The per-agent stats page gained a "first-turn ctx" summary chip (the
input tokens of the most recent fresh session's first turn — a
system-prompt/CLAUDE.md sprawl proxy), but agent.md's chip-row
description didn't mention it. Add it next to the reminder chips, noting
the stats.rs derive and the inert-until-capture behaviour (omitted from
JSON until the sessions / turn_stats.session_id capture has rows).
The C0R3 page PR moved the SYST3M tab off the dashboard onto the
standalone /core.html page, leaving the web-ui docs describing a tab that
no longer exists.
- dashboard.md: retitle the "SYST3M tab" section to "C0R3 page
(/core.html)" and reframe its intro — it's now a standalone page off the
H0M3 hub with a createTabStrip sub-tab nav (default R3BU1LD QU3U3), its
own core.js bundle on /dashboard/stream; the dashboard keeps only the
rebuild-queue state for the SW4RM "building…" badges.
- correct the chrome tab-strip list to the actual in-page tabs
(SW4RM / Y3R C4LL / P3RM1SS10NS / SCH3DUL3S) and note SYST3M / ST4TS /
S3TT1NGS / FL0W / L0GS / M4TR1X are all standalone pages off H0M3.
- fix the container-load poll reference (core.js, while the C0NT41N3R
L04D sub-tab is active) and the stray "SYST3M ›" panel references.
- drive-by: the schedules endpoint doc said "SYST3M scheduled-prompts
panel" — it's the SCH3DUL3S panel; corrected.
- web-ui.md index: same tab-list correction + C0R3 page pointer.
The "What the Rust side does" section predated the isolated-container
egress fix and omitted two now-implemented mechanisms:
- HOST_ADDRESS=<bridge-ip> in the nspawn conf — load-bearing for the
container's default route (nixos-container only installs `default via`
when HOST_ADDRESS is non-empty; bridge mode skips host-side route setup).
- the resolver wiring: hive-priv's bridge-DNS marker + the
hyperhive-isolated-dns oneshot rewriting resolv.conf to the bridge
dnsmasq (nixos-container re-copies the host's resolv.conf each start, so
the fix is runtime, ordered before the harness).
Doc-only; brings network.md in line with the merged behavior.
Follow-up to the isolated-container egress fix. The hyperhive-isolated-dns
oneshot (which rewrites resolv.conf to the bridge resolver) was ordered
before network-online.target + tea-login, but NOT relative to the harness
(hive-ag3nt) or the matrix daemon. hive-ag3nt only declares
`after network.target`, so on first boot its first-turn api.anthropic.com
lookup could race ahead of the resolv.conf rewrite and error (self-heals
next turn, but flaps the first one).
Add hive-ag3nt.service + hive-matrix-daemon.service to the oneshot's
`before` so DNS is fixed before any network consumer starts. The matrix
entry is a harmless no-op when matrix is disabled (unit absent).
Caught by damocles in review of the parent PR.
When isolateContainers=true, claude (and all egress) broke in every
container: agents came up with an IP but no way off the bridge subnet.
Two container-side gaps, both confirmed against nixpkgs
nixos-containers.nix:
1. No default route. hive-priv wrote HOST_ADDRESS= empty in the nspawn
conf. nixos-container's container-side setup only installs
`ip route add default via $HOST_ADDRESS` when HOST_ADDRESS is
non-empty, so the container had an address but no gateway -> nothing
off-subnet (incl. api.anthropic.com) was reachable. Fix: write
HOST_ADDRESS=<bridge-ip>. In bridge mode the host-side address/route
setup is skipped, so this only affects the container's default route.
2. No usable resolver. nixos-container copies the host's /etc/resolv.conf
into the container at every start; the host resolver (e.g. 127.0.0.53)
is unreachable from a private netns and isn't authoritative for the
hive's own zones. Fix: hive-priv drops a marker carrying the gateway
IP only when isolated, and a new harness-base oneshot
(hyperhive-isolated-dns) rewrites resolv.conf to point at the bridge
dnsmasq. Inert in shared-netns mode (no marker), so the shared
container toplevel does the right thing in both modes.
The gateway IP is the address part of HIVE_NETWORK_SUBNET (the bridge IP
verbatim, honouring a non-.1 operator override), via a new validated
bridge_gateway_ip() helper with unit tests.
Unblocks defaulting isolation on.
