The classifier ran every classification on a hard-coded `haiku` with no
`--effort`, and the controller ran at most two requested reviews at once,
also hard-coded, so changing any of them took a code change and a
redeploy.
swarm-controller now keeps one settings document,
`{model, effort, max_concurrent_reviews}`, in its own
`review-classifier-settings` KV bucket under `all`, the way it keeps the
schedule definitions. No document means today's values: haiku, no
`--effort`, two reviews. `GET /api/review-classifier/settings` reads it;
`PUT` replaces it whole, needs the gateway's `Remote-User` like the
try-out POST, and refuses a bad model (the inference route's check), an
effort outside the five levels or a limit outside 1..=8 with
problem+json 400.
Every run request (try-out and requested review) now carries the model
and effort, and the classifier passes them to claude instead of its
constant, which is gone. A try-out reads the settings when it is posted;
a review reads them for the limit before it waits for a place and again
for the model and effort once it has one. The review semaphore is
replaced by a running count under a resizable limit, so a PUT, or the
next review's read, changes how many run without a restart, and a lower
limit lets the runs already going finish. MAX_WAITING_REVIEWS is
unchanged.
Closes#5151, part of #5096.
Every run is recorded when it ends: a try-out when its reply settles it,
a requested review once its review is posted or the post fails, and a
request turned away by a full queue once its comment is posted. A
duplicate request that starts nothing is not recorded. The newest 100
are one JSON array under `recent` in a new controller-owned KV bucket,
`review-classifier-runs`, written with compare-and-set as
`schedule-definitions` is (five attempts). A failed write is logged; the
review is already posted by then.
Each recorded run is also one info event, "review classifier run
finished", with the record's fields as tracing fields. The controller
logs natively to the journal, so they reach VictoriaLogs as F_<NAME>.
GET /api/review-classifier/runs returns the list, newest first, gated
by the gateway like the try-out GET. post_classifier_review now returns
the posted review's html_url, which the record keeps as review_url.
Closes#5153, part of #5096.
An agent's CPU and memory caps, set from the swarm. Keyed by the agent
alone, in a new `agent-limits` KV bucket that swarm-controller alone
writes, the same shape as `agent-inference` (#5034): mara on #4268,
"agents are not tied to a hive. dont tie the limits to the hive."
PUT {"cpu_quota": "400%", "memory_max": "8G"} every field optional,
replaces the stored value
GET the stored caps, {} when there are none
DELETE 204
PUT refuses an agent no hive's wanted state declares (as the inference
route does), unknown fields (422), zero and anything systemd does not take,
before the queue is asked, and stores memory in the spelling systemd takes
(8gb -> 8G). The hive hands these values to systemd verbatim.
The routes sit under /api/, behind the swarm UI vhost's authelia
subrequest like every sibling; the core-toggle module-eval case pins that
gate.
The bucket's reader on the hive comes in the next commit.
Contributes to #3865.
Two read-only routes feed them: GET /api/pulls/repos lists the repos with an
open pull request, and GET /api/pulls?repo=owner/name lists a repo's open
pull requests, highest number first. The page offers the lists through the
browser's own suggestion list and keeps both fields free text, since a run
also works on a merged or closed PR.
Follow-ups from the review of #5142.
A `Done` answer with fewer or more rows than the classifications sent, or
none, now posts the no-verdict comment instead of the classifier's overall
verdict, so a short answer can't approve.
Requests are accepted per repo, PR and head into a set bounded at the two
running plus eight waiting. A request for a key already running or waiting
starts nothing and logs at info; a request with every place taken starts no
run and posts a no-verdict comment saying so. Before, every delivery spawned
a task waiting on the semaphore, so toggling the review request queued one
paid run per toggle.
Table cells get a zero-width space after `#`, `!`, `://` and `www.`, so a
model's reason can't render an issue or PR reference or a bare URL as a link.
The secret publisher's policy denies the classifier's forge token leaf, the
way #5139 denied Grafana's admin password, with a module-eval-bao-grants arm
pinning it. The controller posts reviews with whatever token is there, so a
publisher write picked the account reviews come from.
Closes#5143, part of #5096.
Requesting the forge account `review-classifier` as a reviewer now runs the
repo's rules against the PR and posts the result as that account's review,
on the head commit the run read. A push starts no run; only a fresh review
request does.
- An instance-wide `pull_request_review_request` system hook delivers to
`/webhook/forge/review-request`. A request for any other reviewer, or a
removed request, does nothing.
- The controller reads the PR, loads the rules with the same reader as
`GET /api/review-classifier/rules`, sends one request on the try-out
subject, and posts APPROVED or REQUEST_CHANGES with the table as the body.
- With no clean verdict (PR or rules don't load, the run fails, or any
classification has no answer) it posts a COMMENT saying why, never an
approval. The event is one function, `no_clean_verdict_event`.
- Table cells are flattened, markup-escaped, and every `@` gets a
zero-width space, so a model's reason can't break the table or mention
anyone.
- At most two requested reviews run at once; the rest wait in memory.
- The account is minted the way hive accounts are: a five-minute pass
queues a `MintClassifierForgeToken` node when the token at
`swarm/services/review-classifier/forge/token` is missing or stale. The
token is `write:repository`. The controller's bao policy gains that leaf
(create/read/update), pinned in module-eval bao-grants. The classifier
never sees the token.
The account gets no repo access automatically; a repo admin adds it by
hand. A re-request retires the account's earlier review: Forgejo clears
`official` on a reviewer's earlier reviews when it is requested again
(models/issues/review.go AddReviewRequest, v16.0.5), and approvals and
rejection blocks count official reviews only.
Refs #5137
`GET /api/review-classifier/rules?repo=owner/name` reads
`.hyperhive/review-classifier.yml` or `.json` from the repo's default
branch with the controller's forge token, on every request, and answers
the classifications in the try-out POST's shape, plus the file, the
branch and the commit it was read at. The swarm UI's try-out page loads
them into its editor (#5134).
