hyperhive/nix/templates/harness-base.nix
iris b714ba15c2 harness: drop trailing /bin from systemd path entries so wrapper sudo resolves (#672 fixup)
systemd.services.<name>.path appends /bin to each entry, so the
literal '/run/wrappers/bin' here was being expanded to
'/run/wrappers/bin/bin' inside the unit's PATH — a path that
doesn't exist. 'which sudo' then fell back to
'/run/current-system/sw/bin/sudo' (the non-setuid nix-store binary)
and refused with 'must be owned by uid 0 and have the setuid bit
set' on every agent, despite hyperhive.user.passwordlessSudo = true.

Verified on this container post-rebuild:
  PATH includes /run/wrappers/bin/bin (non-existent)
  /run/wrappers/bin/sudo exists with mode r-s--x--x (real setuid)
  but `sudo` resolves to /run/current-system/sw/bin/sudo and fails.

Fix: drop the trailing /bin from both entries. systemd appends it.

The /run/current-system/sw entry was already correctly
expanding to /run/current-system/sw/bin (because of the same
auto-append), which is why everything else on PATH worked despite
the broken wrappers entry — only sudo (the one binary that needs
the wrapper dir) was affected.
2026-05-31 11:29:18 +02:00

1340 lines
59 KiB
Nix

{
pkgs,
lib,
config,
# Flake inputs routed through _module.args by the agent flake.nix.
# Default to {} so the module evaluates cleanly even when the agent
# flake doesn't set up the routing pattern (e.g. during standalone
# nixos-rebuild without a flake wrapper).
flakeInputs ? { },
...
}:
let
# Agent user metadata (#658). `userName` defaults to `"agent"` when
# the meta-flake doesn't inject the per-agent override (stand-alone
# `nixos-rebuild` against `nixosConfigurations.agent-base` works
# without erroring on a missing per-agent name). `homeDir` derives
# from `userName` to keep them coupled.
userName = config.hyperhive.user.name;
homeDir = "/home/${userName}";
in
{
# Shared scaffolding for any hyperhive harness container — both
# sub-agents (`agent-base.nix`) and the manager (`manager.nix`) extend
# this. The systemd service that actually runs the harness binary
# differs per role and lives in the child module.
# Optional feature modules. Each declares its own `hyperhive.*`
# option(s), default-off, so every agent has them available but
# only opts in from its own `agent.nix`.
imports = [ ./weston-vnc.nix ];
# Per-agent unix user the harness + co-process daemons run as (#658).
# Defaults to `"agent"` so a standalone evaluation (e.g.
# `nix flake check` against `nixosConfigurations.agent-base`) builds
# cleanly; the meta-flake's per-agent module rebinds this to the
# agent name (`"damocles"`, `"iris"`, …) so each container has a
# uniquely-named user matching its agent label. UID auto-assigned
# by NixOS (the auto-allocation range for normal users); no hard-
# coded UID.
options.hyperhive.user.name = lib.mkOption {
type = lib.types.strMatching "^[a-z_][a-z0-9_-]{0,30}$";
default = "agent";
example = "iris";
description = ''
Unix user the harness service runs as inside the container.
The meta-flake overrides this to the agent's own name so the
user inside the container matches the agent label (`HIVE_LABEL`).
Stand-alone evaluation defaults to `"agent"` so module evaluation
without the meta-flake wrapper still builds.
Constraints match `useradd`'s NAME_REGEX: lowercase / `_` start,
total length 31, no special characters. UID is auto-assigned
by NixOS; no `uid =` override surface (intentional pinning
across rebuilds isn't a concern when the home and state dirs
stay bind-mounted from the host).
'';
};
options.hyperhive.user.passwordlessSudo = lib.mkOption {
type = lib.types.bool;
default = true;
example = false;
description = ''
Grant `${config.hyperhive.user.name}` passwordless sudo
(`NOPASSWD: ALL`). True by default so claude's `Bash` tool
keeps working for tools that expect root inside the container
(`systemctl`, package managers in dev shells, etc.) the
same surface the previous root-user shape had, just elevated
explicitly instead of implicitly.
Flip to `false` for agents that should be strictly
unprivileged. Anything claude shells out to that needs root
will then fail loudly with the standard sudo error rather
than silently succeeding easier to spot the leak.
'';
};
options.hyperhive.role = lib.mkOption {
type = lib.types.enum [
"agent"
"manager"
];
default = "agent";
example = "manager";
description = ''
Whether this container runs as a sub-agent (`"agent"`, the
default invokes `hive-ag3nt serve`) or as the swarm's
manager (`"manager"` invokes `hive-m1nd serve` and
defaults the forge notification surface to mentions-only).
meta.rs flips this to `"manager"` for the manager container
and leaves it at the default for every sub-agent. Agents
and `agent.nix` files don't normally touch this option;
it's exposed so a standalone `nixos-rebuild` against
`nixosConfigurations.manager` keeps working without the
meta-flake wrapper around it.
Closes #671: harness + manager templates merged into a
single `harness-base.nix` driven by this option.
'';
};
options.hyperhive.model = lib.mkOption {
type = lib.types.str;
default = "haiku";
example = "sonnet";
description = ''
Claude model for this agent. Sets the `HIVE_DEFAULT_MODEL`
environment variable; the harness applies it at boot and it takes
priority over any persisted runtime override. The operator can still
switch the model at runtime via the per-agent web UI that choice
is tracked in the state dir for the current session but is reset by
any rebuild that changes this option.
Valid values are the short model names that `claude --model` accepts:
`"haiku"`, `"sonnet"`, `"opus"` (or any future identifier). Context
window sizes are looked up at runtime from the
`HIVE_CONTEXT_WINDOW_TOKENS_<KEY_UPPER>` env vars injected by the
meta flake; override sizes via `services.hyperhive.c0re.contextWindowTokens`
on the host.
'';
};
options.hyperhive.allowedBashPatterns = lib.mkOption {
type = lib.types.listOf lib.types.str;
default = [ ];
example = [
"git *"
"ls *"
"cat /agents/*/state/*"
];
description = ''
Shell command patterns auto-approved for the `Bash` built-in tool.
Empty list (the default) grants wholesale `Bash` approval
claude can run any shell command without a prompt. Non-empty list
replaces `Bash` in `--allowedTools` with one `Bash(pattern)` entry
per item; only commands matching a pattern are auto-approved; all
others require confirmation (which in `--print` mode means they
will not run). Use to sandbox agents to a known-safe command
vocabulary.
Patterns use the same glob syntax claude accepts in `Bash()`:
`*` matches any string within a word, shell-style.
'';
};
options.hyperhive.allowedRecipients = lib.mkOption {
type = lib.types.listOf lib.types.str;
default = [ ];
example = [
"alice"
"manager"
];
description = ''
Names this agent is allowed to `send` to via
`mcp__hyperhive__send`. Empty list (the default) means
unrestricted the agent can message any peer, the
operator, or the manager. Non-empty list constrains the
surface: only the listed names + the manager (always
allowed) get through; anything else returns an error
string to claude without touching the broker. The
operator (`operator`) needs to be in the list if the
agent should be able to surface output on the
dashboard.