Builds the read/surface half of the per-session first-turn-tokens metric
(the capture — sessions table + turn_stats.session_id — landed separately).
A fresh claude session's first turn pays the full static prefix (system
prompt + tools + CLAUDE.md + first wake) as uncached input, so its
input_tokens is a clean proxy for prompt / CLAUDE.md sprawl — watching it
over time surfaces creep.
- stats.rs: add `Snapshot.first_turn_ctx: Option<u64>` populated by
`read_first_turn_ctx` — the agreed per-session derive (first turn,
`ORDER BY started_at LIMIT 1`, of the most recent session that started
in the window). Inert-until-capture: `.ok()` maps both "no fresh
session yet" and "older db without the sessions table" to None, the
same decoupling as read_bash_breakdown; the field is skipped from the
JSON when None. Pre-capture rows have a NULL session_id and are excluded.
- agent stats.js: add a "first-turn ctx" summary chip, guarded on a
numeric value so it stays hidden until capture has data.
clippy + cargo fmt clean; agent bundle builds.
The /logs.html header puts a flex:1 sub-tab strip (`.logs-tabbar`) next
to the ← home back-link inside the flex `.page-header`. The strip is
itself a flex container, so its default `min-width: auto` keeps it at its
intrinsic content width and — without an explicit nowrap on the parent —
it gets pushed onto its own row under the back-link.
Mirror the working dashboard chrome (`.dashboard-chrome` / `.tabbar`,
which sets `flex-wrap: nowrap`): make `.page-header` explicitly nowrap and
give `.logs-tabbar` `min-width: 0` so it shrinks to share the row instead
of wrapping. Affects every standalone page that uses `.page-header`
(flow / logs / stats / settings); the nowrap is correct for all of them.
Folds the SYST3M-tab removal into the same change that adds /core.html
(per operator review — it should be one PR, not an additive page + a
follow-up). Drops the tab + pane from dashboard.html and ~600 lines of
now-dead renderers/wiring from tabs.js: renderMetaInputs, renderRebuildQueue
+ renderQueueEntry + the rebuild-queue row cache / glyph maps / fingerprint,
renderTombstones, the container-load poll (cload* + start/stopContainerLoadPolling),
the two rqe-* elapsed tickers, the meta/tombstone state + sync + apply
handlers, their SSE dispatch entries, and the SYST3M tab-count.
Deliberately KEPT in tabs.js: rebuildQueueState + syncRebuildQueueFromSnapshot
+ applyRebuildQueueChanged + inFlightOpsByAgent — these don't render the
SYST3M panel, they drive the "building…" / "meta-updating…" badges on the
SW4RM agent cards, so the dashboard still subscribes to rebuild_queue_changed
to refresh those badges (applyRebuildQueueChanged now only re-renders
containers, no panel render).
core.css still imports dashboard.css for the shared section styles; splitting
those out + de-duping the renderers into a shared module remains a follow-up.
Carves the dashboard's SYST3M tab content out into a standalone page at
/core.html, reached from the H0M3 hub, with a createTabStrip sub-tab
strip (default = Rebuild Queue): rebuild queue, meta inputs, kept state,
container load. Same minimal-chrome standalone-page pattern as
/logs.html (← home back-link + sub-tab nav).
This is the additive first step: the new page is its own esbuild bundle
that cold-loads /api/state and subscribes to /dashboard/stream for the
same live rebuild_queue_changed / meta_inputs_changed / meta_update_running
/ tombstones_changed events the dashboard uses. The four section
renderers are ported from tabs.js; the dashboard's SYST3M tab is left in
place and untouched, so this PR cannot regress the existing dashboard.
Removing the now-duplicate SYST3M tab + de-duplicating the renderers
(into a shared module) + splitting the shared section CSS out of
dashboard.css is the deliberate follow-up.
core.css imports dashboard.css wholesale (transitional) so the ported
sections render identically; dashboard.css does not import common.css so
nothing double-loads. build.mjs gains core.js / core.css / core.html
entries; no Rust change (hive-c0re serves dist/ via ServeDir, and the
/dashboard/stream + /dashboard/history routes are registered ahead of
the fallback).
Page name "C0R3" is a placeholder pending the operator's pick — trivially
renamed (the /core.html URL + the C0R3 label).
