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).
agents get `pkgs.tea` (gitea/forgejo CLI) and a tea-login oneshot
that runs `tea login add --url <hyperhive.forge.url> --token
$(cat /state/forge-token)` before the harness starts. idempotent:
exits 0 when the token file is absent (hive-forge not on) or when
~/.config/tea/config.yml already exists. new
`hyperhive.forge.url` option (default http://localhost:3000) so
operators can point at a non-default forge port. claude can now
shell out to `tea repos create`, `tea pulls create`, etc.
bumped from (read:user,write:repository,write:issue) to also include
write:user (own profile + create repos under own namespace),
write:organization (share namespaces between agents), write:misc
(hooks/attachments). still excludes admin and package scopes.
new forge module probes the hive-forge nixos-container (no-op when
absent), and ensures every agent + the manager has a forgejo user
named after them with an access token at `<state>/forge-token`
(visible inside the container as `/state/forge-token`).
idempotent: skips user creation when forgejo reports 'already
exists', skips token issuance when the file is present, scopes the
token to read:user,write:repository,write:issue. token-name suffixed
with a clock so re-issuing doesn't collide with a stale name. shells
out via `nixos-container run hive-forge -- runuser -u forgejo --
forgejo admin` (runuser instead of sudo since sudo isn't in the
container by default).
hooks: ensure_all sweeps existing containers at hive-c0re startup
(backgrounded), and the actions.rs spawn task calls ensure_user_for
the new agent right after lifecycle::spawn succeeds. failures log a
warning but don't abort spawn — a missing token is recoverable from
the next startup sweep.
avoids fighting an operator-side `services.forgejo` over the
singleton module options. container shares host netns
(`privateNetwork = false`) so agents still dial the forge via
plain `localhost:<httpPort>` and the host firewall is the only
layer that matters. container name is `hive-forge` (no `h-`
prefix) so hive-c0re's lifecycle scanner ignores it — operator
manages it with the standard `nixos-container` CLI. state lives
at `/var/lib/nixos-containers/hive-forge/var/lib/forgejo/` and
survives restarts.
new `services.hive-forge.enable` (off by default) wraps
`services.forgejo` with hyperhive-friendly defaults: sqlite (no
extra db service), built-in ssh on 2222 so it doesn't fight the
host's openssh, http on 3000 (outside hyperhive's 7000/8000/8100-8999
ranges), registration off (operator seeds agent users), private
repos by default. exported as `nixosModules.hive-forge` — operator
imports it on the host alongside hive-c0re. container-side wiring
(MCP tools or a bind-mounted token) is deferred; containers already
share the host netns so they can reach http://localhost:3000 today.
bare set_transient/clear_transient pairs leak the in-memory transient
on task cancellation, panics, or any early return between the two
calls — dashboard then shows the agent stuck in 'rebuilding…'
forever (coder hit this today). add Coordinator::transient_guard
returning a TransientGuard whose Drop clears, and convert every
caller (dashboard lifecycle_action, auto_update::rebuild_agent,
manager_server Update, actions::destroy, actions Spawn task,
migrate phase 4). destroy() now takes &Arc<Coordinator> so it can
hold a guard. existing stuck transients clear on next hive-c0re
restart since transient state is in-memory only.
install_configured now takes an optional notify recipient. on a
non-zero or spawn-failed 'claude plugin install', sub-agents send
the spec + stderr to manager via the hyperhive socket; manager
passes None so it doesn't message itself. boot still proceeds either
way — notification is best-effort.
run_claude now keeps a 20-line stderr ring buffer and bails with it
inline (was just 'exit <status>'). agent serve loop, on Failed (not
PromptTooLong — that's already absorbed by drive_turn's compaction
retry), sends the error body to manager via the normal hyperhive
send. swallows transport errors — failure is already in journald
and the events sqlite. manager-only harness (hive-m1nd) is unchanged
so it doesn't try to notify itself.
