applySessionIdentity}.
*/
@Override
protected Launch buildLaunch(BridgedConfig.Profile cfg, String sessionName, String resumeSessionId) {
protected Launch buildLaunch(BridgedConfig.Profile cfg, LaunchSpec spec) {
// CB-539: a profile may deliberately opt into the subscription (subscription: true) when no
// off-subscription endpoint exists for it — e.g. `sonnet` on `ccs`. That profile gets no
// ANTHROPIC_BASE_URL/AUTH_TOKEN (there is nothing to point them at) and the guard's base_url
@@ -212,7 +183,7 @@ public final class ClaudeCodeLauncher extends HerdrPeerLauncher {
// mutableArgv: argvWithBridge may hand back the profile's own (immutable) List.of when it
// has no MCP — session flags must be added into a list we own.
List<String> argv = mutableArgv(argvWithBridge(cfg));
String agentSessionId = applySessionIdentity(argv, sessionName, resumeSessionId);
String agentSessionId = applySessionIdentity(argv, spec.sessionName(), spec.resumeSessionId());
return new Launch(workerEnv, argvWithModel(argv, cfg), agentSessionId);
}
@@ -44,7 +44,7 @@ import java.util.regex.Pattern;
* <li>{@code namePrefix} (constructor arg) — the label prefix ({@code claude}, {@code opencode})
* that drives both unique naming and the orphan-reap pattern, so each adapter reaps only its
* own kind of pane and never another's.</li>
* <li>{@link #buildLaunch(BridgedConfig.Profile)} — the peer-specific env map + argv, including any
* <li>{@link #buildLaunch(BridgedConfig.Profile, LaunchSpec)} — the peer-specific env map + argv, including any
* subscription/guard check, MCP mount, and instruction injection. The base never sees how the
* peer is configured; it only places and starts the returned {@link Launch}.</li>
* </ul>
@@ -80,15 +80,25 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
private final AtomicLong nameSeq = new AtomicLong(); // per-peer counter (herdr agent names only)
/**
* The {@code fleet.tabLabel} template; a {@code null} supplier or a {@code null}/blank value ⇒
* {@link BridgedConfig.Fleet#DEFAULT_TAB_LABEL}. A profile's own {@code tabLabel} still
* overrides it.
* Live fleet config, read once per spawn. A null supplier or value leaves tab labels at their
* default and supplies no role charter. A profile's own {@code tabLabel} still overrides it.
*
* <p>CB-559: a supplier rather than a String, so a config reload renames the <em>next</em> tab
* without a restart. Existing tabs keep the label they were given — bridged does not rewrite a
* label it already wrote.
* <p>CB-559: a supplier rather than a snapshot, so a config reload affects the next launch
* without a restart. Existing tabs keep the label they were given.
*/
private final Supplier<String> tabLabelTemplate;
private final Supplier<BridgedConfig.Fleet> fleet;
/** The final instruction always requires a bridge reply when the bridge MCP is mounted. */
protected static final String REPLY_CHARTER =
"You are a spawned member in the claude-bridge fleet. Every message you receive arrives "
+ "through the bridge, and the ONLY channel back to the sender is the bridge_reply MCP tool. "
+ "Text you write in your terminal is NOT sent anywhere — the sender cannot see your screen, "
+ "so an in-terminal answer is silently discarded. Therefore you MUST end EVERY turn by calling "
+ "bridge_reply with `content` set to your complete response. This holds for every message without "
+ "exception — tasks, questions, clarifications, acknowledgements, and ordinary back-and-forth "
+ "conversation. Call bridge_reply exactly once, as the final action of your turn, with your full "
+ "answer in `content`; never wait for confirmation first. If you end a turn without calling "
+ "bridge_reply, the sender receives nothing and the exchange stalls.";
/**
* Tab numbers, counted per {@code role/profile} pair (CB-557).
@@ -142,12 +152,11 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
}
/**
* As above, plus the {@code fleet.tabLabel} template (CB-557).
* As above, plus the live {@code fleet} config (CB-557).