The file is one `classifications` list of the try-out's `Classification`.
A missing file, both files, a parse error, any unknown key, an empty list
or more than ten entries is a problem+json error, never a pass. The
checks the POST makes on its classifications move into one function both
paths call.
The doc names the file and its schema, replacing the placeholders, and
documents the route.
Part of #5096.
POST /api/review-classifier/try-out validates the run (owner/name repo,
a PR number, one to ten classifications with distinct names, thresholds
up to 100), reads the PR's head commit, commits and diff with the
controller's forge token, sends one request to the review classifier on
the queue and answers 202 with an id. GET .../try-out/{id} reports the
run as running, done (rows and overall verdict) or failed (the error).
Results live in memory for an hour after the reply; there is no run
store. POST then GET because a run can outlast the gateway's 60-second
proxy read timeout. Both routes are utoipa-annotated, so they show on
/api/docs/.
A replay consumer now has a 5s idle heartbeat, and the row stream ends at
the first consumer error instead of skipping it. When the server drops the
consumer, as on a restart, two missed heartbeats end the SSE response and
the browser reconnects with Last-Event-ID. Before, the response stayed
open and delivered nothing.
A Last-Event-ID above the stream's newest sequence now replays as a fresh
reader. The server accepts a start past the end and delivers nothing
until the stream reaches it, so a stale id stalled the replay.
swarm-controller's stream tests run against a real nats-server on
loopback (`test_nats`, binary from HIVE_NATS_SERVER, exported by
checks.nix and devshell.nix). They cover: ensuring the stream twice, an
existing stream with other limits being opened as it is, the replay
arriving before live rows, a reconnect getting every later row exactly
once, and a dropped consumer ending the row stream.
The limits test now compares against the subagent stream's own config.
Two comments that described the agent's terminal route as live-only are
corrected.
swarm-controller now creates `term-main-<agent>` for every live agent, in
the same one-minute pass and with the same config style as
`term-sub-<agent>` (#4897): subjects `$SWARM.term.<agent>` only, max_age
24h, max_bytes 64 MiB, discard Old, max_message_size from the queue's
max_payload. An existing stream is opened as it is. The agent keeps its
plain publish on `$SWARM.term.<agent>` and gets no JetStream subject.
`GET /api/agents/{name}/term/stream` creates an ephemeral push consumer on
that stream starting 100 rows back, so a swarm-ui terminal opens with
recent history and then follows new rows, in order, from one
subscription. Each SSE event carries the row's stream sequence as its id,
and a reconnect's Last-Event-ID resumes after it. Without the stream the
route falls back to the live core subscription it used before.
The stream is `term-main-<agent>`, not `term-<agent>`: an agent may be
named `sub-x`, whose `term-<agent>` would be agent `x`'s subagent stream.
Its one literal subject cannot match `$SWARM.term.<agent>.sub.>`, which
needs more tokens, so the two streams never overlap.
The reader gains `$JS.API.CONSUMER.CREATE.*.*.$SWARM.term.*`, the
extended create form whose last tokens are the consumer's filter. Checked
against nats-server 2.14.4, the trailing `*` also admits a literal `*` or
`>` filter, so a consumer under it reaches any terminal row, main or
subagent; those are rows the reader already subscribes to and reads with
`STREAM.MSG.GET.*`.
Closes#4826
`check_model` accepted a value like `--dangerously-skip-permissions`,
and the stored model ends up in the agent's `claude` argv. A PUT whose
`model`, or any `subagent_models` entry (same validator), starts with
`-` is now a 400 with the same problem+json shape as the other model
checks. A dash inside a name (`claude-opus-4-8`) is still accepted.
Raised by argus on #5084.
Refs #5034
Each agent now reports the model and effort it runs on, and where each
comes from (`swarm`, `agent_nix`, `local_file` or `default`), in its
turn-state header on `$SWARM.agent-state.<agent>`: `effort`,
`model_source` and `effort_source` join `model`. The shared
`swarm_queue_client::agent_inference::{Effective, Source}` types spell
those fields for both ends.
The Bus keeps a source next to the model and effort. `Bus::new` takes it
from the local fallback chain (`harness_state::local_model` /
`local_effort`, which now return the source with the value), a `/model`
or `/effort` pick sets `local_file`, and the swarm poller sets `swarm`
or, on a clear, the local chain's source. A swarm value equal to the one
the agent already runs is now applied once, so the source moves to
`swarm` and the header is republished; before, `next()` skipped it as
unchanged.
swarm-controller subscribes to every agent's header with the reader
grant it already holds for the state relay, and keeps the latest
inference fields per agent in memory (`inference::Reported`).
`GET /api/agents/{name}/inference` now answers
`{configured, effective}`, `effective` being `null` until the agent's
first header with these fields since the controller started.
`swarmctl agent inference get` prints both; `set` reads `configured`.
Refs #5034
swarm-controller keeps each agent's inference settings in a new
`agent-inference` KV bucket, one key per agent, and is its only writer
(`GET`/`PUT`/`DELETE /api/agents/{name}/inference`; `swarmctl agent
inference get|set|clear`). Every field is optional. Each agent reads its
own key every 10 s through a direct-get grant on that key alone, and a
set `model` or `effort` replaces the Bus value, so the next turn runs on
it and the badge follows.
The swarm value is applied without persisting it: `Bus::set_model` and
`set_effort` are split into a persisting setter and a non-persisting
`apply_*`, and the poller uses the latter. Otherwise the swarm value
would overwrite `hyperhive-model`/`hyperhive-effort`, and clearing it, or
a restart, would no longer find the agent's own value. An unset field
changes nothing; a field cleared after being applied goes back to the
local value (`harness_state::local_model`/`local_effort`, the same
resolution `Bus::new` uses), read again at the clear.
`subagent_models` is validated, stored and shown by `get`; no agent
applies it yet.
Refs #5034
forgejo-api's client has no request timeout, so a forge that accepts a
connection and never answers held a forge-calling job node forever:
the node never reached Done or Failed, and kept its executor task.
