Watch

atlas 0f7b403803 agents get a todo when assigned to an issue or PR, rewritten when unassigned and cleared on close

A new swarm service, swarm-notify-router, in its own container on the host
that sets deploy.notify-router.enable (defaulted on by allSwarmServices where
the host has the router's store leaf, like the claude login minter).

- swarm-controller registers instance-wide hooks for `issues` +
`issue_assign` and `pull_request` + `pull_request_assign`, updates the
event list of a hook already registered with fewer (leaving its active flag as it is), and keeps each
verified body unchanged in the SWARM_EVENTS stream on
$SWARM.events.forge.<X-Forgejo-Event-Type>, deduped by X-Forgejo-Delivery.
A delivery the stream did not keep answers 503.
- swarm-notify-router drains its durable consumer on that stream, runs its
flows (source -> match -> todo sink, stored as nodes plus edges and checked
to be a tree) and queues each op on $SWARM.router.todo.<agent> in the
work-queue stream ROUTER_TODOS. Built-in flows: assigned, unassigned (same
key, rewritten) and closed (prefix-clears every holder's router todos). A
per-issue snapshot in its notify-router-state bucket holds the last
assignees and the todo holders; only agents in the agent-status bucket get
ops, and an assignee that was not one is left out of the snapshot so the
next assign event retries it. After the first event that must be retried,
the rest of the batch is nakked so redelivery keeps stream order.
- hive-agent creates its own durable consumer todo-<agent> and applies each
op to its todo store as subsystem `router`, then acks; a refused op stops
the rest of its batch the same way. The store's (subsystem, key) dedup
makes a redelivery harmless. New Todos::clear_prefix.
- swarm-nats-auth: a --notify-router-client arm; the controller may publish
$SWARM.events.forge.*; each agent gets three {agent} templates (extended
consumer create whose filter the server checks, MSG.NEXT, ACK).
- Credentials: a notify-router store leaf minted by glue-bao-tls, a
swarm-notify-router bao policy (read on its own queue client secret), an
authelia machine client, the secret publisher entry, a journal forwarder
route. No manual step.

hive-forge-notify is unchanged, so agents also still get forge's own
notifications.

Refs #5571

2026-10-22 07:28:53 +02:00

12 KiB

Raw Blame History

hyperhive — claude entry point

Hey claude. This is your starting page. The detailed docs live in docs/ and are written for humans + you both — read them when you need depth on a subsystem. This file is the index.

High-level project intro: README.md.
Open work + backlog: the forge issue tracker at $HIVE_FORGE_URL/hyperhive/hyperhive/issues. Read the host out of the env var — inside an agent container localhost is the agent, not the forge, so a hardcoded loopback address fails to connect.
Operator/agent trust-boundary design: docs/trust-boundary/boundary.md (area/ops issues for the deployment/gateway/privsep work — the separator is a slash, and --label area:ops now fails with did you mean "area/ops"?).
Agent trust model (trust boundary, prompt-injection threat model, capability = accepted risk), credential isolation + sandbox threat model: docs/trust-boundary/security.md.
Odd-looking nspawn / container plumbing usually has a story: docs/process/gotchas.md.

Build & check

Always through the devshell — a bare cargo fails (failed to find tool "cc", cannot find -lsqlite3). Details: docs/process/conventions.md.

nix develop -c cargo clippy --all-targets -- -D warnings
nix develop -c cargo test
nix fmt        # treefmt is the CI gate, NOT bare cargo fmt

Changed a hivectl / swarmctl verb or flag? Regenerate its doc — only CI's nix flake check catches the drift: nix develop -c cargo run --bin hivectl -- markdown-docs > docs/tools/hivectl-cli.md (same for swarmctl → docs/tools/swarmctl-cli.md). CI red but local green → build the suspect check (e.g. nix build .#checks.x86_64-linux.hivectl-docs) before blaming the builder.

Repo map

One line per crate / top-level dir. Each module's authoritative, always-current description lives in its own //! doc-comment — grep/read the module when you need detail. This index is kept deliberately lean: it auto-loads into every turn's context, and a hand-maintained per-file tree drifts out of sync with the code.