*
* @param tabLabelTemplate fleet-wide tab-label template, read per spawn (CB-559); {@code null},
* or a supplier yielding {@code null}/blank ⇒
* {@link BridgedConfig.Fleet#DEFAULT_TAB_LABEL}. A separate constructor
* rather than a new parameter on the one above, so every existing call
* @param fleet live fleet config, read once per spawn; {@code null} ⇒ default tab label and no
* role charter. A separate constructor rather than a new parameter on the one above,
* so every existing call
* site keeps the default without an edit.
*/
protected HerdrPeerLauncher(String namePrefix, AgentControl agents, WorkspaceControl spaces,
@@ -155,8 +164,8 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
Function<String, String> env,
long spawnReadyTimeoutMs,
LongSupplier nowMillis, Runnable sleeper,
Supplier<String> tabLabelTemplate) {
this.tabLabelTemplate = tabLabelTemplate;
Supplier<BridgedConfig.Fleet> fleet) {
this.fleet = fleet;
this.namePrefix = namePrefix;
this.agents = agents;
this.spaces = spaces;
@@ -175,22 +184,7 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
* Any subscription/guard check, MCP mount, and instruction injection happen here. The env map
* and argv are adapter-private; the base only places and starts what is returned.
*/
protected abstract Launch buildLaunch(BridgedConfig.Profile cfg);
/**
* Session-aware variant of {@link #buildLaunch(BridgedConfig.Profile)} (CB-547a). Default
* discards the session identity and delegates to the profile-only form, so an adapter that
* carries no durable peer session (opencode, say) inherits byte-identical behaviour and needs
* no change. An adapter that does (Claude Code) overrides this to mint/resume the id and to
* surface it on the returned {@link Launch#agentSessionId()}.
*
* @param cfg the resolved profile to spawn
* @param sessionName the bridge's logical session name, or null/blank for launcher-derived
* @param resumeSessionId the peer's own prior session id to resume, or null/blank for fresh
*/
protected Launch buildLaunch(BridgedConfig.Profile cfg, String sessionName, String resumeSessionId) {
return buildLaunch(cfg);
}
protected abstract Launch buildLaunch(BridgedConfig.Profile cfg, LaunchSpec spec);
/** Direct transport access for peer-specific, non-turn control operations. */
protected final AgentControl agents() {
@@ -225,6 +219,10 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
}
}
/** All per-spawn values adapters may need, including the base-composed effective charter. */
protected record LaunchSpec(String sessionName, String resumeSessionId, MemberRole role, String charter) {
}
// --- profile surface -----------------------------------------------------------------------
/** The configured peer profile names (what {@code spawn(profile)} accepts). */
@@ -295,18 +293,23 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
}
/**
* Spawn a peer with session identity (CB-547a). {@code sessionName} and {@code resumeSessionId}
* are threaded from the {@link SpawnRequest} into {@link #buildLaunch(BridgedConfig.Profile,
* String, String)}, and the launch's resolved agent-session id is returned alongside the agent
* so the caller can put it on the {@link PeerHandle}.
* Spawn a peer with session identity (CB-547a). The session values, role, and charter are
* threaded from the {@link SpawnRequest} into {@link #buildLaunch(BridgedConfig.Profile,
* LaunchSpec)}, and the launch's resolved agent-session id is returned alongside the agent so
* the caller can put it on the {@link PeerHandle}.
