The hive reads the agent-archetype key when it provisions and rebuilds an
agent and keeps it in swarm-agent-archetypes.json. For an archetype agent
the meta flake has no agent-<name> input and mkAgent imports
hyperhive.hyperhiveArchetypes.<name>.module; provisioning, rebuilds, the
meta-input cascade, a deploy's rev and the startup migration skip the
config repos. The harness gets HIVE_ARCHETYPE and the system prompt says
the agent has no config repo. The create form sends the picked archetype.
Adds the android-dev built-in.
Refs #5386
POST /api/agents takes an archetype the stored list holds, records it in
the new agent-archetype bucket, and queues no CreateRepo, AddRepoMember or
InitAgentConfigRepo for it. A destroy clears the key. Hives may read the
bucket.
Refs #5386
The dashboard vhost, the hive domain vhost and the dashboard's local
name move to hive-c0re/vhosts.nix; the matrix .well-known discovery and
the legacy /matrix/ redirect move to glue-matrix-hive-discovery.nix,
which adds them to the hive domain vhost. Both declare through
gateway.vhosts. hive-gateway/vhosts.nix keeps the catch-all only, and
the gateway no longer reads deploy.matrix or c0re.servedFrontend, nor
deploy.hive-controller.enable for its vhosts.
Part of #3742.
The agent read the operator-profiles bucket with a keys() listing, an
ordered consumer whose grant (STREAM.INFO, CONSUMER.CREATE) and flow-control
reply had never run against a permissioned queue. The controller now writes
the whole list as one JSON array under one key, only when it changed, and
the agent reads it with one DIRECT.GET the way agent_inference::read_own
reads its key. The agent grant shrinks to that one subject.
swarm-controller reads each member of authelia's operator group's forge
profile every five minutes and keeps one key per operator in the new
operator-profiles bucket. Every agent may list and read that bucket, and
hive-agent renders it into the system prompt at harness start, falling
back to today's prompt when the bucket is empty or doesn't answer within
five seconds.
Fixes#5383
A flake declares agent archetypes in its `hyperhiveArchetypes` output,
`<name> = { module; meta; }`. The hyperhive flake ships `reviewer`
(default permissions without `execution`).
swarm-builder answers `$SWARM.builder.archetypes` with the `meta` of
every archetype its own hyperhive source declares. swarm-controller asks
once the queue connects and every 10 minutes, checks each archetype as
`POST /api/agents` checks its settings, and stores the result in the
`agent-archetypes` KV bucket when it changed. `GET /api/archetypes`
serves it, and the create form's archetype select prefills permissions,
resource caps and inference from it.
The agent created is an ordinary agent with its own config repo; the
archetype's module is not applied yet.
Refs #5386
A review request for a head whose newest recorded review approved or
requested changes now runs nothing: the classifier account posts that
verdict and table again, opening "Already reviewed <sha>, verdict
unchanged". A head whose newest review had no clean verdict, a failed
post, or no record gets a fresh run, as before.
An accepted request now keeps its key until its record is written, so a
request right after a run finishes reads that run's verdict instead of
starting a second run.
Part of #5387; the restart-loss half waits on #5219.
The collector's oidc/swarm authenticator verifies the bearer offline as a
JWT. The controller's authelia client was registered without
accessTokenSignedResponseAlg, so authelia mints it an opaque
authelia_at_ handle, which that authenticator cannot verify: by config,
no push from the vcs or jobq exporter can be accepted. Register it with
RS256 like every other client whose token lands on an oidc/* receiver.
The SDK logs a failed export at debug only, so a refused push never
reached the journal. AuthenticatedHttpClient now warns on a failed token
mint, a transport error, or a non-2xx answer (status plus a body
snippet), once per kind of failure and again every 10 minutes with a
count, and logs once when a push succeeds after failing.
Refs #5393
Permission::default_for and hive-sh4re's MANAGER_AGENT are gone: every
agent with no permissions key holds Permission::DEFAULT, ruth included.
