Surfaces the per-container on-disk footprint the c0re sampler now
reports as `disk_bytes` on /api/container-resources (state dir +
container writable rootfs, shared nix store excluded; sampled
out-of-band every few minutes). Renders bytes→human via the existing
cloadFmtBytes helper, with an em-dash until the first sample lands
(disk_bytes is null then). Pairs with the hive-c0re sampler half.
Replace the gateway's bare in-container self-signed leaf with a leaf
signed by a host-held hive CA. A bare self-signed leaf is its own trust
anchor, so every regeneration is a fresh anchor every consumer would have
to re-trust, and a runtime-generated in-container leaf cannot be wired
into an agent's build-time trust store at all. A stable CA fixes both: a
single anchor that agents and federation peers trust once, surviving leaf
rotation.
New hive-tls module: a host oneshot generates a long-lived CA (default
~20y) under services.hyperhive.tls.stateDir and signs a gateway leaf
(default ~10y, SAN covering the bare domain, forge., matrix. and the
wildcard). It is ordered before the gateway container so the leaf exists
when nginx starts. Active only when the gateway uses self-signed TLS
(default) and a domain is set; inert under operator-cert or ACME modes.
Gateway: bind-mount the host CA dir read-only at /run/hive-ca; the
existing in-container cert unit now imports the host leaf into nginx's
state dir (copy as root, key left root:nginx 0640 for the pre-start
config test) instead of generating one. Cert/key paths nginx serves are
unchanged.
Foundational step toward agent + federation trust of self-signed hives;
no behaviour change for agents yet (they still reach the forge over plain
http on port 80). Eval-proven across self-signed, certDir and the inert
default paths.
Replace the per-row save buttons on the P3RM1SS10NS tab with one
page-level "save all (N agents)" button covering both the capabilities
and tool-groups matrices.
Toggling cells stages edits in the DOM; each checkbox carries its
render-time baseline (data-baseline). On save we diff against the
baseline and POST only the perm-types that actually changed per agent
to the new combined endpoint:
POST /api/permissions { changes: [ { agent, tool_groups?, capabilities? } ] }
An omitted field leaves that perm-type untouched (no commit, no diff);
an included array fully replaces it. The backend coalesces caps + groups
for one agent into a single rebuild — no more double-rebuild when an
operator changes both for the same agent. The batch is atomic: on a
clean 200 the bar shows queued -> rebuilding and re-fetches (resetting
baselines); on any validation error nothing is applied and an error note
is shown.
Live capabilities_changed / tool_groups_changed re-renders are skipped
while the section has unsaved edits so a half-finished edit set is not
clobbered; the tab-activation and post-save re-fetches are the recovery
paths.
Pairs with the hive-c0re combined-PermPayload half (damocles).
Fixes#1724. The forgejo ROOT_URL was hardcoded `http://` even when the
gateway terminates TLS, producing broken clone links and mixed-content
redirects for a TLS deployment.
Derive the scheme + port from gateway state: `https` (on httpsPort) when
the gateway terminates TLS (`gateway.selfSignedTls` or `gateway.tls.certDir`
set), `http` (on gateway.port) otherwise — dropping the port suffix on the
canonical port for the scheme (80 / 443). The per-agent in-cluster
HIVE_FORGE_URL is unaffected (it stays http on the gateway's :80 path);
this only changes the public ROOT_URL forgejo advertises.
Refreshes the `rootUrl` option description (the TLS scheme is auto-derived
now, so the manual-override-for-TLS note is gone).
Eval-proven: gateway.selfSignedTls = true → ROOT_URL =
https://forge.<domain>/; false → http://forge.<domain>/.
Per mara's review of #1718: the HIVE_FORGE_URL / HIVE_MATRIX_URL comments
carried "network isolation is now unconditional / shared-netns removed"
backstory that doesn't belong on those keys. Trimmed to describe just the
value (the gateway vhost) + the gatewayHost null-guard. The
isolation-removal narrative stays where it's on-topic — the network
module's own deprecated-option descriptions and warning text.