Every executor arm that awaits the forge now runs its work under the
same deadline the periodic passes use. On elapsing, the node fails
through its existing Outcome::Failed path with
"<node>: no answer from the forge within 120s".
Nodes: create_repo, create_forge_user, add_repo_member,
grant_docs_read, init_agent_config_repo, mint_agent_forge_token,
mint_hive_forge_token.
The bound is per node, not per request: the nodes make 1 to about 15
sequential forge and store requests each (mint_hive_forge_token the
most), and the requests live inside forge::Client methods the periodic
passes share. One 2-minute bound covers both, so #5065's
within_pass_deadline / PASS_DEADLINE become within_forge_deadline /
FORGE_DEADLINE, with a log line that no longer says "retrying next
tick" (true of a pass, not of a node).
The InitAgentConfigRepo arm is lifted out of run_swarm_node, like its
siblings, to keep that function under clippy::too_many_lines.
Refs #5066
A controller that started before the forge's host had published its token
ran without forge access until someone restarted it: on a split-host first
deploy that is a manual step. Now the forge client lives in a slot that is
filled once (`Late<T>`, an `Arc<OnceLock<T>>`):
- `forge::enable` tries the store once at start. On failure it logs once and
re-reads in the background with backoff (30s doubling to a 5-minute cap,
no log per retry), fills the slot and logs once when the token is there.
- The forge passes (#5014's per-hive token mint among them), the config-PR
poll and the webhook registration start from that moment, through
`start_forge`'s callback.
- Job nodes and HTTP handlers read the slot when they run, so they answer
"no forge" before and work after.
A token replaced in the store after the slot is filled still reaches the
controller with its next restart. The round-1 change that shared
`queue_identity::fetch_with_retry` is reverted; nothing else uses it.
Refs #4998
The controller's forge token never entered bao: the forge minted it in its
container, hive-forge-swarm-controller-token copied it to a host file, and the
controller LoadCredential'ed that file. forgeTokenFile defaulted to that path
only where the forge ran, so a controller on any other host had no forge
access unless an operator copied the file by hand.
Now the forge's host publishes it and the controller reads it:
- swarm-bao-forge-publisher (replaces hive-forge-swarm-controller-token)
waits for the in-container token, logs in with its own forge-publisher leaf
and writes it to swarm/controller/swarm-controller/forge/token. It runs on
every forge container start, the old unit's trigger.
- New principal swarm-forge-publisher: leaf minted by glue-bao-tls.nix,
policy create/update on that one path, written by
swarm-bao-forge-publisher-policy, CN reserved as a hive name.
- The controller policy gains read on that path. swarm-controller reads it
at start under its own certificate with the OIDC secret's retry, and runs
without forge access if it still cannot.
- forgeTokenFile, its LoadCredential and the forgeEnv gate are gone;
SWARM_CONTROLLER_FORGE_URL is always set. hostSwarmControllerTokenFile is
removed, and a tmpfiles rule deletes the old host copy on the forge host.
The in-container file stays: the mint unit's only-once guard reads it, and
it is the publisher's source.
Refs #4998
Agent creation gains a GrantDocsRead node (after CreateForgeUser) that adds
the agent as a read collaborator on internal/docs, and forge::docs_grant
runs a pass over the store's hive-agent-* roster at start and every five
minutes that queues the same job for any agent that is not one. The pass only
adds; an unreadable roster or forge read changes nothing.
hive-c0re drops shared_docs_access, its call in sync_agent, and the
SHARED_ORG / SHARED_DOCS_REPO constants only it used.
`hive-<hive>` is each hive's read-only forge account, and an agent's forge
account takes the agent's name, so an agent called `hive-<x>` would claim
hive `<x>`'s account. `hive_types::reserved_agent_prefix_violation` names
the rule, and `broken_name_rules` applies it, so both creation paths that
go through it (`POST /api/agents` and placing an undeclared agent on a
hive) refuse such a name for a new agent. An agent already in the roster
keeps its name with a warning, as for every other naming rule.
The email check between hive and agent accounts stays as the second
guard, for agents created before this rule.
Refs #4851
swarm-controller now gives every hive in its directory a forge account,
`hive-<hive>`, in a new `hives` team of `agent-configs` (read on
`repo.code` + `repo.pulls`, every repo including later ones), and mints
one `read:repository` token named `swarm-hive` into the store at
`swarm/hives/<hive>/forge/token`. A five-minute pass queues a
`MintHiveForgeToken` node for a hive whose token is missing or no longer
matches the forge, whose account is missing, or whose team membership or
team settings drifted. A hive whose store or forge read failed is never
queued, so an outage cannot rotate every hive's token. The token decision
reuses `agent_token`'s classifier and replace step, now parameterised by
name and scopes.
An agent may be called `hive-<x>`. The hive account is created with
`hive-<hive>@hives.hyperhive.local` and the job refuses an account by that
name with any other email before it locks it down, adds it to the team or
mints for it; agent creation already refuses the hive's account by the
same email check.
The controller's bao policy gains the `swarm/hives/+/forge/token` leaf
(create/read/update), pinned by the bao-grants module-eval assertion. The
hive's own policy already reads `swarm/hives/<hive>/*`.
hive-c0re reads that token through `forge::reader_token()`, falling back
to the core token file when the store holds none, and uses it for:
fetching a merged config rev (swarm deploy, `hivectl forge
reconcile-config`), reseeding a purged `applied`, the meta-update
cascade's fast-forward, seeding `proposed` on first deploy, and finding
the PR that merged a refused rev. Those reads no longer require a local
`hive-forge` container. The `/knowledge` clone moves from the forge
host's boot sweep into the knowledge pull, so every hive clones it.
Posting the deploy-failure comment on that PR stays on the core token.
Refs #4851
The CI runner was a hive deploy module: hive-ci.nix ran it beside the
forge, and hive-c0re registered it through the forge's admin API with
the `core` token, writing the registration token through hive-priv. It
could only run on the forge's host, and it kept an admin-scoped forge
token on a hive.