claude-code rejects --dangerously-skip-permissions / defaultMode=
bypassPermissions when running as root, which all hyperhive
containers do. revert to the previous explicit allow-list plumbing
(per-flavor list spliced into permissions.allow + --tools enable
list), keep TodoWrite out of the built-in allow set, and keep the
deny list (TodoWrite, WebFetch, WebSearch, Task) as belt-and-braces
in case anything sneaks past the allow gate.
bus-only note made post-mortems require the web UI / events sqlite;
now stderr lines also land in 'journalctl -M <container> -b' alongside
the existing LiveEvent::Note for the dashboard.
socket client now retries connect/IO failures with 2-4-8-16-30s
backoffs (60s total budget). transparent for non-tool callers via
request(); tool handlers go through request_retried() which also
returns the retry count, then annotate_retries() appends a one-line
note to the tool result so claude knows the slow round-trip was a
c0re flicker, not a content failure — avoids burning tokens on an
LLM-level retry.
new hyperhive.claudePlugins NixOS option (list of strings) rendered
to /etc/hyperhive/claude-plugins.json. both hive-ag3nt and hive-m1nd
shell out 'claude plugin install <spec>' for each entry once at
startup before the turn loop opens. failures log a warning but don't
abort boot.
claude-settings.json now sets permissions.defaultMode=bypassPermissions
with a small deny list (WebFetch, WebSearch, Task, TodoWrite). The
per-flavor allow list and --tools / --allowedTools CLI flags are gone
— anything not denied auto-approves. mcp.rs loses ALLOWED_BUILTIN_TOOLS,
builtin_tools_arg, allow_list, allowed_mcp_tools. The extraMcpServers
allowedTools field is parsed for back-compat but no longer wired
anywhere; restrict via permissions.deny instead.
agents weren't being woken with the 'you were rebuilt — check
/state/ for notes, --continue intact' system message after
several recent rebuild surfaces:
- auto_update::rebuild_agent — used by the dashboard rebuild
button, admin-CLI rebuild via lifecycle_action, the startup
rev-scan, AND the new meta-input update batch loop. kick
moves *into* rebuild_agent's success arm so all four
paths benefit. (the dashboard's lifecycle_action extra
closure was already firing kick — now it's a no-op for the
rebuild path since rebuild_agent does it.)
- actions::run_apply_commit — apply-commit approve flow built
+ tagged deployed/<id> but never kicked. add kick on
success with the more specific 'config update applied' hint.
- server.rs::HostRequest::Rebuild — the admin-CLI direct path
calls lifecycle::rebuild bypassing rebuild_agent. add kick
on success.
dashboard's restart / start lifecycle_action extras still
kick via their own closures since they don't route through
rebuild_agent. stop / kill / destroy intentionally don't
kick — there's nothing to wake.
read_meta_inputs() previously only included direct inputs of
meta's root node — so a manager-added 'inputs.mcp-matrix' in
agent-dmatrix's flake.nix never surfaced in the dashboard
panel even though it's a real fetched input that nix can
update.
now: BFS the flake.lock graph from root to depth 2. emits
one MetaInputView per fetched (non-follows) node, names are
slash-paths from root — 'hyperhive', 'agent-coder',
'agent-dmatrix/mcp-matrix', 'hyperhive/nixpkgs', etc. that's
the same syntax 'nix flake update' accepts for transitive
inputs, so the existing POST /meta-update path needs no
nix-side change.
depth limit of 2 keeps the panel readable — deeper transitives
(nixpkgs's own deps etc.) would explode it; bumping a level-2
entry re-fetches its sub-inputs anyway.
POST /meta-update's 'which agents to rebuild' derivation
updated for the slash names: anything under hyperhive/
fans out to all agents (shared base); 'agent-<n>/...' picks
out the agent name from before the first slash.
read_meta_locked_revs (used by the deployed:<sha> chip per
container) split out into its own straight root-input lookup
since the chip only cares about the agent's own input.
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'.