Replace the unfiltered hyperhiveFlake = "${self}" (handed to hive-c0re
as the flake path it builds agent containers from) with a named,
explicitly-filtered source derivation that drops files no nix or cargo
derivation reads: docs/, scripts/, and root-level *.md. Editing those no
longer changes the hyperhiveFlake store path, so it no longer forces a
rebuild of every agent container.
This is the per-package source-filter pattern (operator request on
#1114): each real source is a named filtered derivation used as src,
not an inline filter at the use site. Companion to the existing rust
cleanSrc (crane cargo-source filter); supersedes the inline nixSrc
approach in the older #1114 branch.
Safe: hive-c0re consumes hyperhiveFlake as a runtime path string and
builds packages/nixosConfigurations from it (nix build <path>#…), which
never forces the checks that reference ./docs — and no package, module,
or container config reads docs/scripts/root-md.
Follow-up to the turn-time feature: the terminal now stamps a wall-clock
time on the turn-start / turn-end rows and an elapsed duration on
turn-end, sourced from a per-event `ts` (unix seconds) that the backend
serializes as a flattened sibling of the event tag on both the live SSE
frame and each history row.
- terminal-rendering.md: add the `.turn-time` row to the taxonomy table
+ a note on the ts source and the graceful degradation when a frame
omits ts.
- web-ui/agent.md: note the ts sibling on the emitted events, and the
time/duration suffixes on the TurnStart / TurnEnd renderings.
The per-agent terminal marks turn boundaries (◆ TURN ← / ✓ turn) but
shows no time. Append a wall-clock HH:MM:SS to both the turn-start and
turn-end rows, plus the elapsed duration on turn-end, rendered as dim
metadata so the boundary glyph stays the focus.
This is the frontend half of the feature; it reads a per-event `ts`
(unix seconds) off the turn_start / turn_end events. The read is guarded
on a numeric `ts`, so until the harness surfaces per-event timestamps
the rows render exactly as before (inert-until-capture). Once the
backend attaches `ts` to the history rows + live SSE frame, the times
light up automatically for both live and scrollback, with no further
frontend change.
H0M3 had tiles for every operator surface except the forge. Add a
Forge tile that links straight to the hive-forge web UI. It mirrors
the Matrix tile's gating: hidden by default, revealed by home.js only
when state.forge_present is true, with the href filled from
state.forge_public_url (the gateway-served vhost) or the direct :3000
port fallback — the same precedence the dashboard uses for forge links.
Operators without a forge never see a dead link.
Drop 'Phase 3 of the matrix-MCP work' (circular inside the matrix-mcp
crate) + the vague issue-thread pointer; keep the self-contained note
that it mirrors damocles-daemon's v0 set.
Clean the legacy backlog so the tracker-tag lint can become a required
gate (mara's warn-during-cleanup -> full-tree-enforcement path). Rewrite
the ~33 real `closes/see #NNN` provenance refs in doc-comments to prose
across hive-forge, hive-c0re, hive-ag3nt, hive-matrix-mcp, hive-sh4re,
and add a `lint:allow` line marker to check-issue-refs.sh for genuine
non-tracker `#<digits>` (a hash-digit heading-detection test input).
Tree is now lint-clean; tracker-tag lint ready to promote to required.
Adds a reasoning-effort quick-picker to the agent page's overflow menu,
a direct sibling of the existing model quick-picker, wired to the backend
seam from #1597/#1600: /api/state carries effort + available_efforts, the
picker POSTs { effort } to /api/effort (operator-only, mirroring
/api/model), and live updates arrive via the effort_changed SSE event.
The available levels are the backend's to declare — the frontend holds no
hard-coded list; availableEfforts is seeded from state.available_efforts
on cold-load and the picker section is omitted until the backend supplies
the set. Mirrors the model picker otherwise: postEffort(),
renderEffortChip() (updates the new amber effort-chip + button active
states), the cold-load seed, and the effort_changed handler. Build green.
Applies on the next claude session (operator uses the existing
clear-session), per the backend contract.
Match the harness reader landed in #1600 (configured_effort reads
HIVE_DEFAULT_EFFORT, mirroring HIVE_DEFAULT_MODEL). Was
HYPERHIVE_EFFORT_LEVEL in the first cut; damocles standardized the
name on the HIVE_DEFAULT_* convention.
Per-agent enum option (medium|high|xhigh, default medium) rendered into
the HYPERHIVE_EFFORT_LEVEL env var. The harness resolves effort as
operator-override-file -> this env -> medium and passes it to
claude --effort at turn launch (harness read+apply tracked separately).