Now the runner is a swarm service:
- swarm-ci-runner.nix runs it in a `swarm-ci-runner` container on the
host that sets deploy.ci-runner.enable, reaching the forge by name.
It needs neither the forge nor a hive there.
- swarm-controller reads the forge's runner registration token with its
own admin token and writes it to swarm/services/ci-runner/
registration-token whenever it differs from what it last wrote. Its
grant on that path is create/update only.
- The runner's own leaf (CN swarm-ci-runner, role and policy of the same
name) reads that one path. swarm-bao-ci-runner.service fetches it into
the runner's EnvironmentFile and restarts the runner on a change.
- A pre-start step removes .runner when the daemon's journal since its
last start holds Forgejo's "unregistered runner" refusal (the daemon
exits 1 when its startup Declare gets it), or when .runner names
another forge address; upstream's pre-start step then registers again.
Removed from the hive side: hive-ci.nix and deploy.forgejo.ci.*,
forge/ci_runner.rs, PrivRequest::RegisterCiRunner and its handler and
ReadWritePaths entry, InfraContainer::Ci with hivectl's --ci and the
dashboard's hive-ci journal entry, HYPERHIVE_FORGE_CI_ENABLED, and
read:admin,write:admin from CORE_TOKEN_SCOPES: nothing else in hive-c0re
calls /api/v1/admin/*. The old ci options, under both swarm.forge.ci and
deploy.forgejo.ci, now fail evaluation naming deploy.ci-runner.
The forge's half of the old ci.enable, the actions/checkout mirror and
DEFAULT_ACTIONS_URL = self, is deploy.forgejo.actionsFromSelf.
Closes#4954
The operator admin sync now also takes homeserver admin away. It records
each user id it sends `make-user-admin` for in
`operator-matrix-admins.json` under the controller's state directory
(written to a temporary file and renamed over it; a missing file is an
empty set). On each pass, a recorded account that is no longer in
authelia's `admins` group but is still joined to `#admins:<server>` is
sent `!admin users force-leave-room <user> <admin room id>` as `@swarm`,
over the same `/_matrix` path the grant uses. Leaving that room is what
ends tuwunel admin.
It never revokes an admin it didn't grant, `@swarm`, tuwunel's server
user (`@conduit:<server>`), an agent's account (the store's
`hive-agent-*` cert roles, the same roster the agent account mint uses),
or anyone in the new `services.hyperhive.swarm.matrix.adminSync.ignoreUsers`
list. That list reaches the daemon as
`SWARM_CONTROLLER_MATRIX_ADMIN_SYNC_IGNORE_USERS`. It also never sends a
revoke that would leave `#admins` with no member besides `@swarm` and the
server user.
A revoke is sent at most once per account per controller process, like a
grant. A recorded account that is out of both the group and the room is
dropped from the file.
Closes#4799
- swarm.serviceDomains lists every swarm service name on every host. The
store's issuing role is built from it, so the store now admits a leaf
for grafana, metrics, otel, logs, the bao UI and the apex whichever host
runs the store. Each gateway's own leaf stays narrowed by
localServiceDomains.
- swarm.matrix.apiUrl defaults to http://<swarm.matrix.gatewayHost>
(null when gatewayHost is null) instead of loopback on the matrix host.
- The authelia vhost proxies /bridge/ to swarm-authelia-bridge with the
prefix stripped, and deploy.swarm-controller.authBridgeUrl defaults to
${swarm.authelia.url}/bridge on every host. swarm.authelia.bridgeUrl is
removed.
- swarm-authelia-bridge admits only swarm-controller's OIDC client on
every request (403 for any other active token), configured by
SWARM_AUTHELIA_BRIDGE_CONTROLLER_CLIENT_ID (renamed from
..._OPERATOR_READER_CLIENT_ID, which gated ListOperators alone).
- swarm.peerHives is removed; swarm-wireguard.nix filters this host's own
hive out of swarm.hives itself.
Closes#4933
swarm-controller keeps every scheduled prompt (id, name, agents, body,
anchor, interval, paused) as one document under one key of a new
`schedule-definitions` KV bucket, so each change is a single
compare-and-set and a prompt exists for all of its agents or none. It
writes each agent's key of a new `agent-schedules` bucket as the prompts
naming that agent, without names or other agents, after every change and
on a one-minute reconcile pass. It is the only writer of both, behind
GET/POST /api/schedules and PATCH/DELETE /api/schedules/{id}.
`swarmctl schedule add|list|edit|pause|resume|rm` drives those routes over
the controller socket.
An agent's own queue credential is granted one more subject, the direct
get of its own key in `agent-schedules`. A KV watch is not granted: its
consumer create subject carries no key, so it would cover every agent's.
hive-agent's new schedule ticker reads that key every 10 seconds and fires
due slots as the same `schedule:<id>` todo hive-c0re's worker pushes. Fire
times are anchor + k*interval; the last fired slot per schedule id is kept
in the state db, so a body edit or list refresh never moves or refires a
slot, downtime fires the latest missed slot once, a restart refires
nothing, a one-shot fires once, and paused slots are skipped.
Refs #4914
swarm-authelia-bridge gains ListOperators, the usernames in authelia's
operator group (`operatorGroup = "admins"`, passed in from
swarm-authelia.nix). Unlike the two agent verbs, which admit any active
token, it answers only the client named by
SWARM_AUTHELIA_BRIDGE_OPERATOR_READER_CLIENT_ID (swarm-controller's
queueClientId) and returns 403 to every other active token: authelia
issues machine tokens to every hive and its agents as well. The bridge's
introspection now models `client_id` to make that check, the same way
swarm-nats-auth's does.
swarm-controller runs a grant-only pass at start and every five minutes:
read the group, map each username to the account an SSO login lands on
(tuwunel lowercases `preferred_username`), skip anyone already joined to
`#admins:<server>` (tuwunel's definition of admin) or without a matrix
account, and send `!admin users make-user-admin <user>` there as @swarm
with the swarm appservice token. Nothing is ever revoked: tuwunel has no
command for it (#4799).