Per mara's review of #1718: this env block only exists when hyperhive is
enabled (it lives inside `config = mkIf cfg.enable`), so the
`if hyperhive.enable then gateway else loopback` branch was dead — the
loopback fallback could never be reached. Set HIVE_FORGE_URL to the
gateway vhost directly.
HIVE_MATRIX_URL drops the same redundant `hyperhive.enable` term but
keeps the gatewayHost null-guard — that one is a real fallback for a
domain-less config, not the dead enable branch.
Addresses argus's review of #1718: the HIVE_FORGE_URL / HIVE_MATRIX_URL
conditionals still keyed off the now-deprecated `network.enable`, so
setting it to `false` would point agents at host loopback (unreachable
from their private netns) even though isolation stays on — contradicting
the "deprecated and ignored" warning. Guard on the top-level
`hyperhive.enable` instead, so `network.enable` truly has no effect
anywhere and the loopback branch only covers a hyperhive-disabled host.
Eval-proven: with `network.enable = false`, HIVE_FORGE_URL now resolves
to the gateway vhost (`http://forge.<domain>`), and the deprecation
warning still fires.
Network isolation is validated working end to end (forge + matrix reach
the gateway under private netns), so the shared-host-netns mode is
removed — there is now one mode.
- hive-network.nix: the bridge/DNS block and the isolation overlay are
now unconditional whenever hyperhive is enabled (gated on
services.hyperhive.enable, not the per-feature toggles). The redundant
isolateContainers-implies-enable guard is dropped.
- network.enable and isolateContainers are kept as DEPRECATED no-op
options so existing configs that set them to true still evaluate
cleanly; setting either to false now warns (gated on hyperhive being
enabled, so a non-hyperhive host gets no spurious warning) and has no
effect. Both are slated for removal in a future release.
- hive-c0re.nix: the HIVE_FORGE_URL / HIVE_MATRIX_URL conditionals drop
the "&& isolateContainers" clause — agents always reach forge/matrix
through the gateway vhost now.
- hive-matrix.nix: refresh one stale comment.
The Rust lifecycle is env-driven (HIVE_NETWORK_ISOLATION), so no Rust
change is needed: the env is now always set, so the existing
private-netns path always runs.
Supersedes the default-on flip (the option it would have defaulted is
gone).
BE-1 (the list endpoint) lives at GET /api/matrix-accounts (reads go under
/api/, matching /api/state and /api/operator-inbox); the page was calling
the un-prefixed /matrix-accounts. Correct the fetch + the contract comment +
the docs reference. POST /matrix-account-login (mutation, root path like
/approve) is unchanged.
Per-account token-failure isolation is now live on the matrix daemon, so a
bad credential entered via this form can no longer take down the agent's
whole matrix session — remove the experimental notice (+ its now-unused CSS).
Also relabel the account list 'configured accounts' -> 'provisioned accounts'
to match the v1 read path (lists accounts with a stored token; a
config-declared-but-unprovisioned account appears once provisioned through
the form). Docs + module header updated to match.
New standalone H0M3 page (/matrix-accounts.html) and tile. An agent
picker drives a list of that agent's configured matrix accounts (name,
homeserver, token-stored status) and a provision form that logs in by
password or stores an existing token.
Frontend half of the per-agent external matrix-account provisioning
work. Built against the v1 backend contract:
GET /matrix-accounts?agent=<name>
-> { accounts: [ { name, homeserver, token_present } ] }
POST /matrix-account-login (x-www-form-urlencoded, operator-auth)
fields: agent, account, homeserver, mode=password|token,
user_id?, password?, token?
-> 2xx { ok, user_id } | 4xx { error }
The token is never echoed back; secret inputs are cleared on submit.
Token-status dot reflects token-stored, not live session (a true
up/down indicator needs the daemon account registry, a follow-up). The
form carries an experimental notice pending per-account failure
isolation on the matrix daemon.
Blocked from merge on the backend endpoints and the daemon
failure-isolation fix; opening for review + to pin the UI/backend wire
contract.
Cosmetic follow-up to the Y3R C4LL extraction (argus nit on #1705): the
approvals + questions sections kept their original 2-space IIFE indentation
from tabs.js, which read as if they were inside a block and was inconsistent
with the unindented operator-inbox section + the other domain modules.