Useful for sandboxing untrusted sub-agents set
`[ "manager" ]` to scope them to manager-only chatter.
The manager itself is always exempt; this option only
affects sub-agent `send`.
'';
};
options.hyperhive.extraMcpServers = lib.mkOption {
type = lib.types.attrsOf (
lib.types.submodule {
options = {
command = lib.mkOption {
type = lib.types.str;
description = "Absolute path to the MCP server binary. Use `\${pkgs.foo}/bin/foo` or `/run/current-system/sw/bin/foo`.";
};
args = lib.mkOption {
type = lib.types.listOf lib.types.str;
default = [ ];
description = "Args passed to the MCP server binary.";
};
env = lib.mkOption {
type = lib.types.attrsOf lib.types.str;
default = { };
description = "Environment variables for the MCP server child process.";
};
allowedTools = lib.mkOption {
type = lib.types.listOf lib.types.str;
default = [ "*" ];
example = [
"send_message"
"join_room"
];
description = ''
Tool names this MCP server is auto-approved to call via
`--allowedTools`. Single entry `"*"` (the default) means
"every tool from this server" convenient but trusting.
Tighten to a specific list when you only want a subset.
Names are bare (e.g. `send_message`); the harness prepends
`mcp__<server-key>__` at build time.
'';
};
};
}
);
default = { };
example = lib.literalExpression ''
{
matrix = {
command = "/run/current-system/sw/bin/mcp-matrix";
args = [ "--config" "/state/matrix.toml" ];
env.MATRIX_HOMESERVER = "https://matrix.example.org";
allowedTools = [ "send_message" "join_room" ];
};
}
'';
description = ''
Extra MCP servers claude sees alongside the hyperhive tool surface.
Keys are the server names (claude addresses tools as
`mcp__<key>__<tool>`). Rendered to `/etc/hyperhive/extra-mcp.json`
at activation time; the harness reads that file at boot and merges
it into `--mcp-config` + `--allowedTools`. Take effect on the
agent's next harness restart (no operator approval needed beyond
whatever brought the new agent.nix into deployed/*).
'';
};
options.hyperhive.matrix.enable = lib.mkOption {
type = lib.types.bool;
default = true;
description = ''
Enable per-agent matrix integration via `hive-matrix-mcp`
(#548 phase 3). When true (the default), the harness:
- runs `hive-matrix-daemon` as a systemd unit that holds a
matrix-sdk Client + sync against the homeserver at
`HIVE_MATRIX_URL` (default `http://localhost:8008` the
in-host tuwunel from `nix/modules/hive-matrix.nix`). The
daemon auto-skips when `<state>/matrix-token` is missing,
and a `systemd.paths` watcher restarts it the moment
hive-c0re provisions the token (mirrors `matrix-avatar-sync`
shape from #571).
- exposes the matrix tool surface (send_message, send_dm,
send_reaction, send_reply, mark_read, list_rooms,
list_room_members, read_room) to claude via an auto-injected
`extraMcpServers.matrix` entry. Claude spawns the stdio
`hive-matrix-mcp` bridge per turn, which forwards each tool
call to the daemon over `/run/hive-matrix/socket`.
- wakes the agent on incoming room events via a short teaser
Wake signal (`[matrix] <sender> in <room>: <first 100c>`)
to the hyperhive control socket; the full event stays
unread server-side until `read_room` consumes it.
Set to `false` for agents that should NOT have matrix tools at
all (e.g. agents on a host without `hyperhive.matrix.enable` on
the meta side). When token file is absent the daemon and MCP
both no-op cleanly anyway, so `false` is rarely necessary.
'';
};
options.hyperhive.matrix.url = lib.mkOption {
type = lib.types.str;
default = "http://localhost:8008";
example = "https://matrix.darkest.space";
description = ''
Matrix homeserver URL the agent's `hive-matrix-daemon` connects
to. Default points at the in-host tuwunel (shared netns).
Override per-agent when an agent should talk to an external
homeserver instead (e.g. a federation-only setup or a remote
hive's tuwunel reached via a vpn).
'';
};
options.hyperhive.frontend.dist = lib.mkOption {
type = lib.types.package;
default = pkgs.hyperhive-frontend;
defaultText = lib.literalExpression "pkgs.hyperhive-frontend";
description = ''
The shipped frontend dist (built by `nix/frontend.nix`). Output
layout: `dashboard/` (used by hive-c0re on the host) and
`agent/` (used here, layered with `extraFiles` below at
activation time). Override to ship a fully custom per-agent SPA;
the JSON contract (`/api/state`, `/events/stream`, the action
endpoints) is the source of truth for any replacement.
'';
};
options.hyperhive.frontend.mergedDist = lib.mkOption {
type = lib.types.package;
readOnly = true;
description = ''
Computed: the merged static tree consumed by the harness via
`HIVE_STATIC_DIR`. Composed at evaluation time by copying
`hyperhive.frontend.dist`'s `agent/` subdir as the base, then
layering each `extraFiles` entry on top. Read-only
consumers (`agent-base.nix`, `manager.nix`) reference this in
their systemd service environment; do not set directly.
'';
};
options.hyperhive.frontend.extraFiles = lib.mkOption {
type = lib.types.attrsOf (
lib.types.submodule (
{ name, ... }:
{
options = {
source = lib.mkOption {
type = lib.types.path;
description = ''
Source file or directory to layer over the default
agent dist. A path (relative to `agent.nix` or
absolute) nix copies its contents into the merged
static tree.
'';
};
target = lib.mkOption {
# First char must be alphanumeric/underscore (rules out
# leading `/`, leading `.`, leading `-`); inner chars
# include `.` and `/` so nested layouts like
# `"games/bitburner"` work. This is the shape check —
# the `..`-segment traversal check is the assertion in
# `config.assertions` below (regex alone can't reject
# mid-path `..` segments without lookahead, which nix
# POSIX regex doesn't support).
type = lib.types.strMatching "^[A-Za-z0-9_][A-Za-z0-9_./-]*$";
default = name;
defaultText = lib.literalMD "the attribute name";
description = ''
Destination path within the merged static tree, used
as both the served URL prefix (`/<target>/...`) and
the on-disk layout in the merged derivation. Defaults
to the attribute name. Use forward slashes for
nested layouts (e.g. `"games/bitburner"`).
Constrained shape: must start with an alphanumeric or
`_`, and only contain alphanumerics, `_`, `.`, `/`,
`-`. `..` segments are separately rejected at config
eval time.
'';
};
};
}
)
);
default = { };
example = lib.literalExpression ''
{
bitburner = {
source = ./bitburner-dist;
# served at GET /bitburner/...
};
}
'';
description = ''
Per-agent additions layered on top of the default frontend
dist. Each entry copies its `source` into the served static
tree under `target`. Useful for shipping a self-contained
agent-specific surface alongside the standard agent UI (e.g.
the bitburner agent's game page at `/bitburner/`).