The admin-room calls (alias, joined members, profile, send) go into
swarm-matrix-client next to whoami, keeping its no-body-in-errors rule.
Closes#4585
`mint_agent_identity_node_runs_end_to_end_and_fails_without_the_agent_pki_named`
runs a `MintAgentIdentity` node, which calls `store::connect()` directly
rather than through `AppState`. It stops before that login only while
`SWARM_CONTROLLER_AGENT_PKI_MOUNT` and `SWARM_CONTROLLER_AGENT_PKI_ROLE`
are unset; with both set and `BAO_*` set it reaches a real store. The test
now refuses to run with them set. It still runs with `BAO_*` set, since
without the agent PKI it never gets as far as the login.
The seven linked-account handlers (matrix, forge and github link and
delete, and the list) read their store from `AppState::accounts` instead
of calling `store::connect()` themselves. In production that is
`Accounts::Live`, which still logs in to the swarm's secret store on
every request: a `SecretStore` holds a token it never renews, and a store
that is down at startup must fail only the requests made while it is
down.
Tests set `Accounts::Fake` over the in-memory `FakeStore`. The eight
handler tests that proved validation passed by expecting the store
connect to fail with 500 now assert that the write or delete landed in
the fake, and lose their `BAO_*`-must-be-unset guards: with `BAO_*` set
they no longer reach a real store.
The agent view in the swarm UI gets a message box. The text reaches the
agent's harness straight from swarm-controller over the swarm queue and
becomes an inbox turn from the operator's authelia username. hive-c0re is
not in the path.
- gateway: `/api/` forwards authelia's `Remote-User`, replacing any the
client sent.
- swarm-controller: `POST /api/agents/{name}/input` takes that header as
the sender (401 without it) and sends `{from, body}` as a NATS request
on `$SWARM.input.<agent>`. No subscriber is a 503.
- swarm-nats-auth: the controller is granted publish on `$SWARM.input.*`;
no hive or agent grant names the family, so the harness can trust
`from`.
- hive-agent: a task subscribes to the agent's own input subject and
pushes each message into the serve loop's existing reminder channel,
renamed `local_msg_*` since it now carries both.
The harness does not reply to the request: no agent grant covers a reply
subject. The controller therefore answers 202 once the request times out
(2s) and 503 on the server's no-responders status.
Closes#4828
swarm-controller now creates `term-sub-<agent>` for every agent a hive is
declared to run, at start and every minute after, with the config
`swarm_queue_client::subagent_term::open_or_create` spells (subjects
`$SWARM.term.<agent>.sub.>`, max_age 24h). An existing stream is opened as
it is, as the controller does for its other streams and buckets, under the
`$JS.API.STREAM.CREATE.*` grant it already holds.
The agent no longer creates the stream: its token is granted publish on
`$SWARM.term.<agent>.sub.>` and no `$JS.API.STREAM.CREATE|INFO` subject,
and the subagent daemon only publishes. A `CREATE` carries the stream's
config in its payload, which no subject grant narrows, so the agent could
otherwise pick the stream's subjects and limits.
An agent's subagent daemon publishes each subagent's output as terminal
rows on `$SWARM.term.<agent>.sub.<subagent>`, as the agent, into a
per-agent stream it creates itself; swarm-controller lists an agent's
subagents from that stream's subjects and relays one subagent's rows as
SSE; the swarm UI lists them under the agent's terminal preview and
reuses AgentTermPreview, full-screen tab included, with no input.
- swarm-nats.nix: the agent token may also publish
`$SWARM.term.{agent}.sub.>` and `$JS.API.STREAM.CREATE|INFO` on
`term-sub-{agent}`, and nothing else of JetStream. A module-eval arm
pins the agent-token grant as an exact list.
- mcp.nix: hive-subagent-daemon loads the agent's store identity
(`hive-agent-bao-cert/-key/-server-ca`, the ones hive-agent loads)
whenever the agent has a store, not only on the opencode preset. The
agent's own queue secret lives in the store, so this is the credential
the harness connects with.
- hive-subagent-mcp: `swarm_term` reads the agent's queue secret under
that identity, connects with the agent token, opens or creates
`term-sub-<agent>` (max_age 24h), and publishes classified rows from
the sink every subagent line already passes through. The sink only
queues (bounded, drop-and-count); a missing store, refused credential,
failed stream create or failed publish is a log line.
- The stream-json classifier (`stream_enrich`) and the `TermMsg` row
types plus `fit` move from the hive-agent binary into hive-sh4re, so
the subagent daemon publishes the rows AgentTermPreview already
renders. hive-agent keeps its LiveEvent classifier on top.
- swarm-controller: `GET /api/agents/{name}/subagents` and
`GET /api/agents/{name}/subagents/{subagent}/term/stream`.
- docs/swarm: what the UI shows and what the queue carries.
Closes#4827
Each row of an agent's linked accounts, except its own `main` matrix
account, gets a delete action. swarm-controller serves DELETE beside each
PUT (matrix-accounts/{account}, forge-accounts/{label}, github-account),
answers 404 for an account the store does not hold, refuses `main`, and
removes every version through `delete_all_versions`.
The matrix confirmation has a revoke checkbox, off by default: the
controller logs the stored token out at its homeserver first, and keeps
the account when that fails or no homeserver is stored.
The controller's policy gains `delete` on each agent's
`metadata/.../matrix/+`, `forge/+` and `github-token`, pinned in
bao-grants.nix.
Refs #4855
A config PR merged in the Forgejo UI changed nothing on the hive: the
hive's webhook ignores `closed`, its poll then cancels the dashboard
card, and `applied/main` stays where it was.
swarm-controller reads `merged`/`merge_commit_sha` off the
`pull_request` delivery it already receives for `agent-configs`, finds
the hive placing the agent by scanning every hive's wanted state (the
scan `declarations_elsewhere` already ran, factored out), and queues a
`TriggerDeploy` carrying the rev. Zero or several claimants deploy
nothing and log the claimants.
