new sqlite table at /state/hyperhive-turn-stats.sqlite on each
agent's state dir. one row per claude turn captures identity
(model, wake_from, result_kind), timing (started/ended_at,
duration_ms), cost (input/output/cache_read/cache_creation token
counts), behaviour (tool_call_count + per-tool breakdown JSON),
and post-turn snapshot metrics (open_threads_count,
open_reminders_count).
wire additions:
- AgentRequest/ManagerRequest::CountPendingReminders +
Broker::count_pending_reminders_for(agent)
- Bus::observe_stream + take_tool_calls — pumps the existing
stdout stream-json, picks out tool_use blocks, accumulates per
turn. bin loops fold the breakdown into each row.
- TurnStats::open_default + TurnStatRow + record() — best-effort
inserts; failures log + don't block the harness.
both ag3nt and m1nd bins capture started_at + duration via
Instant::elapsed, fetch open-thread + reminder counts from
hive-c0re via the existing socket (post-turn, best-effort), and
record one row at turn_end. record_kind splits ok / failed /
prompt_too_long; failures carry the error message in note.
todo entries for host-side vacuum sweep + reading the table back
into agent/dashboard badges.
new LiveEvent variants on the per-agent bus —
status_changed / model_changed / token_usage_changed /
turn_state_changed — replace the per-agent web UI's
/api/state polling for the badge row.
emit sites:
- Bus::set_model → model_changed
- Bus::record_usage → token_usage_changed
- Bus::set_state → turn_state_changed
- turn::wait_for_login → status_changed("online") on creds detect
- post_login_start / post_login_cancel → status_changed("needs_login_*")
per-agent endpoints (post_set_model / post_compact / post_new_session
/ post_cancel_turn / post_login_*) now all return 200; client
drops the post-submit refetch except on login transitions, which
still need /api/state to render the OAuth form + session stream.
client adds dispatch on the four new event kinds, threads
`currentLabel` through so the composer re-enables on a live
status flip, and no longer fires refreshState() from turn_end or
postModel — the events carry the same signal faster.
closes the per-agent half of the dashboard event-channel
refactor; TODO entry dropped.
drop the "● harness alive — turn loop running" paragraph; the
new #alive-badge chip in the state row carries the same signal
across all statuses (loading / online / needs-login / offline)
with colour coding. token-usage chip renamed + restyled as
#ctx-badge — primary number is total context-window tokens
used, mirroring claude code's "N tokens" indicator.
every state-row badge now has hover detail: state-badge gets
per-state tooltips + age suffix, model-chip explains the
/model command, last-turn shows the raw ms duration, ctx-badge
breaks out input / cache_read / cache_write / output.
new todo entry for the per-turn stats sink (start/end/model/
tokens/tool-call-count) the harness should be writing.
new 'qu3u3d r3m1nd3rs' section between approvals and operator
inbox. lists every pending reminder with agent, due-relative
timestamp, body, payload path (path-linkified), and a cancel
button. drives off a new /api/reminders endpoint and a
POST /cancel-reminder/{id} that hard-deletes the row.
failure surface (last_error / attempt_count + retry) deferred —
needs a sqlite migration; tracked in TODO.md.
agents constantly emit pointer strings to /agents/<n>/state/foo.md
since broker bodies cap at 1 KiB. now those tokens linkify in the
message flow, question bodies, answer text, and operator inbox;
clicking expands an inline <details> that lazy-fetches via the
new /api/state-file?path=... endpoint.
endpoint allow-list: per-agent state dirs + shared docs, both
in their container-mount form (/agents/<n>/state, /shared) and
host form (/var/lib/hyperhive/...). 1 MiB read cap; canonicalises
before the prefix check so `..` / symlinks can't escape.
legacy bare `/state/...` is deliberately not matched — ambiguous
from the host's perspective (we'd need to know which agent the
message references to translate). agents should use the qualified
form going forward.