The default agent UI remains served at `/`; entries here only
add new routes and never replace the default. Overwrite
semantics are **hard-fail**: if `target` collides with an
existing file or directory in the default dist (or with a
prior entry's target), the `mergedDist` build aborts with
`refusing to overwrite existing path '<target>' in the
default dist`. To override a default file, fork the dist via
`hyperhive.frontend.dist` instead `extraFiles` is for
pure additions.
`target` must be a relative path inside the static dir. An
assertion rejects leading `/` and `..` segments at config
eval time (string-concat-into-paths safety, even though
agent.nix goes through operator review before deploy).
'';
};
options.hyperhive.forge.url = lib.mkOption {
type = lib.types.str;
default = "http://localhost:3000";
example = "http://forge.internal:3000";
description = ''
Base URL of the hyperhive-managed Forgejo. Used at container
boot by a oneshot systemd unit that calls
`tea login add --url <this> --token "$(cat $HYPERHIVE_STATE_DIR/forge-token)"`
(= `/agents/<name>/state/forge-token`) so the agent's claude can
shell out to `tea` without an extra auth dance. No-op when the
forge-token file is missing (i.e. hive-forge isn't running on
the host).
'';
};
options.hyperhive.forge.keepSubscriptions = lib.mkOption {
type = lib.types.bool;
default = true;
description = ''
When true (the default), the forge notification poller will NOT
auto-unsubscribe from repo watches after delivering a
"subscribed"-reason notification. Sub-agents keep their broad
subscriptions so they stay informed about repos they contribute to.
Set to false for agents (e.g. the manager) that use reason-based
filtering and do not need firehose-level repo visibility they will
auto-unsubscribe after receiving a watched-repo notification.
'';
};
options.hyperhive.forge.skipNotifyReasons = lib.mkOption {
type = lib.types.listOf lib.types.str;
default = [ ];
example = [
"subscribed"
"participating"
];
description = ''
Forgejo notification `reason` values to suppress in the forge
notification poller. Notifications with these reasons are marked
read and silently dropped; all others including notifications
with a null or unrecognised reason are delivered.
Drop-list is safer than an allow-list: directed signals
(`review_requested`, `assigned`, `mention`) are never silently
missed even if Forgejo returns an unexpected reason string.
Empty list (the default) delivers all notifications. Set to
`[ "subscribed" "participating" ]` for agents like the manager
that want only direct mentions and reviews, not the full repo
firehose. Rendered to the `HIVE_FORGE_NOTIFY_SKIP_REASONS`
environment variable consumed by the harness poller at runtime.
'';
};
options.hyperhive.dashboardLinks = lib.mkOption {
type = lib.types.listOf (
lib.types.submodule {
options = {
label = lib.mkOption {
type = lib.types.str;
description = "Display label for the link.";
};
icon = lib.mkOption {
type = lib.types.str;
default = "";
description = "Optional icon emoji or short glyph.";
};
url = lib.mkOption {
type = lib.types.str;
description = "Full URL (may include a different port, e.g. http://localhost:9001/stats).";
};
};
}
);
default = [ ];
example = lib.literalExpression ''
[
{ label = "Stats"; icon = "📊"; url = "http://localhost:9001/stats"; }
]
'';
description = ''
Extra navigation links surfaced on the hive-c0re dashboard card for
this agent. Declare any additional web UI pages the agent exposes
stats pages, custom UIs, etc. hive-c0re reads the JSON file this
option produces at each container-view snapshot and attaches the
links to the agent card without any code changes.
'';
};
options.hyperhive.claudeMarketplaces = lib.mkOption {
type = lib.types.listOf lib.types.str;
default = [ "anthropics/claude-plugins-official" ];
example = [
"anthropics/claude-plugins-official"
"anthropics/claude-plugins-community"
];
description = ''
Claude Code plugin marketplaces to add at harness boot. Each
entry is passed to `claude plugin marketplace add <source>`
(`owner/repo`, full git URL, or local path). Idempotent
re-adding an existing marketplace is treated as success.
Required before `hyperhive.claudePlugins` entries that
reference a marketplace (e.g. `foo@claude-plugins-official`).
Rendered to `/etc/hyperhive/claude-marketplaces.json`.
Defaults to Anthropic's official marketplace; agents get it
out of the box without any per-agent.nix wiring.
'';
};
options.hyperhive.claudePlugins = lib.mkOption {
type = lib.types.listOf lib.types.str;
default = [ ];
example = [
"formatter@my-marketplace"
"thinking-tools@anthropics"
];
description = ''
Claude Code plugins to install at harness boot. Each entry is
passed verbatim to `claude plugin install <spec>` once per
container start, before the turn loop opens. `claude plugin
install` is expected to be idempotent, so reinstalling on every
boot is cheap. Failures log a warning but do not abort boot a
missing plugin is preferable to a non-serving agent. Rendered to
`/etc/hyperhive/claude-plugins.json`; the harness reads it via
`plugins::install_configured`.
'';
};
options.hyperhive.claudePluginsAutoUpdate = lib.mkOption {
type = lib.types.bool;
default = false;
description = ''
When true, the harness runs `claude plugin marketplace update`
before installing plugins at boot, pulling the latest index from
all configured marketplaces. Disabled by default most agents
want pinned plugin versions and the network round-trip adds to
boot time. Enable for agents that should always install the latest
available version of their plugins.
'';
};
options.hyperhive.icon = lib.mkOption {
type = lib.types.nullOr lib.types.path;
default = null;
example = lib.literalExpression "./icon.svg";
description = ''
Path to an SVG file used as this agent's icon shown on the
dashboard and the per-agent web UI (header + favicon). Commit
the SVG into the agent's config repo next to `agent.nix` and
reference it as a relative path (`./icon.svg`).
When null (the default) the agent falls back to the shared
hyperhive logo. The harness serves the icon (configured or
default) at `GET /icon` on the per-agent web port.
'';
};
options.hyperhive.autoCompact = lib.mkOption {
type = lib.types.bool;
default = true;
description = ''
Enable proactive watermark-based compaction. When `true` (the
default) the harness automatically runs a notes-checkpoint turn
followed by `/compact` once the context window crosses 75% of
the model's limit, keeping later turns from hitting the hard
overflow path. Set to `false` to disable proactive compaction
entirely (`HIVE_COMPACT_WATERMARK_TOKENS=0`); the reactive path
(compact-on-overflow when the session is already past the limit)
still applies.
Disable for agents that run large-context models (sonnet/opus)
where the heuristic fires too early and discards useful history
before the session is actually close to the limit.
'';
};
config = {
assertions = [
# Guard the inputs-routed-as-output pattern: the agent flake.nix is
# expected to set `_module.args.flakeInputs = builtins.removeAttrs inputs ["self"]`.
# If `self` leaks into flakeInputs the agent gets a spurious attrset
# entry that can shadow real inputs and is almost certainly a bug.
# Guard with `or {}` so standalone evaluation stays clean when
# flakeInputs is absent from _module.args.
{
assertion = !(builtins.hasAttr "self" (config._module.args.flakeInputs or { }));
message = ''
hyperhive: `flakeInputs` must not contain "self".