`DeployRequest` gains `rev: Option<String>` with `serde(default)`, so
rev-less payloads from either side keep decoding.
hive-c0re, given a rev for an agent it runs: a no-op when
`applied/main` already is the rev (a dashboard merge deploys its own
PR); otherwise it fetches the forge `main` with the core token,
requires the rev to descend from `applied/main` (the ancestry gate,
factored out of `run_deploy_merge_verify`), fast-forwards by CAS and
queues the usual relocking rebuild. No eval-verify on this path, per
mara (#4850 c90075). A refusal is commented on the PR that merged the
rev, found by commit.
swarm-controller's forge-objects pass converges every config repo's
`main` rule to merge whitelist `operators` + `core` and approval
whitelist `operators`. The hive's boot PATCH stops forcing
`enable_approvals_whitelist` off, so the two do not fight.
Refs #4850
GET /api/hives/{hive}/agents/{agent}/linked-accounts returns one row per
account linked to the agent, as kind, name and host: each matrix account
under swarm/agents/<agent>/matrix (with its homeserver, and the agent's own
`main` marked reserved), each forge label under swarm/agents/<agent>/forge
(with its url), and github when swarm/agents/<agent>/github-token exists
(host github.com, which is not stored). No credential field is in the
response type.
Listing those two directories needs a new controller grant: `list` on
secret/metadata/swarm/agents/+/matrix and .../+/forge only, pinned in
bao-grants.nix as the only metadata stanzas under agents/ beside the queue
revocation. Checked against a dev OpenBao 2.6.3: the grant lists those two
directories and is refused on agents/, agents/<agent>/, and a leaf.
The swarm UI agent detail panel shows all rows under "accounts"; the table
view's matrix column shows the matrix rows. The link badges stay.
Refs #4855
An operator links an agent's GitHub personal access token in the swarm UI
(LinkGithubAccountForm, "link github account" on /agents). swarm-controller's
PUT /api/hives/{hive}/agents/{agent}/github-account stores it at
swarm/agents/<agent>/github-token (swarm_secret_client::github), a flat leaf
under the agent's prefix that the agent's existing read grant already covers:
no policy change, and no list grant, since there is one token per agent.
In the agent, hive-agent-github-token (oneshot + 2-minute timer, as the agent
user, under its own store certificate, ordered before hive-github-notify)
reads that path and writes <state>/github-token, 0600 and agent-owned, the
file the gh wrapper, git credential helper and hive-github-notify already
read. It replaces the file by rename only when the bytes changed and never
deletes it: a hive-written github-token stays until a token is linked in the
swarm UI. It is installed only with a store address and
services.hyperhive.agent.github.enable.
Removed: the dashboard's CR3D3NTIALS page (credentials.html/js/css, its
build entries and H0M3 tile; GITHUB was its only tab), hive-c0re's
dashboard/matrix_accounts.rs with GET/POST /api/github-account,
priv_client::write_agent_github_token, the host socket's
SetAgentGithubToken and `hivectl github set-token`, and hive-priv's
WriteAgentGithubToken with write_agent_state_file, its only caller gone.
Docs: integrations/github.md and swarm/ui.md describe the swarm path,
swarm/credentials.md gains the store-path row, and the hive UI docs,
hivectl docs and security.md's hive-priv table drop the removed pieces.
Closes#4347
An operator now links an agent's external forge account (label, base URL,
token) in the swarm UI. swarm-controller stores it at
swarm/agents/<agent>/forge/<label>. There is no index: the store's
listing of the agent's forge/ directory is the set of accounts.
In the agent, hive-agent-forge-accounts (oneshot + 2-minute timer, as
the agent user, under its own store certificate) lists
swarm/agents/<agent>/forge/ with the `list` #4866 grants an agent on its
own metadata subtree, reads each account, and writes
<state>/forge-<label>-token and forge-<label>.json in the names and shape
hive-forge -f already reads. An empty listing (a 404, which `bao kv list
-format=json` answers with `{}` and an empty stderr) is zero accounts; a
denial or an unreachable store fails the unit. It never deletes: files
for labels not listed, including ones the hive wrote, stay as they are.
Removed: the dashboard FORGES tab (credentials.js/html section and its
CSS), hive-c0re's extra_forges.rs and its routes, priv_client's
extra-forge calls, and hive-priv's WriteAgentExtraForgeAccount /
DeleteAgentExtraForgeAccount with their helpers. The GITHUB tab and
WriteAgentGithubToken stay.
Also: persistence.md's matrix avatar note names the exit-75 restart on a
changed account listing, not the dashboard, as what brings a linked
account up.
Refs #4348
render_agent gains a second stanza: list on
secret/metadata/swarm/agents/<agent>/*, next to the existing read on
secret/data/swarm/agents/<agent>/*. An agent can now learn which
credentials it holds by listing its own subtree. Metadata read, writes
and every other principal's paths stay refused.
An agent's policy was only written when it was minted, so existing agents
would never get the new stanza. swarm-controller now rewrites every
agent's policy at start (read_policy::ensure_agent_policies), with the
same 30s / 24h retry as ensure_hive_access. The roster is the store's
hive-agent-* cert-auth roles, listed with the controller's existing
`list` on auth/cert/certs; the writes use its existing grant on
sys/policies/acl/hive-*. Only the policy is written: mint_and_verify
also reissues the certificate, so the pass does not call it.