*/
protected Spawned spawnInternal(String profileName, String requestedCwd, String callerCwd,
String sessionName, String resumeSessionId, MemberRole role) {
BridgedConfig.Profile cfg = requireProfile(profileName);
Launch launch = buildLaunch(cfg, sessionName, resumeSessionId);
BridgedConfig.Fleet liveFleet = fleet == null ? null : fleet.get();
String roleCharter = liveFleet == null ? null : liveFleet.charterFor(role);
String replyCharter = cfg.hasMcp() ? REPLY_CHARTER : null;
String charter = roleCharter == null ? replyCharter
: replyCharter == null ? roleCharter : roleCharter + "\n\n" + replyCharter;
Launch launch = buildLaunch(cfg, new LaunchSpec(sessionName, resumeSessionId, role, charter));
String cwd = resolveCwd(requestedCwd, cfg, callerCwd);
Agent agent = cfg.tabPlacement()
? spawnInTab(cfg, launch.env(), launch.argv(), cwd, role)
? spawnInTab(cfg, launch.env(), launch.argv(), cwd, role, liveFleet)
: spawnAsPane(cfg, launch.env(), launch.argv(), cwd);
return new Spawned(agent, launch.agentSessionId());
}
@@ -382,7 +385,7 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
/** Dedicated worker space → own tab (carrying cwd+env) → start the peer into the seed pane. */
private Agent spawnInTab(BridgedConfig.Profile cfg, Map<String, String> workerEnv,
List<String> argv, String cwd, MemberRole role) {
List<String> argv, String cwd, MemberRole role, BridgedConfig.Fleet liveFleet) {
Workspace space = spaces.ensureWorkspace(cfg.workspace());
Tab.Created tab = spaces.createTab(space.workspaceId(), cwd, workerEnv);
log.info("spawning {} profile={} space={} tab={} cwd={}",
@@ -415,7 +418,7 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
tidy("label tab " + tab.tab().tabId(),
() -> spaces.renameTab(tab.tab().tabId(),
cfg.renderTabLabel(
tabLabelTemplate == null ? null : tabLabelTemplate.get(),
liveFleet == null ? null : liveFleet.tabLabel(),
role, nextLabelSeq(role, cfg.profile()))));
log.info("{} started pane={} tab={} terminal={}",
namePrefix, started.agent().paneId(), started.agent().tabId(), started.agent().terminalId());
@@ -54,38 +54,9 @@ public final class OpenCodeLauncher extends HerdrPeerLauncher {
/** Writer for the generated {@code opencode.json}. */
private static final ObjectMapper JSON = new ObjectMapper();
/**
* Standing instruction written to the charter file and mounted via the config's
* {@code instructions} so the worker returns its result through {@code bridge_reply}. Kept on
* disk (not a launch flag) because opencode's {@code instructions} takes file paths, not inline
* text — the file is regenerated per spawn and never touches the worker's own profile.
*/
static final String REPLY_CHARTER =
"You are an off-subscription worker in the claude-bridge fleet, running under opencode. "
+ "Every message you receive arrives through the bridge, and the ONLY channel back to the "
+ "sender is the bridge_reply MCP tool. Text you write in your terminal is NOT sent "
+ "anywhere — the sender cannot see your screen, so an in-terminal answer is silently "
+ "discarded. Therefore you MUST end EVERY turn by calling bridge_reply with `content` set "
+ "to your complete response. This holds for every message without exception — tasks, "
+ "questions, clarifications, acknowledgements, and ordinary back-and-forth conversation. "
+ "Call bridge_reply exactly once, as the final action of your turn, with your full answer "
+ "in `content`; never wait for confirmation first. If you end a turn without calling "
+ "bridge_reply, the sender receives nothing and the exchange stalls.";
/** Root under which per-spawn opencode config dirs are created (injectable for tests). */
private final Path configRoot;
/**
* The current spawn's resume-target session id, threaded from {@link #spawn(SpawnRequest)} to
* {@link #buildLaunch} across the base's {@code spawn -> spawnInternal -> buildLaunch} chain,
* which carries no request. A plain field would race under concurrent spawns (the base supports
* them), so it is thread-local: each spawn captures its own request's id on its own thread, and
* {@code buildLaunch}, synchronous and same-thread, reads exactly that one. Set only around the
* {@code super.spawn} call and cleared in {@code finally}, so a paused/leftover value can never
* bleed into the next spawn.
*/
private final ThreadLocal<String> resumeSessionId = new ThreadLocal<>();
/**
* Session discovery against opencode's on-disk storage ({@link OpenCodeSessionDiscovery}) —
* the one seam that knows opencode's private session-file layout. Its root is injectable for
@@ -126,10 +97,10 @@ public final class OpenCodeLauncher extends HerdrPeerLauncher {
Map<String, BridgedConfig.Profile> profiles, String defaultProfile,
Function<String, String> env,
long spawnReadyTimeoutMs, long spawnReadyPollMs,
Supplier<String> tabLabelTemplate) {
Supplier<BridgedConfig.Fleet> fleet) {
this(agents, spaces, profiles, defaultProfile, env, spawnReadyTimeoutMs,
System::currentTimeMillis, () -> sleepUninterruptibly(spawnReadyPollMs),
defaultConfigRoot(), defaultDiscoveryRoot(), tabLabelTemplate);
defaultConfigRoot(), defaultDiscoveryRoot(), fleet);
}
/**
@@ -164,8 +135,7 @@ public final class OpenCodeLauncher extends HerdrPeerLauncher {
/**
* Full testability constructor, plus the fleet-wide tab-label template (CB-557).