In your agent flake.nix, use:
_module.args.flakeInputs = builtins.removeAttrs inputs [ "self" ];
'';
}
# hyperhive.model must be a non-empty string — an empty value causes
# the harness to pass an invalid model flag to claude.
{
assertion = config.hyperhive.model != "";
message = "hyperhive.model must not be empty (set it to e.g. \"haiku\" or \"sonnet\")";
}
# hyperhive.forge.url must look like an HTTP URL when non-default.
{
assertion =
config.hyperhive.forge.url == ""
|| lib.hasPrefix "http://" config.hyperhive.forge.url
|| lib.hasPrefix "https://" config.hyperhive.forge.url;
message = "hyperhive.forge.url must be an http:// or https:// URL (got: \"${config.hyperhive.forge.url}\")";
}
# hyperhive.icon must reference an SVG file when set.
{
assertion = config.hyperhive.icon == null || lib.hasSuffix ".svg" (toString config.hyperhive.icon);
message = "hyperhive.icon must point to an .svg file";
}
# hyperhive.frontend.extraFiles[*].target is concatenated into
# $out during the mergedDist build. The option's strMatching
# type already rejects leading `/`, leading `.`, and the
# weirder characters; this assertion catches mid-path `..`
# segments (e.g. `foo/../etc/passwd`) that the type's regex
# can't easily express without lookahead. agent.nix is
# operator-reviewed, so this is belt-and-braces — but it's the
# kind of mistake that's easy to make and hard to spot.
{
assertion = lib.all (entry: !(builtins.any (seg: seg == "..") (lib.splitString "/" entry.target))) (
lib.attrValues config.hyperhive.frontend.extraFiles
);
message = ''
hyperhive.frontend.extraFiles: `target` must not contain
`..` path segments.
'';
}
];
# Per-agent unix user (#658). Runs the hive-ag3nt / hive-m1nd
# harness + co-process daemons (hive-matrix-daemon) under a
# non-root principal. The user name follows
# `hyperhive.user.name` — defaults to `"agent"` for standalone
# eval, overridden per-agent by the meta-flake to the agent's
# own label so each container has a uniquely-named user.
#
# UID auto-assigned by NixOS (per mara's #8109: "no hardcoded
# uids"). Home is `/home/${userName}`. `wheel` membership +
# the sudoers rule below grants `NOPASSWD: ALL` when
# `passwordlessSudo` is true — same blast radius as the
# previous root-by-default shape, just explicit.
users.users.${userName} = {
isNormalUser = true;
home = homeDir;
createHome = true;
group = userName;
extraGroups = lib.optional config.hyperhive.user.passwordlessSudo "wheel";
# Matches /bin/bash on NixOS — the harness's claude shell-outs
# expect a POSIX shell at $SHELL; bashInteractive is already
# the system default for the root user too (see SHELL env
# var declaration below).
shell = pkgs.bashInteractive;
};
users.groups.${userName} = { };
# `NOPASSWD: ALL` for the agent user. Lets claude's Bash tool
# keep working with anything that expected root (systemctl,
# nix-env, etc.) without prompting — same surface as the
# previous root-by-default shape, just elevated explicitly.
# Flip `hyperhive.user.passwordlessSudo = false` to drop both
# the wheel-group membership and this sudoers entry; anything
# that needs root then fails loudly instead of silently
# succeeding.
security.sudo.extraRules = lib.mkIf config.hyperhive.user.passwordlessSudo [
{
users = [ userName ];
commands = [
{
command = "ALL";
options = [ "NOPASSWD" ];
}
];
}
];
# First-boot migration from the legacy root-run shape (#658).
# Runs on every activation; marker-guarded so the move only
# happens once. The bind mount that hive-c0re sets up has
# already moved from `/root/.claude` to `${homeDir}/.claude`
# by the time we get here (per `lifecycle::CONTAINER_CLAUDE_MOUNT`
# — the host-side path stays the same, the container-side
# mount target shifts), so the bulk of the data is already at
# the new location. This script just:
#
# - ensures `${homeDir}` exists with correct ownership (covers
# the very first boot before useradd's `createHome` has
# anything to chown);
# - migrates any leftover `/root/.claude` content that an
# operator might have populated before #658 deployed (the
# bind mount didn't exist in that lifecycle, so claude
# would have written into the root user's empty home —
# nothing important typically, but safer to move than to
# strand);
# - chowns the bind-mounted state dir (`/agents/*/state`) so
# the agent user can read/write it.
system.activationScripts.hive-agent-user-migrate = lib.stringAfter [ "users" "specialfs" ] ''
homeDir=${lib.escapeShellArg homeDir}
userName=${lib.escapeShellArg userName}
# Always ensure the home dir exists with the right ownership
# useradd's createHome handles the very first creation but
# doesn't re-chown if a rebuild changes the user name (rare
# but possible if the meta-flake's per-agent name evolves).
mkdir -p "$homeDir"
chown "$userName:$userName" "$homeDir"
# One-time migration of pre-#658 /root/.claude content into the
# new home. Marker-guarded so the move only runs once per
# container lifetime subsequent activations skip the legacy
# path even if claude were to repopulate /root/.claude for any
# reason.
marker=/var/lib/hive-agent-user-migrated
if [ ! -e "$marker" ] && [ -d /root/.claude ] && [ "$(ls -A /root/.claude 2>/dev/null)" ]; then
mkdir -p "$homeDir/.claude"
# `mv -n` (no-clobber) so any pre-existing files at the
# destination (e.g. from the bind mount) win we never
# blow over data already at the new location.
if cp -an /root/.claude/. "$homeDir/.claude/" 2>/dev/null; then
rm -rf /root/.claude
echo "hive-agent-user-migrate: moved /root/.claude $homeDir/.claude"
fi
fi
mkdir -p "$(dirname "$marker")"
: > "$marker"