Refs #4348
A hive whose homeserver runs on another host has no local
matrix-appservice-token, so hive-c0re's matrix sweep returned before
reaching the store read in ensure_hive_user: no @hive-<name>: token, no
Space, no chat room, no invites, and a sweep-health banner.
swarm-controller now mints @hive-<name>: with the swarm appservice
token for every hive in its directory, as a MintHiveSenderToken job
node queued by a five-minute pass, and stores it at
swarm/hives/<name>/matrix/sender-token, the same matrix::Credential
swarm-matrix-ctl writes there. It is keep-if-live, reusing agent_token's
classify/plan: a stored token whoami confirms as @hive-<name>: is left
alone, so only an absent or dead one is minted. agent_token's probe and
mint steps are lifted into probe_at/mint_at so both passes share them.
swarm-matrix-ctl mint still writes the path for its own hive when it is
empty. If both mint an empty path at once, one token is invalidated
(same pinned device); the next pass classifies it Revoked and re-mints.
hive-c0re's ensure_all no longer returns when there is no local
as_token. ensure_hive_user reads the store first on every sweep and
overwrites its token file when the store's token differs, keeps the
file when the store has none, mints with the local as_token only when
neither holds one, and fails with one error when there is nothing at
all. The decision is sender_source, unit-tested.
The controller's bao policy gains create/read/update on
swarm/hives/+/matrix/sender-token (`+`, since `*` is a glob only at the
end of a path), pinned in module-eval.
Refs #4427
The sweep ran once at startup and was never retried: a boot where
store::connect or the roster read failed (e.g. the controller up
before bao) left the tokens live until the next restart. It now runs
off forge::agent_token::spawn's five-minute mint pass, reusing that
pass's roster observation instead of a second store/roster read, so a
failed first tick retries on the next one. Idempotent, so a re-run
after a partial sweep deletes nothing extra. Drops the one-shot
startup spawn; one call path.
Also updates the forge.md and agent_token.rs docs that still said the
legacy hyperhive-<seconds> tokens stay until manually removed.
Refs #4644
Before b5d07d4d, hive-c0re minted a new `hyperhive-<unix-seconds>` token
for an agent on every spawn and rebuild and never revoked one, so every
live agent's forge user carries a pile of write-scoped tokens nothing
holds. Nothing in the tree lists or deletes them.
On each start, swarm-controller now walks the store's hive-agent-*
roster (the one the swarm-agent mint pass walks), and for every agent
whose swarm-agent token that pass would keep, deletes each token named
exactly `hyperhive-<digits>`. It logs the count per agent and a total.
- `core` is refused by name in both the roster filter and the per-user
delete: hive-c0re still names core's live admin token
`hyperhive-<unix-seconds>`.
- An agent whose swarm-agent token is not current is skipped, because
consumers fall back to `<state>/forge-token`, the last hyperhive-*
token, until the swarm token is fetched.
- A failed list or delete is logged and skipped; the sweep does not
retry and never blocks startup. A second start deletes nothing.
Tokens on forge users of agents no longer on the store roster (already
destroyed) are not reached.
Closes#4644
Exempting offline and paused from the name rules let a reserved name be
placed anyway: a first `offline` declared it, and the `up` after it
passed as an agent already declared on the hive. `paused` alone sufficed,
since the hive deploys an absent agent declared paused. A first
declaration in any placing state now runs the name rules and the
placed-elsewhere check; an agent already declared on the hive skips both.
A rule-breaking name whose roster read fails, including when no identity
bridge is configured, was accepted with a warning nobody sees, as in
create_agent. No later step on this route refuses the name, so it now
refuses with 503.
The gate is taken only for a placing declaration, so a destroy and its
credential revocation no longer wait on creations.
Refs #4804
The state route ran the name rules and the placed-elsewhere check on
every up/offline/paused declaration. An agent already declared on the
hive whose name breaks a rule, and which is not in the roster, could then
only be destroyed from the swarm.
Now an agent this hive already declares in a placing state passes both
checks. For a first declaration the placed-elsewhere check still applies
to up, offline and paused alike, and the name rules to up only: offline
and paused are how an operator stops an agent. The route reads this hive's
declaration to tell, and refuses with 503/500 when it cannot.
Refs #4804
PUT /api/hives/{hive}/agents/{agent}/state wrote any identifier into any
hive's wanted state, and the hive first-deploys a declared agent it has
no container for. That skipped create_agent's checks on the name.
A declaration that places the agent (up, paused, offline) is now refused
with 400 for a new name that breaks a naming rule, and with 409 for a
name the swarm has placed on another hive. Both reuse create_agent's
helpers (broken_name_rules/name_verdict, placements_elsewhere), under the
same gate, and an unreadable wanted state on another hive refuses with
503/500 as creation does. `destroyed` places nothing and is not checked,
so the revocation path is unchanged.
An agent with no declaration at all is still accepted: swarm-ui declares
state for agents that predate swarm-level creation, which is how the swarm
adopts them. Whether to refuse such names instead is left open on #4804.
Refs #4804
create_agent read the other hives' wanted state first and snapshotted the
queued SetAgentWanted nodes second. A node finishing between the two was
in neither — not yet declared at the first read, already terminal at the
second — so a second hive could get the same name.
The queue is now snapshotted first: a SetAgentWanted node only turns
terminal after its declaration is written, so anything terminal by then is
visible to the read that follows. The order lives in
`placements_elsewhere`, and a test that finishes a node between the two
reads fails with them swapped.
Refs #4396
swarm-controller's POST /api/agents now refuses (409) a name the swarm
has already placed on a different hive: a non-Destroyed declaration in
that hive's wanted state, or a SetAgentWanted node still queued for it.
The same name on the same hive is that agent being re-created and goes
through. A wanted state that cannot be read refuses (503/500) instead of
reading as "placed nowhere". Creations are serialised from that read to
the graph insert so two concurrent creations of one name cannot both
pass.
Hive-level creation is removed: hivectl `agent create` / `request-create`,
HostRequest::Spawn / RequestSpawn, the dashboard POST /api/request-spawn
route, and ApprovalKind::Spawn with its approve/resolve arms and the
approval-carrying `templates::spawn`. The swarm path (deploy request or
wanted-state sweep -> queue_first_deploy -> templates::first_deploy) used
none of them. Old `spawn` approval rows are skipped by collect_lenient,
as `init_config` rows were in a3b672d1.
policy.rs's comment on agent_object_name stated swarm-wide name
uniqueness as a fact; it now says where it is enforced and what that
check cannot see.