Same declarative-default + runtime-override shape as hyperhive.model.
Umbrella: operator-controlled claude effort.
Per mara's review: instead of measuring the banner height in JS and
offsetting the chrome's sticky top, put the warning banner and the page
chrome in the same sticky div so they stack naturally.
common.js builds the wrapper: ensureStickyTop() wraps the page's chrome
(.dashboard-chrome / .page-header) in a single .sticky-top container and
injects the warning banner as its first child. The banner and the chrome
are no longer individually sticky — .sticky-top owns the stickiness, so
they pin together in one context instead of two top:0 stickies colliding
(the banner used to overlay the tab bar). Pages without a chrome (the
H0M3 hub) get a banner-only sticky region. No per-page markup needed; no
JS height measurement. Build green.
The schedules table builds a column per target, including agents that
have since been destroyed (their past schedules linger). Those columns
showed identically to live agents, so the operator couldn't tell a
schedule still targets something that no longer exists.
Flag them: renderSchedulesTableHead now checks each agent column against
the live roster (containersState) and, for any that isn't there (and
isn't 'operator', which is always valid), greys + strikes through the
column label and titles it 'no longer exists (gone)'. The rows stay
visible + cancellable — the operator can see the stale targets and clear
them rather than be surprised by phantom columns. Build green.
Largest step of the tabs.js split: the SCH3DUL3S tab — reminders +
scheduled-prompts (the lists, inline create row, per-row edit form, and
fire-now / cancel actions) — moves into a new schedules.js, ~1060 lines
out of tabs.js. It owns its own module state (schedules list +
edit-in-progress tracking) and reads the shared agent roster from
state.js for the target-chip pickers; render + format helpers come from
util.js, so the keystones (state.js, util.js) paid off here.
Exports the four entry-wired functions — applySchedulesChanged /
applyRemindersChanged (mutation dispatch), refreshSchedules /
refreshReminders (tab activation + cold-load) — plus activeScheduleCount
(the count pill). The 19 other functions stay module-private. Also drops
the now-unused paintAtomic import from tabs.js (its only callers were in
this block).
Behaviour-preserving code motion; esbuild inlines schedules.js into the
tabs.js bundle. Build green; tabs.js drops ~1060 lines (4323 → 2969).
Each home-hub tile now leads with an icon next to its label: 🖥 Dashboard,
📡 Flow, 📜 Logs, 📊 Stats, ⚙ Settings, 💬 Matrix. Matches the existing
emoji nav-icon convention (the per-agent link strip uses 📊/🖥/⬡/↳).
Markup: the label is wrapped with the icon in a .home-tile-head flex row
(icon + label on one line, description below). The icon is aria-hidden —
decorative, the label carries the accessible name. CSS-only layout
addition; no behaviour change. Build green.
Forgejo reported "does not have a signing key" on PRs despite the
forgejo-gpg-init service generating one. Two causes, both fixed:
- SIGNING_KEY = "default" resolves through the forgejo process's git
config (user.signingkey), not by scanning GNUPGHOME — and the keygen
never set it, so "default" found no key. forgejo-gpg-init now sets the
forgejo user's git user.signingkey to the generated key (+ commit and
tag gpgsign), with HOME pinned to the state dir so the global config
lands where forgejo reads it.
- The keygen guard was stamp-file based, so a partial state wipe that
lost the key while keeping the stamp would never regenerate. It's now
keyed on the actual secret key (regenerate iff gpg shows none) and
runs idempotently before forgejo on each start.
Also drops the heredoc for the gpg params in favour of printf (avoids
nix-string indentation fragility) and corrects the stale comment that
claimed "default" scans GNUPGHOME.
The tabs.js split's render-heavy domains (schedules, system, containers,
questions/approvals) all share a handful of pure helpers that lived in
the tabs.js IIFE: paintAtomic (atomic-swap render) and the fmtAgo /
fmtElapsed / fmtDuration / truncate formatters (used 10/15/4/6/6 times
across the file). Pull them into a new dashboard-internal util.js so the
upcoming per-tab modules can import them instead of depending on the
entry's closure — the helper analogue of the state.js roster keystone.