*
* @param tabLabelTemplate {@code fleet.tabLabel}; {@code null}/blank ⇒
* {@link BridgedConfig.Fleet#DEFAULT_TAB_LABEL}
* @param fleet live fleet config, read once for each spawn
*/
public OpenCodeLauncher(AgentControl agents, WorkspaceControl spaces,
Map<String, BridgedConfig.Profile> profiles, String defaultProfile,
@@ -173,9 +143,9 @@ public final class OpenCodeLauncher extends HerdrPeerLauncher {
long spawnReadyTimeoutMs,
LongSupplier nowMillis, Runnable sleeper,
Path configRoot, Path discoveryRoot,
Supplier<String> tabLabelTemplate) {
Supplier<BridgedConfig.Fleet> fleet) {
super(NAME_PREFIX, agents, spaces, profiles, defaultProfile, env,
spawnReadyTimeoutMs, nowMillis, sleeper, tabLabelTemplate);
spawnReadyTimeoutMs, nowMillis, sleeper, fleet);
this.configRoot = configRoot;
this.discovery = new OpenCodeSessionDiscovery(discoveryRoot);
}
@@ -199,14 +169,15 @@ public final class OpenCodeLauncher extends HerdrPeerLauncher {
* with {@code -m}.
*/
@Override
protected Launch buildLaunch(BridgedConfig.Profile cfg) {
protected Launch buildLaunch(BridgedConfig.Profile cfg, LaunchSpec spec) {
Map<String, String> workerEnv = baseEnv(cfg);
// A config file is needed for the bridge MCP mount, for a pinned endpoint (CB-508), or both.
if (cfg.hasMcp() || hasCustomProvider(cfg)) {
workerEnv.put("OPENCODE_CONFIG", writeConfig(cfg).toString());
}
applyGitToken(workerEnv, cfg);
return new Launch(workerEnv, argvWithResume(argvWithModel(argvWithAuto(cfg), cfg)));
return new Launch(workerEnv,
argvWithResume(argvWithModel(argvWithAuto(cfg), cfg), spec.resumeSessionId()));
}
/**
@@ -243,11 +214,9 @@ public final class OpenCodeLauncher extends HerdrPeerLauncher {
* The launch argv plus, on a resumed spawn, opencode's {@code -s <id>} flag to continue a prior
* conversation by its session id. {@code -s, --session <id>} resumes an existing session; on a
* fresh spawn (no resume target) no flag is added, letting opencode start a brand-new session.
* The id comes from the current spawn request's {@code resumeSessionId}, threaded per-thread by
* {@link #spawn(SpawnRequest)}.
* The id comes from the base launch spec.
*/
private List<String> argvWithResume(List<String> argv) {
String id = resumeSessionId.get();
private List<String> argvWithResume(List<String> argv, String id) {
if (id == null || id.isBlank()) {
return argv;
}
@@ -380,31 +349,11 @@ public final class OpenCodeLauncher extends HerdrPeerLauncher {
return afterScheme.contains("/") ? trimmed : trimmed + "/v1";
}
/**
* {@inheritDoc}
*
* <p>adds this adapter's session-identity work around the base's spawn — as opencode cannot be
* told its session id at spawn (see {@link Capability#SESSION_RESUME} vs
* {@link Capability#SESSION_NAME}), identity is only ever adopted after the fact:
* <ul>
* <li>the request's {@code resumeSessionId} is remembered for {@link #buildLaunch} to turn
* into {@code -s <id>}; and</li>
* <li>the returned handle is wrapped so its
* {@link dev.ltms.bridged.peer.PeerHandle#agentSessionId()} performs lazy session
* discovery against opencode's storage (see {@link OpenCodeSessionDiscovery}) — always
* non-blocking, {@code null} until opencode has persisted the session record.</li>
* </ul>
*/
/** Add lazy on-disk session discovery to the base handle. */
@Override
public PeerHandle spawn(SpawnRequest req) {
resumeSessionId.set(req.resumeSessionId());
try {
PeerHandle inner = super.spawn(req);
return new SessionAwareHandle(inner, discovery, effectiveCwd(req));
} finally {
// Never let a paused/leftover resume id bleed into the next spawn on this thread.