The agent-name parameters that only fed default_for go with it
(permissions::effective, HiveCopy::before_read/after_read and their
callers).
swarm-controller seeds the grant ruth held by name: at start, a declared
ruth whose permissions are unset gets DEFAULT plus manage_root_agent
written into her key, once, logged. A set key is never touched. It is a
one-shot to delete once every swarm has passed it.
The swarm UI setup assistant opens the create form with
manage_root_agent checked, so a new ruth is created with the grant.
permissions_import no longer adds ruth to the import or splits her
default. Docs and help text no longer say ruth holds the grant by
default; swarmctl-cli.md regenerated.
hh-swarm-queue-connect-failed matched "swarm queue connect failed; this
agent publishes nothing upward", which nothing has logged since the
harness started retrying its first connect and exiting (235bc481b). The
rule now matches the error the harness logs when every attempt has
failed, CONNECT_EXHAUSTED in swarm_queue.rs, and its comment and summary
describe the retry-then-exit behaviour.
A failed MintAgentIdentity node logged nothing: the error reached only
the job graph. mint_identity now warns with the agent and the error, and
hh-agent-identity-mint-failed fires on that line.
Refs #4651
Nothing sets the variable once otel.metricIntervalMs is gone, so the
turn-metrics, jobq-metrics and vcs-metrics exporters export at a fixed
60s, as hive-c0re's container-resource exporter already does. The
observability doc and the removal warning now say so.
swarm-secret-publish and its path unit now run inside authelia's container,
beside the mint, and read each client secret at the path authelia writes it.
The host unit and host path unit, which read the secrets through the
container's root on the host, are gone, and so is
deploy.authelia.hostClientSecretDir, which only they read. The publisher keeps
its store identity, the secret-publisher leaf. The host hands the leaf and the
store's CA into the container as nspawn system credentials, and the unit loads
them; where this host mints the leaf, the container waits for swarm-bao-pki.
Authelia does not wait for the publisher, which retries every 30 seconds while
the store is down.
The container also runs the journal forwarder the other service containers
run, pushing to a receiver of its own on the swarm collector (serviceProducers
swarm-authelia) as a new machine OIDC client (swarm.authelia.otel.clientId)
that authelia registers for itself and the publisher carries to the store. The
fetch unit reads it back under the publisher's leaf, also handed to it as a
system credential, so the swarm-secret-publisher policy gains read on that one
leaf; its stanza keeps create and update, which the exact path would otherwise
take from the services/* prefix. Without the leaf the forwarder is off.
The container drops --link-journal=host, and the host collector stops reading
it. The services' "nothing at <path>" errors name the publisher in authelia's
container.
A swarm-ci-metrics host unit on a 5-minute timer lists every forge
repo's Actions runs as the new read-only ci-metrics account and pushes
each finished run's duration and queue wait over OTLP, stamped at the
run's stop time. The first pass backfills from run 1. swarm-controller
mints the account's token into bao and keeps a ci-metrics team with
read on the Actions unit in every organisation.
Closes#5376
`forgejo-swarm-controller-account`, in the forge's container, now puts
the swarm-controller's forge token into the swarm secret store itself,
under the forge publisher's leaf. The container already receives that
leaf as nspawn system credentials for its journal forwarder; the mint
loads the same three. The token reaches bao on stdin (`value=-`) and
the forge's API through a header file, never as an argument, and never
touches disk.
The mint's guard is a marker holding the published token's name, at
/var/lib/forgejo/swarm-controller-token-name, written last: after the
put, and after every other `swarm-controller-boot-*` token of the
account is revoked through forgejo's admin API, authenticated as the
new token. So a put that failed, or a marker lost while the store kept
an older token, leaves no live token behind once a mint completes. The
mint logs in to the store and checks that the account is a site admin
before it mints, so a store that is down or an account the admin API
would refuse mints nothing on each retry. Failures retry under
store-retry; forgejo does not wait for the mint. No bao read and no
grant change: the forge publisher keeps create/update on the token.
