Adds five expert profiles to the existing hermes pod, managed from git:
hestia (home automation), atlas (platform), apollo (media), clio
(documents and web), daedalus (dev and CI). Their main purpose is to be
used by the default profile, which acts as coordinator.
- experts/<name>.md: each expert's role and domain (apps, repos, domain
rules). ConfigMap hermes-experts.
- New non-fatal init container `experts`: creates each profile once
(`hermes profile create`), then on every start writes its SOUL.md
(shared SOUL.md + role file), its description (the kanban routing hint),
the Bot Mode title in profile.yaml (`ui_meta.hermes-bots`, which also
enables message_agent in Bot Chats), skills.external_dirs, and links the
hermes-bot tea login into the profile HOME (container profiles get
HOME=<profile>/home), and copies default's Home Assistant MCP entry
(#1404) into each expert, read from the root config.yaml so there is one
source; ${HA_TOKEN} resolves from the pod env in expert workers,
verified. It also enables the kanban toolset for
default on matrix, cli and api_server.
- SOUL.md: the HA MCP server added to "What you can touch" (#1404 left
it out; every profile now has it). "The team" section, shared by all profiles: the roster; the
coordinator hands in-domain work out as self-contained kanban cards and
relays the outcome; experts stay in their domain and finish with
kanban_complete / kanban_block.
- README: the HA-impact reviewer stays the owner until the owner hands
that review to the new `hestia` profile.
How it runs: the kanban dispatcher already runs inside the default gateway
and spawns `hermes -p <expert>` workers for cards. Experts have no
messaging platforms of their own (secondary profiles read platform
credentials only from their own .env), so nothing double-binds Matrix or
the API server. They use the root Anthropic OAuth login; refreshes are
written back to root.
Same access as default, by owner decision: read-only kubectl/talosctl/
bao, PRs as hermes-bot, Home Assistant through ha-updates and the HA MCP
server. HA MCP write tools stay behind trust: untrusted; kanban workers
run as `hermes chat -q`, where the trust gate never auto-accepts
(tools/approval_prompt.py request_elicitation_consent: single-query
sessions decline, gateway prompts nobody answers end as cancel). All
profiles share the pod and its mounted secrets, so this is not a
credential boundary.
Verified locally in the current image (build.1792397861) on a fresh
volume: the init script runs clean twice (idempotent) and produces the
profiles with the composed SOUL.md, title, description, external_dirs
and tea link; the gateway boots with all six profiles; a kanban card
assigned to hestia is claimed and spawned as a worker, and that worker
connects to its MCP server with ${HA_TOKEN} resolved from the env.
Trade-off: `hermes tools enable kanban --platform X` snapshots the current
default toolset list for that platform into config.yaml, so toolsets
Hermes adds by default later will not appear on matrix/cli/api_server
without another enable.
After merge: nothing manual. Check `hermes profile list` and the experts
init log in the new pod, then give default one domain question on Matrix.
Review 1765/1766 on #1400 (merged before the review was acted on):
- Window enforced in ha.py, not only by the prompt: install refuses outside
02:00-05:00 Europe/Berlin or with < 30 min left, restart with < 20 min
left, before any HA call. install stops waiting 5 min before 05:00
(timeout capped to the time left) instead of the 60 min default.
- restart: "running" only when HA was observed going down; otherwise
result not_restarted, exit 1, so HACS updates are not reported as applied
while their code is not loaded. Poll every 2 s.
- SKILL.md: the hard limits, not_restarted handling, a refused rollback is
reported for the owner, and config-sync's live->main PRs in
konairius/homeassistant as an expected side effect of HACS frontend
updates.
- README: records the owner decision (automatic HACS/firmware updates and HA
restarts by Hermes inside the window, UniFi/SMLIGHT excluded) next to the
existing exception for HA image bumps.
- Offline unittests (skills/tests/test_ha.py): window edges incl. CET/CEST,
refusals before any HA call, error-signature masking, compare (no change,
recovery, regressions, safe mode), restart observed vs not_restarted.
Settled from the review's unverified list: tzdata is present in the image
(ZoneInfo("Europe/Berlin") works in the hermes pod); the live HA endpoints
(core/state, check_config, error_log, template config_entry_*) answered with
the real token; Hermes cron runs in UTC (no timezone configured, pod TZ UTC).