resumeSessionId.remove();
}
PeerHandle inner = super.spawn(req);
return new SessionAwareHandle(inner, discovery, effectiveCwd(req));
}
/**
@@ -796,14 +796,18 @@ class ClaudeCodeLauncherTest {
.toList();
}
private static BridgedConfig.Fleet fleet(Map<String, String> charters, String tabLabel) {
return new BridgedConfig.Fleet(Map.of(), Map.of(), Map.of(), Map.of(), charters, tabLabel);
}
/** A profile with no {@code tabLabel:} of its own — the fleet template decides. */
private ClaudeCodeLauncher labelService(FakeHerdr herdr, Supplier<String> fleetTemplate) {
private ClaudeCodeLauncher labelService(FakeHerdr herdr, Supplier<BridgedConfig.Fleet> fleet) {
BridgedConfig.Profile cfg = new BridgedConfig.Profile(
"sonnet", "http://gx00.gw:8000", "sonnet", null, "BRIDGED_WORKER_TOKEN",
List.of("claude"), "tab", "bridged-workers", null, null, null, null);
return new ClaudeCodeLauncher(new AgentControl(herdr), new WorkspaceControl(herdr),
new SubscriptionGuard(Set.of("gx00.gw")), Map.of(cfg.profile(), cfg), cfg.profile(),
_ -> null, 0, 0L, fleetTemplate);
_ -> null, 0, 0L, fleet);
}
/**
@@ -814,7 +818,7 @@ class ClaudeCodeLauncherTest {
@Test
void theFleetTemplateNamesTheRoleTheMemberWasSpawnedFor() {
FakeHerdr herdr = new FakeHerdr();
ClaudeCodeLauncher svc = labelService(herdr, () -> "{role}: {profile} #{n}");
ClaudeCodeLauncher svc = labelService(herdr, () -> fleet(null, "{role}: {profile} #{n}"));
svc.spawn(new SpawnRequest("sonnet", null, null, null, null, MemberRole.REVIEWER));
@@ -825,7 +829,7 @@ class ClaudeCodeLauncherTest {
@Test
void theCounterRunsPerRoleAndProfileNotPerFleet() {
FakeHerdr herdr = new FakeHerdr();
ClaudeCodeLauncher svc = labelService(herdr, () -> "{role}: {profile} #{n}");
ClaudeCodeLauncher svc = labelService(herdr, () -> fleet(null, "{role}: {profile} #{n}"));
svc.spawn(new SpawnRequest("sonnet", null, null, null, null, MemberRole.DEV));
svc.spawn(new SpawnRequest("sonnet", null, null, null, null, MemberRole.REVIEWER));
@@ -839,7 +843,7 @@ class ClaudeCodeLauncherTest {
@Test
void aBlankFleetTemplateFallsBackToTheRoleFirstDefault() {
FakeHerdr herdr = new FakeHerdr();
labelService(herdr, () -> null).spawn(
labelService(herdr, () -> fleet(null, null)).spawn(
new SpawnRequest("sonnet", null, null, null, null, MemberRole.ARCHITECT));
assertEquals(List.of("architect: sonnet #1"), tabLabels(herdr));
@@ -855,7 +859,7 @@ class ClaudeCodeLauncherTest {
List.of("claude"), "tab", "bridged-workers", "pinned {profile}", null, null, null);
new ClaudeCodeLauncher(new AgentControl(herdr), new WorkspaceControl(herdr),
new SubscriptionGuard(Set.of("gx00.gw")), Map.of(cfg.profile(), cfg), cfg.profile(),
_ -> null, 0, 0L, () -> "{role}: {profile} #{n}")
_ -> null, 0, 0L, () -> fleet(null, "{role}: {profile} #{n}"))
.spawn(new SpawnRequest("sonnet", null, null, null, null, MemberRole.REVIEWER));
assertEquals(List.of("pinned sonnet"), tabLabels(herdr));
@@ -870,7 +874,7 @@ class ClaudeCodeLauncherTest {
void theTemplateIsReadOnEverySpawnSoAnEditTakesEffect() {
FakeHerdr herdr = new FakeHerdr();
AtomicReference<String> template = new AtomicReference<>("{role}: {profile} #{n}");