Removed: the host's `swarm-bao-forge-publisher` service and path unit,
which read the token file through the container's root, and their
entries in the store and policy ordering glue. The forge publisher's
policy unit stays; `swarm-bao-forge-oidc` still uses the leaf.
On deploy, an existing forge host has no marker, so the mint runs once:
it mints a new token, publishes it, revokes the old one, and deletes
the old /var/lib/forgejo/swarm-controller-token. The controller's next
forge call gets a 401, re-reads the store and carries on.
The dashboard, its API, the swagger UI and the agent routes move to their
own vhost, `swarm.dashboardDomain` = `<hive>.hives.<swarm.domain>`, under
the swarm domain so authelia's session cookie reaches it. `/`, `/api/` and
`/api/docs/` ask authelia through `auth_request`, and authelia's
`operatorsOnly` rule for each hive's dashboard name lets only the admins
group in. The hive domain keeps matrix discovery and redirects every other
path to the new name.
The name rides as a SAN on the hive's own gateway leaf: the per-hive bao
role allows it, and hive-tls asks for it and checks coverage.
HTTP Basic auth is gone: the htpasswd writer and its test, the
`hivectl gateway` verbs and host-socket requests, the `gateway.auth`
options (now a removed-option error), the unauthorized page and the bcrypt
dependency. tmpfiles deletes a leftover gateway.htpasswd, which held
bcrypt hashes at 0644.
c0re hands out dashboard and agent URLs on the new name
(HYPERHIVE_DASHBOARD_DOMAIN), and the swarm controller's hive roster
carries each hive's dashboard name for the swarm UI's hive list.
Closes#3469
Declaring an agent Destroyed now deletes its `agent-permissions` key,
next to the queue-credential revoke and with the same failure handling:
logged at error, the destroy goes ahead, and declaring it destroyed
again re-runs it. Agent names are reusable after a destroy, and a create
left on the default stores no key, so the old key handed a new agent of
the same name the destroyed agent's grant (argus on #5293).
Also: migrate.rs's create-time-setup note named ruth's tool-group seed
in `ensure_root_agent`, which this PR deletes; it now says she holds the
code-constant default when the swarm sets nothing.
Part of #5170.
mara on #5170: "dont split caps and perms. its all perms. / dont include
the now empty tool groups / nothing set -> default / also add perms to
agent creation screen".
- hive-sh4re: one `Permission` enum (messaging, meta, inbox, execution,
web_tools, manage_root_agent) replaces `ToolGroup` and `Capability`;
`lifecycle`/`forge` are gone. `Permission::default_for` is the default
for an agent the swarm sets nothing for.
- The record is `{ "permissions": [...] }`. Missing key, unset field or
no successful read yet = default; `[]` = none; a failed read after a
good one keeps the last value.
- hive-agent reads its own key every 10s and builds each turn's --tools,
--allowedTools and MCP config (rewritten before every turn) from it;
the subagent daemon reads it on each start/continue. Agents get a
read-only grant on their own key.
- hive-c0re restarts an agent only when its manage_root_agent changes.
The import now merges meta/capabilities.json and meta/tool-groups.json
into one list per agent and deletes both; the "import pending, don't
read" decision is `ImportOutcome::read_allowed`, table-tested.
- swarm-controller: `GET /api/permissions` (names + default) and an
optional `permissions` on `POST /api/agents`, stored before the job is
queued. swarm-ui's create-agent form gets a permissions field.
- Deleted: HIVE_TOOL_GROUPS and its forward to the subagent daemon,
tool_groups.rs, ruth's tool-group seed, the permissions rebuild job
(PermPayload/WritePermFile/perm_change), the dashboard permissions tab
and its stale-entries panel.
Part of #5170.
An agent's capabilities move from the hive's meta repo
(`meta/capabilities.json`, injected as `HIVE_CAPABILITIES`) into the
`capabilities` field of its key in a new `agent-permissions` KV bucket,
which swarm-controller alone writes.
- swarm-queue-client: `agent_permissions` (bucket, key, `AgentPermissions`,
direct-get read, the import subject and its wire types).