Uniform dedent — no logic change, build unaffected.
Final domain: move the operator question queue (questionRowCache, the
sync/apply/filter/fingerprint/buildQuestionLi/renderQuestions fns, formatTtl,
and the .q-ttl countdown ticker) from tabs.js into call.js. tabs.js imports
syncQuestionsFromSnapshot, applyQuestionAdded, applyQuestionResolved,
renderQuestions, and the activeQuestionCount getter; refreshTabCounts now
sums activeApprovalCount()+activeQuestionCount()+operatorInboxCount().
The .status-age (SW4RM) and .approval-ts/.reminder-due/.sched-due
(approvals+schedules) tickers stay in tabs.js — only the questions-specific
.q-ttl ticker moved. questionsState stays in state.js (shared with the SW4RM
badges). Behaviour-preserving; no visual change.
Net: tabs.js 2323 to 1603 lines; the Y3R C4LL pane now lives in call.js,
alongside permissions.js / schedules.js as a per-domain module.
Begin splitting the tabs.js monolith: lift the Y3R C4LL domain into a new
call.js module (mirrors permissions.js / schedules.js).
- questionsState + QUESTION_HISTORY_LIMIT move to state.js: they're read by
both the SW4RM container rows (per-agent question-count badges) and the
Y3R C4LL questions domain, so they're cross-domain shared state.
- operator-inbox domain (state + refresh/render/mark/append) → call.js.
- approvals domain (state + sync/apply/render/diff-panel/history) → call.js.
- call.js exposes count getters (activeApprovalCount, operatorInboxCount);
the entry's refreshTabCounts pulls them. Live-mutation paths call an
injected onCountsChanged callback (registered via initCall at boot)
instead of reaching into the coordinator (avoids a circular import).
- renderPeerHives, physically interleaved in the moved range but part of
the SW4RM/peers domain, stays in tabs.js.
Behaviour-preserving; no visual change. Questions domain follows next.
(Recreated after a harness-restart git-object corruption; identical content.)
The verb that lists issues/PRs with filters is `list`, but `issues` and
`issue-list` are the names people reach for first — and clap's
"did you mean" tip points at `issue` / `issue-create`, never at `list`.
Two agents independently failed to find it that way.
Add `issues` and `issue-list` as visible clap aliases so the verb is
discoverable under the names people actually type (and they show in
`--help`). No behaviour change; `list` stays canonical.
The forwarded host env vars (HIVE_FORGE_URL, HIVE_FORGE_PUBLIC_URL,
HYPERHIVE_HIVE_DOMAIN, ...) were emitted only into the harness
service's own unit environment. Under network isolation that left the
bash-task runner, the matrix daemon, tea-login and interactive shells
without HIVE_FORGE_URL, so `hive-forge` and friends fell back to the
loopback default (localhost:3000) — unreachable from a private netns,
where the in-cluster gateway URL (forge.<domain>) is the only path.
Emit the forwarded set into systemd.globalEnvironment too, which every
unit + shell in the container inherits, so all in-container surfaces
see the same forge/matrix endpoints. The existing harness-service
forward is kept (harmless redundancy). Shared netns is unaffected (the
localhost default still works there).
Adds a regression test asserting the forwarded vars land inside the
globalEnvironment block, and the generated flake parses.
Under network isolation, agents lost DNS resolution of the hive's own
zones (forge.<domain>, matrix.<domain>): the hyperhive-isolated-dns
oneshot points resolv.conf at the bridge resolver, but resolvconf
regenerated it from host-tracking right after (dhcpcd re-triggering that
when the veth comes up), reverting to the non-authoritative host
resolver. Agents then couldn't reach the in-hive forge or matrix through
the gateway, which broke an isolate-by-default rollout.
Take resolvconf and dhcpcd out of the resolv.conf loop: disable
resolvconf and tell dhcpcd not to touch resolv.conf (without disabling
dhcpcd itself, so the veth still gets its address). Then the last writer
owns it -- the nixos-container host-copy in shared netns, or the oneshot
in isolated mode. Mirrors the approach the matrix container already
takes. Eval-checked; flake check passes.