questions pane now shows both operator-targeted threads
(target IS NULL) and agent-to-agent threads (target = some
agent). filter chips above the list: all / @operator / @peer /
per-participant. peer rows get a mauve left rule + a 0V3RR1D3
button that POSTs the same /answer-question endpoint
(OperatorQuestions::answer already permits the operator as
answerer on any target).
wire changes: OperatorQuestions gains pending_all +
recent_answered_all; QuestionAdded + QuestionResolved events
carry target: Option<String>; emit sites drop their
target.is_none() guard. answered-history rows show the
answerer prefix so override answers are auditable at a glance.
new DashboardEvent::ContainerStateChanged + ContainerRemoved
close the last refetch loop on the dashboard. Coordinator's
rescan_containers_and_emit diffs a fresh container_view::build_all
against a cached last_containers map and fires per-row events.
called from actions::approve (post-spawn), actions::destroy,
the lifecycle_action wrapper, auto_update::rebuild_agent, and
the existing 10s crash_watch poll.
ContainerView extracted to its own module so coordinator and
dashboard can both build it. dashboard endpoints flip to 200;
container-lifecycle forms carry data-no-refresh. client drops
the periodic poll entirely — initial cold load + SSE for
everything afterwards. pending overlay reads from the existing
transientsState since the new event payload doesn't carry it.
PURG3 + meta-update keep the post-submit refetch since
tombstones + meta_inputs aren't event-derived yet; tracked in
TODO.md.
system prompts now describe the hyperhive Forgejo at localhost:3000,
the per-agent user, the pre-configured tea CLI, and the REST API
fallback with /state/forge-token. todo gains the shared docs/skills
RO-repo follow-up (org-shared + per-agent read membership).
new NixOS option in harness-base.nix:
hyperhive.allowedRecipients = [ 'alice' 'manager' ]; # whitelist
hyperhive.allowedRecipients = [ ]; # default = unrestricted
module writes the list as JSON to /etc/hyperhive/send-allow
.json at activation. AgentServer::send reads the file before
issuing the broker request; if the list is non-empty and
`to` isn't on it, the tool returns a claude-readable refusal
string without touching the broker. the manager is always
implicitly permitted regardless of the list — otherwise a
misconfigured allow-list could strand a sub-agent without an
escalation path.
enforcement is in the in-container MCP server (not on the
host's per-agent socket) because the agent's nix config is the
trust boundary anyway — the operator audits agent.nix at
deploy time, the activation-time /etc/hyperhive/send-allow
.json is r/o under /nix/store, so the agent can't tamper at
runtime without going through a new approval.
agent prompt mentions the option + tells claude to route
through the manager when refused. retires the matching TODO
under Permissions / policy.
after reverting the 3-column attempt (74ba8a6), keep the
single-column layout but put related sections adjacent:
swarm: containers → kept-state → meta-inputs
decisions: questions → approvals
messages: operator-inbox → message-flow + compose
this is a free improvement — the operator scrolls through one
logical group at a time instead of bouncing between swarm /
decisions / messages mid-page. follow-up improvements
(collapsing rarely-active sections, multi-column at wide
viewports done less aggressively) captured in TODO under
'Dashboard layout overhaul'.
every snapshot source backing /api/state used .unwrap_or_default()
— sqlite errors, broker errors, nixos-container list failures,
operator_questions decode crashes all degraded to empty lists
without a log line. the 'pending question doesn't render'
bug we've been chasing was likely a row-decode panic in
OperatorQuestions::pending() being swallowed this way.
new log_default(what, result) replaces each call site: same
default value on Err but emits target=api_state warn with the
source name + dbg error first. five sources covered:
nixos-container list, approvals.pending,
approvals.recent_resolved, broker.recent_for(operator),
questions.pending. next time the question goes missing the
journal will say which source failed and how.
todo updated — pending-question entry now points at the new
log instead of three suspect paths.