- swarm-controller: creates the bucket at start; `GET|PUT|DELETE
/api/agents/{name}/permissions` (deny_unknown_fields and unknown
capability ⇒ 422, duplicate ⇒ 400, agent must be declared by a hive);
answers each hive's one-shot import, writing a key only where there is
none and only for agents that hive's wanted state declares.
- swarmctl: `agent permissions get|set|clear`.
- swarm-nats-auth: a hive reads every key of the bucket and sends its
import on its own subject; the controller subscribes to the imports and
may answer them (`responses`).
- hive-c0re: reads each of its agents' keys every 30s into
/var/lib/hyperhive/swarm-agent-permissions.json and restarts an agent,
gracefully, when its capabilities change. Unset is the compiled-in
default (ruth: manage_root_agent; everyone else: none). A failed read
keeps the stored value. On first boot it seeds that file from
meta/capabilities.json, sends the grants to the controller, and deletes
the file once the controller answers.
Deleted: the capabilities.json reader/writer, `HIVE_CAPABILITIES` in the
rendered meta flake, ruth's capability seed, the capabilities half of the
permissions rebuild job (`PermPayload::Capabilities`/`Combined`), the
dashboard's capabilities table and `capabilities_changed` event, and the
now-unused `Capability::description`.
Part of #5170.
swarm-bao-grafana-controller-token, on the host that runs Grafana, logs in
to Grafana's API over its socket as the built-in admin (the password
swarm-bao-grafana-admin minted, #5138) and keeps a `swarm-controller`
service account that is an enabled Viewer. It mints a token for that
account and writes it to bao at
swarm/controller/swarm-controller/grafana/token under Grafana's leaf. A
ten-minute timer runs it again. The unit keeps a stored token Grafana
accepts. If Grafana refuses the stored token with 401, or bao holds
none, the unit deletes every token of the account and mints a new one.
Grafana's grant gains create/read/update on that leaf, and the
controller's gains read.
swarm-controller serves GET /api/alerts from Grafana's
/api/prometheus/grafana/api/v1/alerts. It reads the token from bao,
keeps it in memory, and re-reads bao once on a 401. The HTTP client has
5s connect and 15s request timeouts. Any failure is a 503 that names its
cause. swarm-ui gets an /alerts page listing them, built the same way as
the hives page.
This is option (c) on #5129, the operator's choice.
A store answers the revision it wrote and what it replaced, and restoring
compares against that revision, so a write that landed after the creation's
own is kept instead of overwritten. A write that changed nothing rolls back
nothing, and a key that held no value is removed rather than set to an empty
one. A test fails the second store after the first and checks the earlier
caps come back.
create_agent puts back the caps and inference settings it replaced if a
later store or the job insert fails, so a refused creation leaves nothing
behind for the name. The shared field components take an id prefix so the
create dialog and an open editor never share a DOM id. The queue check now
follows the hive roster check, the 422 is documented, and the copy says
empty rather than closed and that destroy keeps stored settings.
POST /api/agents takes optional limits and inference objects, checked by
the same rules as the two PUT routes before anything is queued and stored
ahead of the job graph. The create form gets collapsed sections for both,
reusing the fields of the detail-panel editors.
The three issue-report handlers awaited the forge bare, so a forge that
accepts a connection and never answers hung the request. Wrap each in
within_forge_deadline, as pulls.rs does, and answer 503 on overrun.
A new swarm service, `swarm-claude-login`, in its own container on the host
that sets `deploy.claude-login.enable` (defaulted on by `allSwarmServices`
where the host has the minter's store leaf, like the review classifier).
- swarm-ui: an agent card showing "login needed" gets a "log in" button.
Its dialog starts the login, shows the URL claude printed, and takes the
pasted code.
- swarm-controller: `POST/DELETE /api/agents/{name}/claude-login` and
`POST /api/agents/{name}/claude-login/code`, behind the swarm UI's
authelia-gated `/api/`, relay the steps to the minter as requests on
`$SWARM.claude-login.request`.