# Chown the bind-mounted state dir so the agent user can
# read/write it. `/agents/*/state` is the canonical mount
# point set by hive-c0re's `set_nspawn_flags`. Wildcard
# because each container only sees its own
# `/agents/<name>/state` (one match); -h to avoid following
# any symlinks the agent might have planted in there.
for stateDir in /agents/*/state; do
[ -d "$stateDir" ] || continue
chown -hR "$userName:$userName" "$stateDir" 2>/dev/null || true
done
# Same treatment for the bind-mounted `~/.claude/` dir. Pre-#658
# the harness ran as root and `claude` wrote `.credentials.json`
# there 0600 root:root; post-#658 the harness reads
# `~/.claude/` as the agent user to decide Online vs
# NeedsLogin (`login::has_session`), and the host-side bind
# source is still root-owned 0700 from those legacy writes.
# Chown recursively so the existing credentials are readable
# under the new identity instead of getting silently treated
# as "no session" and re-prompting login every boot.
if [ -d "$homeDir/.claude" ]; then
chown -hR "$userName:$userName" "$homeDir/.claude" 2>/dev/null || true
fi
'';
# Auto-inject the matrix MCP entry when matrix is enabled (#548
# phase 3). Operator can override or disable by setting their own
# `extraMcpServers.matrix` (nix submodule merge takes the operator's
# value) or by flipping `hyperhive.matrix.enable = false`.
hyperhive.extraMcpServers = lib.mkIf config.hyperhive.matrix.enable {
matrix = lib.mkDefault {
command = "${pkgs.hyperhive}/bin/hive-matrix-mcp";
args = [ ];
# Same socket path the hive-matrix-daemon service binds
# via its `RuntimeDirectory = "hive-matrix"` (#658). Keeps
# the bridge + daemon in sync without baking the new path
# into the Rust default — the env override wins for both.
env.HIVE_MATRIX_SOCKET = "/run/hive-matrix/socket";
allowedTools = [ "*" ];
};
};
environment.etc."hyperhive/extra-mcp.json".text = builtins.toJSON config.hyperhive.extraMcpServers;
# Operator-set per-agent icon (hyperhive.icon). When configured, the
# SVG lands at /etc/hyperhive/icon.svg; the harness serves it at
# GET /icon, falling back to the bundled hyperhive logo when absent.
environment.etc."hyperhive/icon.svg" = lib.mkIf (config.hyperhive.icon != null) {
source = config.hyperhive.icon;
};
environment.etc."hyperhive/bash-allow.json".text =
builtins.toJSON config.hyperhive.allowedBashPatterns;
environment.etc."hyperhive/send-allow.json".text =
builtins.toJSON config.hyperhive.allowedRecipients;
environment.etc."hyperhive/claude-plugins.json".text =
builtins.toJSON config.hyperhive.claudePlugins;
environment.etc."hyperhive/claude-marketplaces.json".text =
builtins.toJSON config.hyperhive.claudeMarketplaces;
environment.etc."hyperhive/claude-plugins-auto-update.json".text =
builtins.toJSON config.hyperhive.claudePluginsAutoUpdate;
# Merged frontend static tree. Base = `${frontend.dist}/agent/`,
# then each `extraFiles` entry is laid on top at its `target`
# path. The runCommand derivation aborts on overwrite so a
# filename collision with the default dist surfaces as a build
# failure rather than a silent override (operator gets a clear
# nix error rather than a confusing 404 / silent dist swap).
hyperhive.frontend.mergedDist = pkgs.runCommand "hyperhive-agent-frontend-merged" { } (
''
mkdir -p $out
cp -r ${config.hyperhive.frontend.dist}/agent/. $out/
chmod -R u+w $out
''
+ lib.concatMapStrings (entry: ''
mkdir -p $(dirname $out/${entry.target})
if [ -e $out/${entry.target} ]; then
echo "hyperhive.frontend.extraFiles: refusing to overwrite existing path '${entry.target}' in the default dist" >&2
exit 1
fi
cp -r ${entry.source} $out/${entry.target}
'') (lib.attrValues config.hyperhive.frontend.extraFiles)
);
# HIVE_DEFAULT_MODEL seeds the initial model selection when no persisted
# model choice exists in the state dir. SHELL must be set so claude's
# Bash tool finds a POSIX shell.
# HIVE_ASSETS_DIR points at the project's static runtime assets
# (branding + claude prompts; see `nix/assets.nix`). Set here so
# both the harness binary and any user-shell `cargo run` inside the
# container resolve them from the same path.
# HIVE_CONTEXT_WINDOW_TOKENS_* are injected by the meta flake from the
# host-level `services.hyperhive.c0re.contextWindowTokens` option — not set here.
environment.variables = {
HIVE_DEFAULT_MODEL = config.hyperhive.model;
HIVE_ASSETS_DIR = "${pkgs.hyperhive-assets}/share/hyperhive";
SHELL = "${pkgs.bashInteractive}/bin/bash";
}
// lib.optionalAttrs (!config.hyperhive.autoCompact) {
# Zero watermark disables proactive compaction; the reactive path
# (compact-on-overflow) still fires when the session is truly full.
HIVE_COMPACT_WATERMARK_TOKENS = "0";
}
// lib.optionalAttrs config.hyperhive.forge.keepSubscriptions {
HIVE_FORGE_KEEP_SUBSCRIPTIONS = "1";
}
// lib.optionalAttrs (config.hyperhive.forge.skipNotifyReasons != [ ]) {
HIVE_FORGE_NOTIFY_SKIP_REASONS = lib.concatStringsSep "," config.hyperhive.forge.skipNotifyReasons;
};
boot.isNspawnContainer = true;
# Every agent gets flakes + the modern `nix` CLI out of the box.
# Equivalent to passing `--extra-experimental-features 'nix-command
# flakes'` on every invocation. Agents shell out to `nix build` /
# `nix flake` constantly (devshells, ad-hoc evals, fetching their
# own MCP-server flakes); without this they hit the "experimental
# feature not enabled" wall on the first try.
nix.settings.experimental-features = [
"nix-command"
"flakes"
];
# Containers bind-mount the host's nix-daemon socket. The host daemon
# may be configured with remote builders or strict sandbox settings
# (sandbox-fallback = false) that make local `nix build` invocations
# fail inside the container. Enable sandbox-fallback so builds that
# can't set up the sandbox (no user-namespaces in nspawn) fall back
# to unsandboxed local builds rather than failing outright.
# mkForce overrides the nixpkgs nix module which sets this to false
# at normal priority -- without it agents get a conflicting definition
# error on rebuild. Security implications: see docs/security.md.
nix.settings.sandbox-fallback = lib.mkForce true;
# `claude-code` is unfree. Each per-agent container's nixosConfiguration
# evaluates its own `nixpkgs` instance, so the operator's host-level
# `nixpkgs.config.allowUnfreePredicate` does not propagate into here —
# we have to allow it inside the container's config as well.
nixpkgs.config.allowUnfreePredicate = pkg: builtins.elem (pkgs.lib.getName pkg) [ "claude-code" ];