Refs #4396
rustdoc runs with -D rustdoc::broken-intra-doc-links and the type is
imported inside the function body, not at module scope, so the bare
link resolved to nothing and failed the workspace-doc derivation.
Pins the three properties the revocation rests on and cannot check
against a store: that only `Destroyed` revokes (a revocation on
`Offline` or `Paused` would give an agent that stops and never
restarts), that a 404 is absence while a 403 stays a failure, and that
the delete addresses `secret/metadata/` -- the path that takes every
version, which is the string the grant has to match.
docs/swarm/credentials.md gains the revocation section and its table
cell stops describing the deletion as something an operator does by
hand.
A per-agent queue credential is minted at agent creation and nothing has
ever removed it. An agent declared destroyed loses its container and
keeps its credential: a bearer secret recovered from a snapshot or a
stale capture still authenticates as that agent, so the set of usable
credentials only grows.
Delete the path the mint published, on the one transition that ends an
agent's life. It mirrors step 3 of `mint_and_verify` and no other step:
the leaf, the ACL document and the cert role are what a hive uses to
collect an agent's secrets and are re-minted on every run of the mint.
Every version, not the newest. The mint rewrites the path when the
principal it names needs correcting, so KV v2's plain delete would leave
the identical secret readable at ?version=N. That is a separately-ACL'd
path, hence the second stanza in the controller's grant -- `delete` on
metadata discloses nothing, and `update` on the data path already lets
this principal destroy any agent credential's usability.
The destroy is not blocked by a failed revocation: the declaration is
already published and refusing the call would leave an operator with an
agent they cannot tear down. The failure is logged at error instead,
naming the agent, since a silent orphan is the fault being removed.
A five-minute pass over every agent some hive's wanted state declares as
anything but destroyed queues, per agent:
- `MintAgentIdentity` (the node agent creation uses) when the stored
certificate at swarm/agents/<agent>/bao-mtls is past half its validity,
read from its own notBefore/notAfter: day 45 of the role's 90;
- the new `RenewAgentQueueCredential` node when the queue secret at
swarm/agents/<agent>/queue is 45 days old or has no mint time. The node
re-decides, writes a fresh value with `minted_at`, reads it back, and logs
the agent and the old age.
When both are due the secret node runs after_any the certificate node,
because mint_and_verify compares the queue secret it read with the one it
reads back. A credential that is not stored is never created here.
`queue::AgentCredential` gains an optional `minted_at` (unix seconds);
agent creation now sets it. Stored objects without it decode unchanged and
count as due, so every existing queue secret is re-minted on the first pass.
Both replacements reach the agent at its next start. The old certificate
stays valid until it expires; the old queue secret does not, so a queue
reconnect before that restart is denied.
Adds x509-cert 0.2 (with der_derive and flagset) to read the validity.
docs/swarm/credentials.md: the renewal column splits into automatic re-mint
and automatic re-pull, filled from the code as it stands.
The controller's OIDC client secret (client `swarm-controller`, used for
the queue connection, the auth-bridge bearer and the OTLP push) came from
an operator-placed file, `deploy.swarm-controller.queue.clientSecretFile`,
handed in by `LoadCredential=`.
Now `swarm-secret-publish`, which already copies authelia's minted OIDC
secrets into the store, also publishes this one, to
`swarm/controller/swarm-controller/oidc/client`. That path sits under
`controller/`, which no hive's policy reads. The controller reads it once
at start with its existing store certificate and holds it in memory, as
`swarm_queue_client::ClientSecret::Value`. If the store is down, it
retries for about a minute and then fails the start, so `Restart=` tries
again.
Policy delta: the controller gets `read` on that leaf, and the publisher
gets `create`/`update` on that leaf.
Removed: the `queue.clientSecretFile` option (both spellings, now removed
options with a message), its singleHostSwarm default, the credential and
placeholder, and the path watcher plus its restart oneshot. A controller
without a store identity is now an eval error, because it has no other
way to get the secret.
The auth callout grants an agent that presents its own queue credential
one more subject, `$KV.agent-icons.<agent>`: its own key in the
agent-icons bucket and no other. The hive's shared agent client is
granted none of the bucket, since every agent on a hive presents it.
hive-agent writes `/etc/hyperhive/icon.svg`, the file its `GET /icon`
serves, to that key once per start, as a JetStream publish straight to
the subject (what `kv::Store::put` sends, minus the bucket lookup), so
the one subject is the whole grant. No icon deletes the key. A failed
write, including one that arrives before the bucket exists, is retried
with backoff until acked. An agent connected with the hive's shared
client publishes nothing.
swarm-controller creates the bucket as soon as its queue connection is
up, instead of on the first icon read, so an agent's write does not
wait for someone to look.
Measured against a local nats-server with a user allowed publish on
`$KV.agent-icons.atlas` only: the write to its own key is stored and
readable, a write to `$KV.agent-icons.argus` is refused (the ack times
out), the DEL marker makes the key read as absent, and a write before
the bucket exists fails with "no responders".
An agent is not fixed to a hive, so its icon cannot be resolved as
hive -> agent. This adds the swarm-level half: an `agent-icons` KV
bucket keyed by the agent name alone — no hive token, so an agent that
moves hives keeps its icon and one that is stopped still has one — and
`GET /api/agents/<name>/icon` on swarm-controller serving it
same-origin, like every other `/api/*` route swarm-ui calls.
404 is the "this agent has no icon" answer, the same contract the
per-agent harness's own `GET /icon` has for an unconfigured agent.
Until the agent-side publisher lands, that is every agent's answer:
the publisher runs inside the container and an agent's NATS grants are
hive-scoped, which cannot authorise a write to a single-token agent
key. The read side needs no grant change — the controller already
holds `$KV.*.>` and `$JS.API.DIRECT.GET.*.>`.
The response carries `Content-Security-Policy: sandbox` and `nosniff`:
the body is an operator-authored SVG served from this daemon's own
origin, and an SVG can carry script.
Hive-side icon serving is untouched.
Refs #4502