- swarm-claude-login: runs `claude auth login` over piped stdio with HOME
and CLAUDE_CONFIG_DIR in a per-login scratch dir, scrapes the first
https URL, writes the code to stdin, and stores the `.credentials.json`
at `swarm/agents/<agent>/claude-credentials` with a KV v2 write, a new
version over any earlier one. The scratch dir goes on every exit path; a
login waits 15 minutes for its code.
- hive-agent: while parked in `wait_for_login`, reads that leaf every 30 s
under the agent's own cert and installs it only when its KV version is
newer than the one recorded in `<harness>/claude-credentials-version`.
An equal or older version is never put back, because claude rotates the
refresh token and the stored copy is stale after the first refresh.
- swarm-secret-client: `claude_login` paths and `SecretStore::read_versioned`.
- Credentials: a `claude-login` store leaf minted by glue-bao-tls, a
`swarm-claude-login` policy (create/update on
`swarm/agents/+/claude-credentials`, read on its own queue client
secret), an authelia machine client, and a `--claude-login-client`
grant in the queue's auth-callout responder. The unit runs as a static
system user.
The hive's own web-UI login is unchanged.
Refs #5268
The wide diff's context drops from 200 to 25 lines, so a block's
enclosing gate stays in view when it opens within 25 lines above the
change (the forge's own diff carries 3). The fallback note, the doc and
the gate fixture (now 20 lines out, not 47) follow.
A failure reading the changed files (a forge error on a tree or blob, a
truncated tree, a non-base64 blob) or joining the diff task made the PR
read fail, so the review posted no verdict. It now falls back to the
forge's 3-line .diff with "reading the files failed: …" as the note.
render wrote each file's whole diff before wide compared it with
MAX_BYTES. It now writes through a writer that refuses once past the cap,
which stops similar's formatter mid-file, so one huge file costs about
MAX_BYTES of output, not its full diff.
The classifier read the forge's PR `.diff`, whose context is fixed at 3
lines. A block moved into a glue module lost its enclosing gate (a
`config = lib.mkIf …` 13 to 191 lines up in the misfired moves #5199,
#5200, #5201), so the glue restating that gate read as a new condition.
The controller now builds the diff itself: it lists the PR's changed files,
reads each at the merge base and at the head through the git tree and blob
API with the token it already holds, and diffs them with 200 lines of
context. It sends the forge's `.diff` instead when the PR changes more than
100 files, the forge names no merge base, or the wide diff is over 256 KiB,
and the posted review then says so with the reason.
Fixes#5227
The retries covered about a minute, and the swarm-bao restart that lost
#5286's review lasted 3m19s. Each try now gets its own forge deadline,
a try with no answer by then is retried like a refused connection, and
the waits are 5s, 15s, 45s, 90s and 150s: six tries over about five
minutes.
The docs also said to request the review again after a post-failed run.
That does nothing while the account's earlier request is still pending,
so they now say to remove the request and then request it again.
A review request's verdict was lost when the post failed: on 2026-10-14
the classifier approved #5286, but its post hit swarm-bao mid-restart
(cert login: connection refused), the run was recorded post-failed, and
nothing on the PR showed it.
post() now tries the post up to four times, waiting 5s, 15s and 45s,
when it fails because the connection to the secret store or the forge
failed or timed out. Any other error fails at once, as before. The
verdict is computed once and reposted as is. All tries share the one
forge deadline the single post had, and the warn log carries the number
of tries.
A separate $SWARM.control.<agent> subject carries commands, apart from
the text subject. swarm-controller exposes POST /api/agents/{name}/cancel
as a NATS request; the harness answers whether a turn was actually
stopped, through the same cancel path as the dashboard button. The agent
credential gets subscribe on its control subject and allow_responses to
answer; only the controller may publish there.
Refs #3803
The review body's table is now: result | check | certainty / threshold |
reason. The result cell is a symbol for what the row did to the review,
after polarity (✅ counted and passed, ❌ counted and failed, ❔ not
counted or no answer), then the model's verdict. The overall line carries
✅ / ❌. Polarity, threshold and counted are folded into those cells.