environment.systemPackages = with pkgs; [
hyperhive
claude-code
bashInteractive
coreutils-full
# procps for pkill — used by the web UI's /api/cancel to SIGINT the
# in-flight claude turn.
procps
# tea: gitea/forgejo CLI client. Configured at boot by the
# tea-login oneshot below if /state/forge-token is present, so
# claude can `tea repos create`, `tea pulls create`, etc.
tea
# jq: JSON processing in shell — useful for parsing API responses,
# forge REST calls, sqlite output, etc.
jq
# curl: HTTP client for forge REST API and other web requests.
curl
# hive-forge <verb>: CLI wrapping common Forgejo REST API operations
# (view, pr, issue, comment, assign, close, labels, branches, etc.)
(pkgs.callPackage ../packages/hive-forge-tools.nix { })
];
# One-shot: write tea's config.yml from the seeded forge token so
# the agent can use `tea` without interactive prompts. Runs on
# every boot so a rotated token (hive-c0re remints on each agent
# rebuild) is always reflected. *Always* exits 0 — never fail a
# NixOS switch-to-configuration over a missing/temperamental forge.
systemd.services.tea-login = {
description = "configure tea CLI from hive-forge token (best-effort)";
wantedBy = [ "multi-user.target" ];
after = [ "local-fs.target" ];
serviceConfig = {
Type = "oneshot";
RemainAfterExit = true;
};
path = [
pkgs.curl
pkgs.python3
pkgs.coreutils
];
environment.HOME_DIR = homeDir;
environment.AGENT_USER = userName;
script = ''
# No `set -e`: any subshell failure must not propagate.
# A failed unit aborts `nixos-container update` which blocks rebuilds.
FORGE_URL=${lib.escapeShellArg config.hyperhive.forge.url}
# $HYPERHIVE_STATE_DIR is set system-wide by the meta flake
# (systemd.globalEnvironment, /agents/<name>/state per agent
# including manager post-#604).
TOKEN_FILE="$HYPERHIVE_STATE_DIR/forge-token"
if [ ! -f "$TOKEN_FILE" ]; then
echo "tea-login: no forge-token at $TOKEN_FILE; skipping"
exit 0
fi
TOKEN=$(cat "$TOKEN_FILE")
# Resolve the agent username from the forge API.
USER=$(curl -sf --max-time 5 \
-H "Authorization: token $TOKEN" \
"$FORGE_URL/api/v1/user" \
| python3 -c 'import sys,json; print(json.load(sys.stdin).get("login",""))' \
2>/dev/null || true)
if [ -z "$USER" ]; then
echo "tea-login: could not resolve username from forge API; skipping"
exit 0
fi
# tea reads config from ~/.config/tea/config.yml. The
# agent user's home is $HOME_DIR (set by NixOS via
# hyperhive.user.name). Write the config under that
# home + chown to the agent user so tea reads it when
# invoked as that user. Still runs as root (this
# service stays root-owned to avoid bootstrap
# ordering issues see comment near serviceConfig
# below), but the artefact it produces is for the
# agent user.
CONFIG="$HOME_DIR/.config/tea/config.yml"
mkdir -p "$(dirname "$CONFIG")" || true
cat > "$CONFIG" << EOF
logins:
- name: forge
url: $FORGE_URL
token: $TOKEN
default: true
ssh_host: ""
ssh_key: ""
insecure: false
ssh_agent: false
user: $USER
preferences:
editor: false
flag_defaults:
remote: ""
EOF
chown -R "$AGENT_USER:$AGENT_USER" "$HOME_DIR/.config" 2>/dev/null || true
echo "tea-login: configured for $FORGE_URL as $USER (config at $CONFIG)"
'';
};
# One-shot: upload the agent's configured icon to its Forgejo user avatar
# so the icon shows up on commits / PRs / issue comments in the forge.
# Only runs when `/etc/hyperhive/icon.svg` is present (set via
# `hyperhive.icon`). No-op when the forge is unreachable or the icon
# is not set. *Always* exits 0.
systemd.services.forge-avatar-sync = {
description = "sync agent icon to Forgejo user avatar (best-effort)";
wantedBy = [ "multi-user.target" ];
after = [ "tea-login.service" ];
serviceConfig = {
Type = "oneshot";
RemainAfterExit = true;
};
path = [
pkgs.curl
pkgs.coreutils
pkgs.jq
pkgs.librsvg
];
script = ''
ICON=/etc/hyperhive/icon.svg
if [ ! -f "$ICON" ]; then
echo "forge-avatar-sync: no icon configured; skipping"
exit 0
fi
FORGE_URL=${lib.escapeShellArg config.hyperhive.forge.url}
# $HYPERHIVE_STATE_DIR is set system-wide by the meta flake
# (systemd.globalEnvironment) to `/agents/<name>/state`.
TOKEN_FILE="$HYPERHIVE_STATE_DIR/forge-token"
if [ ! -f "$TOKEN_FILE" ]; then
echo "forge-avatar-sync: no forge-token found; skipping"
exit 0
fi
TOKEN=$(cat "$TOKEN_FILE")
# Rasterize SVG PNG (Forgejo's Go image library can't decode SVG).
PNG=$(mktemp --suffix=.png)
if ! rsvg-convert -f png -w 512 -h 512 "$ICON" -o "$PNG" 2>/dev/null; then
echo "forge-avatar-sync: rsvg-convert failed; skipping"
rm -f "$PNG"
exit 0
fi
IMAGE=$(base64 -w 0 < "$PNG")
rm -f "$PNG"
# Forgejo POST /user/avatar expects {"image":"<base64>"} just the
# raw base64 string, NOT a data URI (data:image/png;base64,...).
# Use jq to build the payload so the large base64 value is safely quoted.
PAYLOAD=$(jq -n --arg img "$IMAGE" '{image:$img}')
RESP=$(curl -sf --max-time 10 \
-X POST "$FORGE_URL/api/v1/user/avatar" \
-H "Authorization: token $TOKEN" \
-H "Content-Type: application/json" \
-d "$PAYLOAD" \
-w "\n%{http_code}" 2>/dev/null || true)
CODE=$(printf '%s' "$RESP" | tail -1)
if [ "$CODE" = "204" ] || [ "$CODE" = "200" ]; then
echo "forge-avatar-sync: avatar uploaded (HTTP $CODE)"
else
echo "forge-avatar-sync: upload returned HTTP $CODE skipping (non-fatal)"
fi
'';
};
# Long-running matrix-sdk Client + sync per agent (#548 phase 3).
# Holds the unix socket the stdio `hive-matrix-mcp` bridge talks
# to, and emits hyperhive wake signals on incoming room events
# via `/run/hive/mcp.sock`. Conditional on `hyperhive.matrix.enable`
# AND token-file presence (the daemon binary itself exits 0 on
# missing token, but the path watcher below restarts it the
# moment the token lands — same first-boot-ordering pattern as
# matrix-avatar-sync.path / #571).
systemd.services.hive-matrix-daemon = lib.mkIf config.hyperhive.matrix.enable {
description = "long-running matrix-sdk Client + MCP daemon socket";
wantedBy = [ "multi-user.target" ];
after = [ "network-online.target" ];
wants = [ "network-online.target" ];
environment = {
HIVE_MATRIX_URL = config.hyperhive.matrix.url;
# Socket path lives inside the systemd-managed runtime dir
# (`RuntimeDirectory = "hive-matrix"` → `/run/hive-matrix/`,
# owned by the agent user) so the daemon can bind it without
# needing root over `/run/` itself (#658). The stdio bridge
# picks up the same path via its own `HIVE_MATRIX_SOCKET` env
# in `extraMcpServers.matrix` below.
HIVE_MATRIX_SOCKET = "/run/hive-matrix/socket";
RUST_LOG = "info";
};
serviceConfig = {
ExecStart = "${pkgs.hyperhive}/bin/hive-matrix-daemon";
Restart = "on-failure";
RestartSec = 5;