They stay out of the cross-page common.js (their phrasing is
dashboard-specific) but are now a shared dashboard module. All five are
pure, so this is behaviour-preserving code motion; esbuild inlines util.js
into the tabs.js bundle. Build green.
The runner hardcoded a 3h per-job timeout. Surface it as
services.hyperhive.forge.ci.jobTimeout so a stuck or runaway job is
bounded by a sane default while staying overridable for operators who
legitimately run longer jobs.
Default drops to 1h (comfortably covers a cold-cache nix build while
bounding a hang far sooner than the old 3h); raise it via the option
for genuinely long jobs. Accepts a Go duration string.
Resolves the stuck-runner concern from the CI-outage follow-up with a
runner-enforced job timeout rather than an external watchdog, per
operator direction.
The tracker-tag lint's trailing class `[^0-9a-fA-F]` treated a non-hex
letter as a valid terminator, so it flagged hash-route fragments like
#24h / #1h (the stats-page window-selector routes) as tracker tags — four
false positives in the frontend. A real tracker tag is never immediately
followed by a letter, so widen the trailing class to [^0-9a-zA-Z]: still
catches every real tag (followed by space / punctuation / EOL) while
excluding any digit-run glued to a letter. Letter-bearing and 6-8-digit
hex stay excluded as before.
Also write theme.css's --crust fallback in six-digit form (#000 -> #000000)
to dodge the one remaining pure-numeric-short-hex residual the lint
documents. Pixel-identical colour. With both, the frontend tree
contributes zero lint hits; only the legacy backlog (Rust) remains.
Pattern + cases reviewed and approved by atlas on the issue.
.dashboard-chrome was a plain block, so the inline-block home back-link
and the block-level .tabbar stacked on two separate rows. Make the chrome
a flex row (align-items: center) so the '← home' link and the tab strip
sit side by side. The chrome now owns the horizontal gutter and the
single full-width bottom divider; .dash-home-back and .tabbar drop their
own side padding / border accordingly. CSS-only — the markup already had
both as siblings of the chrome header.
Second module of the tabs.js split (after the roster keystone). The
P3RM1SS10NS tab — the per-agent capabilities + tool-groups matrices — is
the cleanest leaf: six contiguous functions with no module-scoped state
(each render builds fresh from the fetched payload) and no cross-domain
references except the agent roster, which it now imports from state.js.
Moves applyCapabilitiesChanged / applyToolGroupsChanged (the live-update
handlers wired into the entry's mutation dispatch table) and
fetchAndRenderCapabilities / fetchAndRenderToolGroups (called on tab
activation) into a new permissions.js, exporting those four; the two
renderers stay module-private. tabs.js imports the four — the dispatch
table and tab-activation call-sites resolve unchanged.
Behaviour-preserving: pure code motion. esbuild inlines permissions.js
into the tabs.js bundle, so the static output is unchanged. Build green;
tabs.js drops 258 lines.
The server-warning rows composited their level tint (amber/red) over
`transparent`, so only 16-18% of the colour was opaque. Because the bar
is `position: sticky; top: 0`, page content scrolled up behind it showed
through the see-through background. Composite the tint over the elevated
surface grey (--bg-elev) instead, so the bar reads as a solid grey strip
with the level tint on top — no bleed-through. Affects every page that
shows the banner (the flow page is where it was spotted).
First step of splitting the 4.3k-line tabs.js monolith into per-tab
modules. The live agent roster (containersState) is the one piece of
genuinely cross-domain state — read by the container tree, the
capabilities + tool-group matrices, the schedule target-chips, the
container-load poll, the selection bar and the operator inbox. Move it
(and its snapshot-sync) into a new dashboard/state.js that each domain
module imports, so the upcoming per-tab extractions have a single source
of truth to depend on instead of a shared IIFE closure.
Minimal by design — only the roster moves; the container row cache and
the apply-handlers (which also re-render) stay with the container domain.
Behaviour-preserving: the imported Map is mutated in place exactly as
before. esbuild inlines state.js into the existing tabs.js bundle, so the
static output is unchanged. Build green.
Add a per-job timeout so a wedged build fails in minutes instead of
hanging until the runner's 3h cap. 30 min on the nix flake check job
(well above a cold-cache rebuild, well under the hard cap) and 5 min
on the seconds-long tracker-tag lint. Complements the runner-level
self-heal (a watchdog that restarts a deadlocked runner): the job
timeout catches a hung job, the watchdog catches a wedged runner.