Rust workspace (`Cargo.toml` members)

Per-hive (host + in-container):

hive-c0re/ — host daemon (runs as the unprivileged hive-core user), daemon-only — the operator CLI is hivectl. Owns the sqlite broker, reminder queue, meta flake, lifecycle (nixos-container shellouts), gateway wiring, per-container stats and the axum operator dashboard. Agent forge/matrix accounts are swarm-controller's. Largest crate.
hivectl/ — operator CLI; talks to hive-c0re over the host admin socket, does NOT link it. Verb reference (CI-enforced): docs/tools/hivectl-cli.md.
hive-priv/ — minimal root helper, socket-activated at /run/hive/priv.sock; the few root ops (bind-mount edits, nsenter) hive-c0re delegates. See docs/trust-boundary/boundary.md.
hive-agent/ — the in-container serve loop: turn-loop policy (turn.rs) over hive-runtime, per-agent web UI, event + turn-stats sinks, system-prompt renderer, claude launch config (mcp_config.rs). With hive-agent-mcp it's what "hive-ag3nt" refers to — there is no such directory.
hive-agent-mcp/ — the embedded MCP server (hive-mcp-http unit).
hive-runtime/ — the Runtime interface an agent's turns run on: a claude backend (over hive-claude) and an acp backend whose stream is translated into claude's stream-json shape. Links no hyperhive binary crate, so the subagent daemon uses it too.
hive-bash-mcp/, hive-matrix-mcp/, hive-subagent-mcp/ — per-agent MCP daemons serving streamable-http directly (no stdio bridge): background bash tasks, matrix-sdk tools, nested subagent sessions (optionally goal-driven over several turns; daemon state is in-memory, the claude session survives a restart).
hive-screen-mcp/ — stdio MCP bridge for GUI agents: screenshot, keyboard, mouse. All userspace.
hive-forge/ — hive-forge Forgejo CLI; one module per verb.
hive-forge-notify/ — notification pollers → todos on the in-agent socket. Two binaries: hive-forge-notify, hive-github-notify.
hive-jobq/ — the job-DAG scheduler library. Runtime-only — nothing writes the graph to disk, so node ids are stable within a run, not across restarts. hive-c0re's job_queue/ is being removed in its favour — new scheduler-shaped code goes here.
hive-jobq-wire/ — presentation of a hive-jobq graph to viewers (WireNode / WireResource); kept out of hive-jobq on purpose.
hive-jobq-metrics/ — OTEL export of a graph's state rollup on a timer.
hive-log/ — the one log-sink chooser: journald or stdout, never both (under a unit, stdout already is the journal). Do not grow it into a utility crate.
hive-metric/ — one-shot CLI pushing a labeled metric over OTLP.

Wire types + socket plumbing:

hive-types/ — leaf newtypes, chiefly the validated Ident.
hive-sh4re/ — shared payload types (Message, HelperEvent, …).
hive-host-sock/ (hivectl ↔ c0re), hive-priv-sock/ (c0re ↔ priv), hive-core-agent-sock/ (harness ↔ c0re, /run/hive/mcp.sock) — per-socket wire types, split out so each client links only its protocol.
hive-agent-sock/ — the in-agent socket the harness serves to in-container producers (todos, reminders); also holds the extra_mcp spec. ⚠️ This socket never leaves the container — hive-c0re is not in the path, unlike hive-core-agent-sock.
hive-sock-client/ — the shared JSON-line unix-socket client; retry is a policy value (Retry::None / Retry::RideOutRestart). Also holds the bounded request-line reader (read_request_line) the socket servers use.

Swarm-level:

swarm-controller/ — owns what is true across hives; opt-in, one per swarm. Serves HTTP over a unix socket nginx proxies to — ⚠️ the socket's directory is its access control; the crate README has the constraint and a unit test pins the path.
swarm-controller-api/ — serde-only request and response bodies of the controller's HTTP API, shared with swarmctl (forge, schedules, AGENTS.md edits, agent create and forge-token, inference, limits and permissions).
swarmctl/ — swarm operator CLI; runs as root and acts directly (README says why). Doesn't link swarm-controller: every body it sends or reads comes from swarm-controller-api. ⚠️ Edits authelia's users.yml in place, shared with swarm-authelia-bridge; that crate's README says what both writers must uphold. Verb reference (CI-enforced): docs/tools/swarmctl-cli.md.
swarm-authelia-bridge/ (+ swarm-authelia-bridge-sock/ wire types) — the other writer of authelia's users DB; runs inside swarm-authelia.
swarm-secret-client/ — OpenBao client over vaultrs. ⚠️ Reads the BAO_ env spellings explicitly — vaultrs's VAULT_* defaults are set by no unit here, and falling through to them yields a client with no identity and a TLS failure that names no cause.
swarm-queue-client/ — the one authenticated connect to the swarm NATS queue.
swarm-wire/ — serde-only request and reply contracts services exchange over the queue (builder, claude login, review classifier, agent inference, login, permissions and terminal streams), so naming one links neither the queue client nor the store client.
swarm-nats-auth/ — NATS auth-callout responder; without it the queue is fail-closed and admits nobody.
swarm-agent-verify/ — checks a swarm-agent.<agent>.<secret> token against the store; shared by swarm-nats-auth and swarm-logs-auth.
swarm-logs-auth/ — nginx auth_request responder for agent access to the log store, and to the metrics store as swarm-metrics-agent-auth.
swarm-metrics-auth/ — nginx auth_request responder for Grafana's bearer on the metrics store's query API.
swarm-logs/ — agent CLI: a LogsQL query in, log lines out.
swarm-metrics/ — agent CLI: a MetricsQL query in, the store's JSON out.
swarm-matrix-client/ — client-server API calls against the swarm homeserver, shared by swarm-controller, swarm-matrix-ctl, hive-c0re and hive-matrix-mcp.
swarm-matrix-ctl/ — the one binary run inside containers.hive-matrix; new in-container jobs are verbs here, not crates.
swarm-builder/ — builds agent config PR heads for commit status, and evaluates the agent archetypes the create form lists. Design: docs/swarm/builder.md.
swarm-ci-metrics/ — polls the forge's Actions runs and pushes each finished run's duration and queue wait to the metrics store. Design: docs/swarm/ci-metrics.md.
swarm-claude-login/ — runs claude auth login for one agent and stores the credentials at swarm/agents/<agent>/claude-credentials. Design: docs/swarm/claude-login.md.
swarm-notify-router/ — forge events from swarm-controller's webhooks to todos on the agents they concern, through each agent's own consumer on the ROUTER_TODOS stream. Design: docs/swarm/notify-router.md.
swarm-review-classifier/ — one claude --print per PR classification, no tools, no forge credential. Design: docs/swarm/review-classifier.md.

External dependencies with no directory here

hive-claude — app-agnostic driver for headless claude --print (stream-json, Telemetry, the self-compacting InfiniteSession). ⚠️ Its own repo, not a workspace member — there is no hive-claude/ directory and grep here will not find it.

Other top-level dirs

frontend/ — npm workspaces built hermetically by nix/packages/frontend.nix: shared, dashboard (operator SPA), agent (per-container UI), swarm-ui (Preact; packaged separately by nix/packages/swarm-ui.nix, not in packages.default's closure).
nix/ — host-modules/ (the host stack), agent-modules/ (per-agent harness features), container-modules/ (guest side of service containers), templates/agent.nix (the container entry point), archetypes/ (the built-in agent archetypes), packages/, docs/ (options doc), module-eval/ (NixOS module-eval tests), builder/ (check-flake), script-tests/, plus sources.nix / rust.nix / checks.nix / devshell.nix / treefmt.nix behind the thin flake.nix.
docs/ — subsystem reference docs (see Reading paths below).
branding/, scripts/ — static assets + helper scripts.

Reading paths

docs/README.md is the single index, grouped by topic. Don't add a second copy of it here — a parallel list used to live in this file and drifted.

Conventions & process

Never add #[allow(clippy::…)] — fix the lint instead (extract a helper, add backticks, etc.). Worked examples: docs/process/conventions.md.
Pre-push lint hook (catches tracker-tag and comment-block failures before CI does — install once per clone): ln -sf ../../scripts/pre-push .git/hooks/pre-push
Comment/PR-description length: size it to how non-obvious the thing is, not to how much investigation it took to find it — a mechanical fix gets a one-line comment, a genuinely non-obvious invariant or decision earns real prose. Don't restate a module's //! doc in a comment further down the same file.
Operator-facing doc with real implementation detail mixed in: wrap the implementation part in a <details><summary>…</summary> block (blank line after <summary> so the markdown inside renders) rather than a sibling page — see docs/integrations/github.md's "Implementation" section. It only collapses in a rendered browser, and the <summary> line isn't markdown-processed, so keep it plain prose with no backticks.

12 KiB Raw Blame History