# Run as the per-agent unix user (#658). The runtime dir
# (`/run/hive-matrix/`) is owned by that user via
# `RuntimeDirectory`; claude (also as that user) can
# connect to the socket inside it when the stdio bridge
# spawns per turn.
User = userName;
Group = userName;
RuntimeDirectory = "hive-matrix";
};
};
# Path-trigger sibling so hive-matrix-daemon fires the moment
# `<state>/matrix-token` appears (#548 phase 3, mirrors the
# matrix-avatar-sync.path pattern from #571). On clean boot
# hive-c0re provisions the token AFTER agent containers come up;
# without the trigger the daemon would exit 0 quietly and the
# MCP would have no backend until next restart. With the watcher
# the daemon comes alive in the same boot cycle as provisioning.
# The glob matches every agent (manager sees its own state at
# `/agents/hm1nd/state/` via the `/agents` bind).
systemd.paths.hive-matrix-daemon = lib.mkIf config.hyperhive.matrix.enable {
description = "trigger hive-matrix-daemon when matrix-token appears";
wantedBy = [ "multi-user.target" ];
pathConfig.PathExistsGlob = "/agents/*/state/matrix-token";
};
# Path-trigger sibling so matrix-avatar-sync fires the moment
# `<state>/matrix-token` appears (#571 closes argus's first-boot
# ordering nag on #567). On a clean boot hive-c0re's matrix
# provisioning runs concurrently with the agent container coming
# up; without this the oneshot would skip silently because the
# token didn't exist yet, and avatar would only land on the next
# restart. With the path watcher the appearance of the token
# triggers a re-fire of the service so the avatar is set in the
# same boot cycle as provisioning completes. The glob matches
# every agent (manager sees its own state at `/agents/hm1nd/state/`
# via the `/agents` bind).
systemd.paths.matrix-avatar-sync = {
description = "trigger matrix-avatar-sync when matrix-token appears";
wantedBy = [ "multi-user.target" ];
pathConfig.PathExistsGlob = "/agents/*/state/matrix-token";
};
# One-shot: upload the agent's configured icon to its matrix profile
# avatar so the icon shows up next to messages in matrix rooms (#548
# phase 2.5). Mirrors the forge-avatar-sync flow above, only differs
# in protocol: matrix avatars are a two-step `media upload` → `set
# avatar_url` dance, both authenticated by the access_token written
# by hive-c0re's `matrix::ensure_user_for`. No-op when the icon
# isn't configured, the matrix token isn't present, or the
# homeserver isn't reachable. *Always* exits 0.
#
# Triggered by EITHER boot (`wantedBy = multi-user.target`) OR
# the path-watcher above (`matrix-avatar-sync.path`) firing on
# token appearance (#571). Both paths re-run the oneshot
# idempotently — running the avatar set twice is harmless.
systemd.services.matrix-avatar-sync = {
description = "sync agent icon to matrix profile avatar (best-effort)";
wantedBy = [ "multi-user.target" ];
# No `after = [ "tea-login.service" ]` — matrix has no
# equivalent prerequisite; we just need the homeserver up.
serviceConfig = {
Type = "oneshot";
# NB: NOT `RemainAfterExit = true` — we want re-runs from
# the path trigger to actually re-execute. With
# RemainAfterExit, systemd treats the service as "still
# running" after the first exit and the second trigger
# becomes a no-op.
RemainAfterExit = false;
};
path = [
pkgs.curl
pkgs.coreutils
pkgs.jq
pkgs.librsvg
];
script = ''
ICON=/etc/hyperhive/icon.svg
if [ ! -f "$ICON" ]; then
echo "matrix-avatar-sync: no icon configured; skipping"
exit 0
fi
# Token written by `hive-c0re::matrix::ensure_user_for` to the
# agent's bind-mounted state dir. $HYPERHIVE_STATE_DIR is set
# system-wide by the meta flake (systemd.globalEnvironment) to
# `/agents/<name>/state`.
TOKEN_FILE="$HYPERHIVE_STATE_DIR/matrix-token"
if [ ! -f "$TOKEN_FILE" ]; then
echo "matrix-avatar-sync: no matrix-token at $TOKEN_FILE; skipping"
exit 0
fi
TOKEN=$(cat "$TOKEN_FILE")
# Local tuwunel reachable on shared host netns at the
# default matrix-spec port. Override via the future
# `hyperhive.matrix.url` if the operator ever runs the
# homeserver elsewhere (deferred to #548 phase 4).
MATRIX_URL=http://localhost:8008
# whoami user_id. Needed to scope the avatar set call.
# Tolerant of the homeserver being unreachable (`-f` makes
# curl fail on 4xx/5xx; `|| true` swallows the exit).
USER_ID=$(curl -sf --max-time 5 \
-H "Authorization: Bearer $TOKEN" \
"$MATRIX_URL/_matrix/client/v3/account/whoami" 2>/dev/null \
| jq -r '.user_id // empty' || true)
if [ -z "$USER_ID" ]; then
echo "matrix-avatar-sync: whoami failed or homeserver unreachable; skipping"
exit 0
fi
# Rasterize SVG PNG (matrix media accepts any image type
# but we already standardise on PNG for the forge sync).
PNG=$(mktemp --suffix=.png)
if ! rsvg-convert -f png -w 512 -h 512 "$ICON" -o "$PNG" 2>/dev/null; then
echo "matrix-avatar-sync: rsvg-convert failed; skipping"
rm -f "$PNG"
exit 0
fi
# Step 1: upload bytes mxc:// URI.
MXC=$(curl -sf --max-time 10 \
-X POST "$MATRIX_URL/_matrix/media/v3/upload" \
-H "Authorization: Bearer $TOKEN" \
-H "Content-Type: image/png" \
--data-binary "@$PNG" 2>/dev/null \
| jq -r '.content_uri // empty' || true)
rm -f "$PNG"
if [ -z "$MXC" ]; then
echo "matrix-avatar-sync: media upload failed; skipping"
exit 0
fi
# Step 2: set avatar_url on the profile.
PAYLOAD=$(jq -n --arg url "$MXC" '{avatar_url:$url}')
CODE=$(curl -s --max-time 10 \
-X PUT "$MATRIX_URL/_matrix/client/v3/profile/$USER_ID/avatar_url" \
-H "Authorization: Bearer $TOKEN" \
-H "Content-Type: application/json" \
-d "$PAYLOAD" \
-o /dev/null -w "%{http_code}" 2>/dev/null || true)
if [ "$CODE" = "200" ]; then
echo "matrix-avatar-sync: avatar set on $USER_ID"
else
echo "matrix-avatar-sync: avatar PUT returned HTTP $CODE skipping (non-fatal)"
fi
'';
};
# Write declared dashboardLinks to the state dir so hive-c0re can read
# them without accessing the container's /etc/ from the host.
# Runs every boot; idempotent (overwrite). Always exits 0.
systemd.services.hive-dashboard-links = lib.mkIf (config.hyperhive.dashboardLinks != [ ]) {
description = "write declarative dashboardLinks to agent state dir";
wantedBy = [ "multi-user.target" ];
serviceConfig = {
Type = "oneshot";
RemainAfterExit = true;
};
environment.LINKS_JSON = builtins.toJSON config.hyperhive.dashboardLinks;
script = ''