The prompt asks for a reason under 350 characters; the enforced cap
stays 500, so an overshoot is truncated less often.
The agent's turn-state header ($SWARM.agent-state.<agent>) gains
`needs_login`, set by hive-agent from its own LoginState, so the flag
reaches the swarm without hive-c0re (mara on #5268: "id prefer the agent
container sending the info it needs login without hive").
- swarm-queue-client: `agent_login::Login`, the shared wire type, beside
`agent_inference::Effective`. `needs_login: Option<bool>` with
`#[serde(default)]`: a header from an older agent decodes as `None`
(unknown), never as "needs login" or "logged in".
- hive-agent: the header publisher reads the shared LoginState cell.
Every writer of that cell emits a status on the bus afterwards, so a
login change goes out on the next rebuild, not the 60 s refresh.
- swarm-controller: `agent_login::Reported` follows every agent's header
and keeps the latest flag; `GET /api/agents/status` merges it into each
row as `needs_login` (null when unknown).
- swarm-ui: AgentCard shows a warning-tone "login needed" badge when the
row's `needs_login` is true.
Refs #5268
The commit author's email was the forge account's own address, read with
the site-admin token, which put an email the account may keep private
into the target repo's history. It is now the address Forgejo itself
uses for such an account's web commits, `<lowercased login>@<NO_REPLY_ADDRESS>`
(models/user/user.go GetPlaceholderEmail). NO_REPLY_ADDRESS defaults to
`noreply.<DOMAIN>` (modules/setting/service.go) and the forge module
leaves it unset, with DOMAIN set from services.hyperhive.swarm.forge.domain,
the same option SWARM_CONTROLLER_FORGE_URL is built from; so the domain is
that URL's host. The account is still read, to refuse a user with no
forge account, but its email no longer is.
Refs #5259
push_classifier_branch's request moves into push_branch, which takes the
forge client, so a test can run it against a stand-in forge. The tests
pin the POST to /repos/o/r/contents and its body: base and new branch,
no force overwrite, the message, the operator as author with the
service account as committer, the base64 content, and update with the
blob sha or create without one.
The page's open-a-PR gating becomes rowsProblem in TryOutTypes.ts,
beside the other form checks. swarm-ui has no test runner, so the
typecheck is the only check on it.
Refs #5259
The try-out page gets an "open a PR" button. It posts the page's
classifications to the new POST /api/review-classifier/open-pr, which
merges them by name into the repo's rules file on the default branch
and has the review-classifier forge account commit the result on a new
branch, review-classifier/<user>-<8 hex>, through the contents API. The
commit's author is the forge account named like the session's
Remote-User, read with the controller's admin token. The answer links
to the forge's compare page, where the user opens the PR as themselves.
The push uses the account's existing token from the store
(swarm/services/review-classifier/forge/token, scope write:repository),
so the account needs write access to the repo; a refused push answers
403 naming that. In a YAML rules file, the entries the merge leaves
alone keep their text, as long as the result reads back as the merged
list; otherwise the file is written out whole.
DefaultBranchFiles now carries each file's blob sha, which the forge
needs to replace a file, and the repo's web URL for the link.
Fixes#5259
Refs #5096
TestNats picked a port by binding 127.0.0.1:0, dropping the listener
and handing the number to nats-server. Until that server bound it, the
port was free for a concurrent test's bind(0) to return again. Both
servers were then started on it (with the same store dir, since that was
keyed by pid and port); the loser exited silently and both tests'
clients reached the winner. term_stream's tests all use the stream
term-main-iris, and one of them creates it with max_message_size 4096,
so a test sharing its server got "stream name already in use with a
different configuration" (10058) from open_or_create's
get-then-create.
The server now binds a port of its choice (-p -1) and announces it in a
ports file, and each server has its own tempdir for the store and that
file, so a test can only reach the server it started.
Fixes#5261
#5177 added five forge requests under the controller's token that went
straight to the forgejo client: the commit status in
`Client::create_commit_status`, and in `forge/builder_account.rs` the
hives-team add, the user read, the repo-creation lockdown and the team
membership read. Each now goes through `Api::call`, so a 401 re-reads
the token from the store and sends the request once more, like the
other controller forge calls.