ClaudeCodeLauncher svc = labelService(herdr, template::get);
ClaudeCodeLauncher svc = labelService(herdr, () -> fleet(null, template.get()));
svc.spawn(new SpawnRequest("sonnet", null, null, null, null, MemberRole.DEV));
template.set("[{profile}] {role} {n}");
@@ -77,7 +77,7 @@ class CompositePeerLauncherTest {
}
@Override
protected Launch buildLaunch(BridgedConfig.Profile cfg) {
protected Launch buildLaunch(BridgedConfig.Profile cfg, LaunchSpec spec) {
return new Launch(Map.of(), List.of());
}
@@ -0,0 +1,71 @@
package dev.ltms.bridged.member;
import dev.ltms.bridged.config.BridgedConfig;
import dev.ltms.bridged.herdr.AgentControl;
import dev.ltms.bridged.herdr.FakeHerdr;
import dev.ltms.bridged.herdr.WorkspaceControl;
import dev.ltms.bridged.peer.Capability;
import dev.ltms.bridged.peer.MemberRole;
import dev.ltms.bridged.peer.SpawnRequest;
import org.junit.jupiter.api.Test;
import java.util.ArrayList;
import java.util.List;
import java.util.Map;
import java.util.Set;
import java.util.concurrent.atomic.AtomicReference;
import java.util.function.Supplier;
import static org.junit.jupiter.api.Assertions.assertEquals;
class HerdrPeerLauncherCharterTest {
@Test
void readsAndComposesTheFleetCharterForEachSpawn() {
AtomicReference<BridgedConfig.Fleet> fleet = new AtomicReference<>(fleet(Map.of()));
CapturingLauncher launcher = new CapturingLauncher(fleet::get);
launcher.spawn(new SpawnRequest("mcp", null, null, null, null, MemberRole.DEV));
fleet.set(fleet(Map.of("dev", "role charter")));
launcher.spawn(new SpawnRequest("mcp", null, null, null, null, MemberRole.DEV));
launcher.spawn(new SpawnRequest("no-mcp", null, null, null, null, MemberRole.DEV));
assertEquals(HerdrPeerLauncher.REPLY_CHARTER, launcher.specs.get(0).charter(),
"without a role charter, MCP profiles receive only the reply charter");
assertEquals("role charter\n\n" + HerdrPeerLauncher.REPLY_CHARTER, launcher.specs.get(1).charter(),
"the changed supplier value is read for the next spawn and the reply rule is last");
assertEquals("role charter", launcher.specs.get(2).charter(),
"a role charter does not depend on an MCP mount");
}
private static BridgedConfig.Fleet fleet(Map<String, String> charters) {
return new BridgedConfig.Fleet(Map.of(), Map.of(), Map.of(), Map.of(), charters, null);
}
private static final class CapturingLauncher extends HerdrPeerLauncher {
private final List<LaunchSpec> specs = new ArrayList<>();
CapturingLauncher(Supplier<BridgedConfig.Fleet> fleet) {
super("test", new AgentControl(new FakeHerdr()), new WorkspaceControl(new FakeHerdr()),
Map.of("mcp", profile("mcp", "http://bridge"),
"no-mcp", profile("no-mcp", null)),
"mcp", _ -> null, 0, () -> 0L, () -> { }, fleet);
}
@Override
protected Launch buildLaunch(BridgedConfig.Profile cfg, LaunchSpec spec) {
specs.add(spec);
return new Launch(Map.of(), List.of("test"));
}
@Override
public Set<Capability> capabilities() {
return Set.of();
}
private static BridgedConfig.Profile profile(String name, String mcpUrl) {
return new BridgedConfig.Profile(name, "http://gx00.gw:8000", null, null,
"BRIDGED_WORKER_TOKEN", List.of("test"), "pane", null, null, mcpUrl, null, null);
}
}
}