old behavior: omitted wait_seconds fell through to the 30s
RECV_LONG_POLL_DEFAULT — claude calling 'is there anything in
my inbox right now?' between actions blocked the turn for half
a minute. flip the semantics: None (or 0) returns immediately,
positive value parks up to MAX (180s, unchanged). cleaner
'peek vs wait' distinction; tool descriptions + agent/manager
prompts updated to point at the new shape.
harness's own serve loops in hive-ag3nt + hive-m1nd relied on
the old default for their inbox poll. they now explicitly pass
wait_seconds: Some(180) to opt into the full park — same
effective behavior as before, just spelled out.
retires the matching TODO under Turn loop.
new tabs above the approvals list: 'pending · N' and
'history · M'. active tab persists in localStorage so the
operator can park on history if they prefer. on a fresh
dashboard the default is pending (matches the prior shape).
history view shows the last 30 resolved approvals — newest
first by resolved_at — with one row per approval: status
glyph (✓ approved / ✗ denied / ⚠ failed), id, agent, kind,
short sha, status label, and a relative time chip. when the
row has a note (deny reason or build error), it renders
below in a muted block with line wraps preserved.
backend: Approvals::recent_resolved(limit) queries by
status IN ('approved', 'denied', 'failed') ORDER BY
resolved_at DESC. StateSnapshot gets approval_history (a
lean ApprovalHistoryView without diff_html — rendering 30
git diffs per state poll would be expensive and the operator
already saw the diff at decision time). dashboard's
history_view fn projects the sqlite row.
retires the matching TODO entry.
new entry under UI/UX for an approval history tab on the
P3NDING APPR0VALS section — sqlite already has every row + the
applied repo's annotated denied/failed tags carry the
human-readable reasons, so this is a render-side change.
retire the 'stream nixos-container stdout' entry — landed in
6f1b664. run() now pipes child output line-by-line into
tracing so 'slow build' no longer looks like 'wedged daemon'.
new NixOS option in harness-base.nix:
hyperhive.extraMcpServers.<key> = {
command = "/path/to/server";
args = [ ... ];
env = { KEY = "value"; };
allowedTools = [ "send_message" "join_room" ]; # or ["*"]
};
declared as attrsOf submodule so agents stack arbitrarily many.
the module writes the whole map as JSON to
/etc/hyperhive/extra-mcp.json at activation; the harness reads
that file in mcp::render_claude_config and merges each entry
into the rendered --mcp-config under its own mcpServers.<key>
block. allowed_mcp_tools(flavor) extends the --allowedTools
arg with mcp__<key>__<pattern> for every entry — "*" (the
default) becomes mcp__<key>__* so every tool from that server
is auto-approved, or pass a concrete list to tighten.
collision guard: an extra server keyed "hyperhive" is dropped
with a warn-log so user config can't shadow the built-in
surface. malformed JSON / missing file fall back to "no
extras" silently.
prompt note added: agents see "(some agents only) extra MCP
tools surfaced as mcp__<server>__<tool>" and learn they're
declared via agent.nix. retires the matching TODO under
Per-agent extension. matrix-chat agents + bitburner-agent
migration unblocked.
submit_apply_commit now diffs the freshly-tagged proposal/<id>
against applied/main and refuses if flake.nix is in the
changeset. flake.nix is fixed boilerplate the meta flake
depends on (it exports nixosModules.default = import ./agent
.nix); silent edits there would break the nixosConfiguration
in subtle ways. the manager prompt already says don't touch
it; this is the host-side belt — clear error to the manager
on submit, row marked failed in sqlite, no orphan pending
approval to chase. diff-failure is logged + ignored: the
build path surfaces concrete errors if flake.nix is actually
broken.
surfaced by a real hang investigation today — lifecycle::run
uses .output() which buffers stdout/stderr until exit, so a
multi-minute nix build through nixos-container update looks
identical to a wedged daemon. line-buffered streaming into
tracing (and ideally the per-agent live event bus when the
agent is known) makes 'still building, just slow' visible
without strace gymnastics.