The read-back of the builder's newly minted token in
`ensure_builder_forge_token` stays a direct client: it sends the
builder's own token, not the controller's, the same as the read-backs
in `agent_token`, `classifier_account` and `hive_token`.
The controller read its forge token from the swarm secret store once, when
its client slot filled, and kept sending it until the daemon restarted. A
token the forge re-minted or revoked left every forge-backed path failing
with 401 until someone restarted the controller, and a restart drops the
runtime-only job graph.
Every forge request now goes through `forge::api::Api::call`. On a 401 it
asks a `StoredSecret` (the queue clients' debounced, single-flight re-read
with backoff) for a token to try instead, rebuilds the forgejo client with
it, and sends the request once more. A second 401, or a store still holding
the refused token, surfaces the error as before. Only 401 counts: unlike an
IdP, the forge answers 400 to a malformed request, and a re-read started by
one would put a real 401 that follows inside the backoff.
forgejo-api bakes the token into its own reqwest client's default headers
and takes no client or middleware from outside, so there is no single
request function to hook. Each `self.api.<endpoint>(..)` site becomes
`self.api.call(|api| api.<endpoint>(..).into_future())`; option structs a
site moved into the call are cloned, since the closure can run twice. The
`OnceLock<Arc<Client>>` slot and its `get()` callers are unchanged.
Refs #5017
The swarm settings gain `runtime` (`api-key`, the default, or `acp`) and
an `acp` provider record: an https base URL, the provider's model, and
its prices per million tokens. A PUT now changes only the fields its
body names, so the settings form that sends model, effort and limit
leaves the runtime alone; `null` clears `effort` or `acp`. The body stays
strict, and the stored document keeps #5192's lenient read.
The controller sends the runtime in each run request. On `acp` the
classifier runs one prompt on a new `opencode acp` session per
classification through hive-runtime's AcpRuntime, in a scratch
directory, with a config rendered from the settings. opencode runs
`read` and other tools without asking, so OPENCODE_PERMISSION
{"*":"deny"} takes every tool away from the model, and a deny-all policy
refuses any request on top. The title call, which resends the prompt,
is turned off. The deadline cancels the turn, and a turn that does not
end within 15s of that is dropped, which also covers an agent stuck
before its prompt. The answer is the final assistant text, as ACP has
no result event.
The provider key is read per run from
swarm/services/review-classifier/acp-provider, which the bao UI's
admins login may create and update, and reaches opencode through
AcpRuntime::with_provider_key and the trampoline's allowlist.
Each ACP classification's cost is computed from the configured prices
and the reported tokens, and added to
hyperhive.review_classifier.cost.usage labelled repo, pr, model and
currency.
Part of #5096
swarm-controller README, the swarmctl inference doc comment (and the
regenerated swarmctl-cli.md) and the hive-agent swarm_inference module doc
still listed only model, effort and subagent models.
Nothing capped how many subagents an agent runs at once; the operator
wants to set that per agent from swarm-ui (#5234, shape (b)).
AgentInference gains `max_subagents: Option<u32>`. An absent field
decodes as None, so every stored value still parses and means "no cap",
which is today's behaviour.
swarm-controller accepts 1..=32 on PUT. 0 would refuse every start; 32
is far more concurrent claude sessions than one agent container has
memory for, so anything above it is a typo, not a cap.
The subagent daemon already read the agent's agent-inference key for
subagent_models; it now reads it on every start and continue (not only
those naming a model) and refuses a new run once the cap's worth of
names is tracked, counted under the same lock as the name reservation.
Every tracked slot counts: running, starting, a goal run between turns,
and an interrupted one still exiting. A goal re-prompt never goes
through the reservation, so it is never refused. A failed read applies
no cap, as it falls back for models.
swarmctl's `inference set` does GET, merge, PUT; its mirror type now
carries the field so a set from the CLI keeps a cap set in swarm-ui
instead of dropping it, and `get` prints it.