# Sub-agents have their state dir bind-mounted at /agents/<name>/state.
# Use a glob exactly one match per container at runtime.
STATE_DIR=$(echo /agents/*/state)
if [ ! -d "$STATE_DIR" ]; then
echo "hive-dashboard-links: no state dir found at /agents/*/state; skipping"
exit 0
fi
printf '%s' "$LINKS_JSON" > "$STATE_DIR/hyperhive-dashboard-links.json"
echo "hive-dashboard-links: wrote $(printf '%s' "$LINKS_JSON" | wc -c) bytes to $STATE_DIR/hyperhive-dashboard-links.json"
'';
};
# Git is needed by claude's Bash tool (for the agent <-> manager config
# request flow) and by hive-c0re's own setup_applied / setup_proposed.
# The per-agent `applied/<name>/flake.nix` overrides `user.name` and
# `user.email` with the agent's identity — values here are `mkDefault`
# so the per-agent override wins without needing `mkForce`.
programs.git = {
enable = true;
config = {
user = {
name = lib.mkDefault "hyperhive";
email = lib.mkDefault "hyperhive@local";
};
init.defaultBranch = lib.mkDefault "main";
};
};
# Manager-only forge defaults (#671): skip the
# subscription/participation firehose so the manager's inbox
# only carries direct mentions, reviews, and assignments. Sub-
# agents keep the noisier defaults (`keepSubscriptions = true`,
# `skipNotifyReasons = [ ]`). `mkDefault` so any agent that
# wants to invert it can.
hyperhive.forge = lib.mkIf (config.hyperhive.role == "manager") {
keepSubscriptions = lib.mkDefault false;
skipNotifyReasons = lib.mkDefault [
"subscribed"
"participating"
];
};
# Harness systemd unit. Role-driven so the same `harness-base.nix`
# covers both `nixosConfigurations.agent-base` (`hive-ag3nt serve`)
# and `nixosConfigurations.manager` (`hive-m1nd serve`) without a
# second template file (#671). Per-agent HIVE_PORT / HIVE_LABEL
# come from the meta-flake's generated `applied/<name>/flake.nix`;
# the manager has hardcoded fallbacks here so `nixosConfigurations.manager`
# still builds standalone.
systemd.services.${if config.hyperhive.role == "manager" then "hive-m1nd" else "hive-ag3nt"} =
let
isManager = config.hyperhive.role == "manager";
# Post-#598 there is exactly one harness binary (`hive`), and
# it picks its surface from `HIVE_ROLE` at startup. We still
# name the systemd unit `hive-ag3nt` / `hive-m1nd` so dashboard
# log queries + ExecStartPre paths + ancestor PR diffs keep
# working without a unit rename cascade.
binary = "hive";
in
{
description = "${binary}${lib.optionalString isManager " manager"} harness";
wantedBy = [ "multi-user.target" ];
after = [ "network.target" ];
# systemd units get a minimal PATH by default and don't inherit
# `environment.systemPackages`. Pointing at `/run/current-system/sw`
# gives the harness (and any tools claude shells out to via Bash)
# access to everything declared in `systemPackages` — including
# anything an agent adds to its own `agent.nix` — without having
# to touch the service definition.
#
# `/run/wrappers/bin` prepended so the `security.wrappers`
# setuid shims (notably `sudo`) resolve before the bare
# nix-store binaries in `/run/current-system/sw/bin`.
# Post-#658 the harness runs as the per-agent user — without
# the wrapper dir on PATH, `sudo` resolves to the un-setuid
# nix-store binary and refuses with "must be owned by uid 0
# and have the setuid bit set" even when
# `hyperhive.user.passwordlessSudo = true` is configured
# (#672 fixup pulled forward into this PR to avoid the
# regression argus flagged on #676).
#
# `systemd.services.<name>.path` appends `/bin` to each entry,
# so the bare prefixes here resolve to `/run/wrappers/bin` +
# `/run/current-system/sw/bin` inside the unit's PATH. Passing
# the trailing `/bin` ourselves (the natural-looking spelling)
# would yield `/run/wrappers/bin/bin` + `/run/current-system/sw/bin/bin`,
# neither of which exists — that's how #672 originally landed
# broken: every agent had a PATH pointing at non-existent dirs
# and `which sudo` kept falling back to the un-setuid binary.
path = [
"/run/wrappers"
"/run/current-system/sw"
];
environment = {
SHELL = "${pkgs.bashInteractive}/bin/bash";
# `HOME` defaults to `/` for systemd services without a User=
# set. With #658 the harness runs as the agent user — set HOME
# explicitly so claude (which the harness spawns) finds its
# `~/.claude/` session dir at the bind-mounted location.
HOME = homeDir;
# Path to the merged agent static dist. The harness serves this
# via `tower_http::ServeDir` for any request it doesn't route to
# an API endpoint. `mergedDist` is the agent-default dist with
# `hyperhive.frontend.extraFiles` layered on top.
HIVE_STATIC_DIR = "${config.hyperhive.frontend.mergedDist}";
# Static runtime assets (branding + claude prompts). Set on the
# unit directly — `environment.variables` only populates
# /etc/profile, which systemd services don't inherit.
HIVE_ASSETS_DIR = "${pkgs.hyperhive-assets}/share/hyperhive";
# Post-#598: the unified `hive` binary picks its surface from
# this env var at startup. Default (`"agent"`) matches the
# binary's standalone fallback when this is unset.
HIVE_ROLE = config.hyperhive.role;
}
// lib.optionalAttrs isManager {
# Standalone-eval fallbacks for `nixosConfigurations.manager`.
# meta.rs overrides both via the per-agent generated
# `applied/hm1nd/flake.nix` (see `lifecycle::setup_applied`);
# the values here keep the container sensible if anyone
# evaluates the standalone config.
HIVE_PORT = "8000";
HIVE_LABEL = "hm1nd";
};
serviceConfig = {
ExecStart = "${pkgs.hyperhive}/bin/${binary} serve";
Restart = "on-failure";
RestartSec = 2;
# `/run/hive-config/` is a per-service runtime dir owned by
# the agent user (`User=` below), auto-cleared by systemd on
# stop. The harness writes its regenerated
# claude-{mcp-config,settings,system-prompt} files there
# (see `paths::config_dir`). Kept separate from `/run/hive`
# — that bind comes in root-owned from the host and holds
# hive-c0re's `mcp.sock` we only connect to (#658 fixup).
RuntimeDirectory = "hive-config";
# Run the harness as the per-agent user (#658). claude itself
# spawned by the harness then runs as that user too — drops
# root inside the container while sudo (`NOPASSWD: ALL` by
# default, see `hyperhive.user.passwordlessSudo`) keeps the
# previous root-by-default surface available explicitly for
# tools that need it.
User = userName;
Group = userName;
};
};
system.stateVersion = "25.11";
};
}