Already fixed in #1401: token volume optional, init container non-fatal, CI
step that runs skills/tests (this PR's test file is picked up once #1401 is
merged).
Not changed: the admin token. update.install, restart, check_config,
template and error_log need an HA admin; the owner chose to run this in
Hermes rather than the narrower proxy design of #1043.
Cold-rebuild and HA-rollback runbooks, README and the image-automation README
name main and konairius/homelab. The plan records the Flux ref patch and the
cron API calls Cut-over 2 needed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Removes the Paperclip deployment (namespace paperclip, CNPG cluster
pclp-pg, ingresses, image automation, server image build) and the
infrastructure that existed only for its agents:
- agent sandbox: paperclip-renner namespace and CNPs, the
kubernetes-sigs/agent-sandbox controller with its namespace, CNP and
Renovate rules
- agent-runtime-{go,ops} images and the build-agent-runtime pipeline
- matrix bridge (Element <-> Paperclip agents) and synapse's appservice
registration mount
- monitoring loki-read-gateway (only called by the paperclip Role)
- MQTT account agents-read and the controller's RBAC in ns paperclip
- the OpenBao secret-exists-check policy file
- agent admits in the wiki, lidarr, shelly-manager, mosquitto and
hubble-relay CNPs; the paperclip-api Gitea webhook host; the wiki API
key; the ClusterSecretStore namespace entry; the pclp-pg RECOVERY row
- the Paperclip design spec and plan
Authentik group "Paperclip Users" is switched to state: absent so the
blueprint deletes it. package-retention's EXCLUDE_PACKAGES becomes
optional, since its only value was the agent-runtime prefix.
The shelly-manager api-proxy sidecar stays: mqtt-account-controller
calls shelly-manager-api too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The stock agent-runtime-claude base has no /usr/share/zoneinfo:
TZ=Europe/Berlin date prints UTC and Python ZoneInfo raises
ZoneInfoNotFoundError (REN-1195). k8s-base and k8s-go run these images.
Install tzdata and fail the build if Europe/Berlin does not resolve.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Built static in the go stage with go install (module checked against
sum.golang.org); added to the default-PATH check and the README table.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
npm install -g ignored Paperclip's pnpm override and pulled
claude-agent-sdk 0.3.257. Install from a package.json with the override
(0.3.280) into /opt/claude-agent-acp and link the bin into /usr/local/bin.
The build fails if the SDK version is not 0.3.280.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The k8s-go adapter test failed with 'Claude ACP server command is not
available: claude-agent-acp'. Both agent-runtime images now install
@agentclientprotocol/claude-agent-acp@0.73.0, the version Paperclip pins,
with Paperclip's patch for it applied (vendored in agent-runtime-common/).
go and gofmt are also linked into /usr/local/bin, so every tool is found
with the default PATH, not only with the image ENV. The build checks that
with env -i. Kaniko context is now ci/images so both images can COPY the
shared patch.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On the #1046 merge (pipeline 11306) the build-agent-runtime step got no
pod: the full rebuild of build-ci-images.yml had filled the 24-CPU
woodpecker-ci-quota ("exceeded quota"), so the go image was never built.
Move the step to .woodpecker/build-agent-runtime.yml. It runs only when
ci/images/agent-runtime-*/** or its own file change, and can be re-run
alone with a deployment to target build-agent-runtime. Step content is
unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Two pod images for the Paperclip Kubernetes sandbox (REN-284 step 2):
the stock agent-runtime-claude, pinned by digest, plus
- agent-runtime-go: Go 1.27.1, golangci-lint v2.14.0
- agent-runtime-ops: kubectl v1.36.5, flux v2.9.5, kustomize v5.8.1, helm 4.3.0
Tools are copied from their upstream images pinned by tag and digest, so
the existing Renovate dockerfile manager bumps them. Pushed to
konairius/agent-runtime-{go,ops}/agent-runtime-claude:v1 because the plugin
keeps that name and swaps only the registry prefix.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Owner decision on REN-240 (protection rule 5):
- README.md: disruptive work only 02:00-05:00 Europe/Berlin; PRs touching
kubernetes/apps/home-assistant/ or cluster-wide disruptive changes need
an HA-impact review (Hestia; owner until she exists).
- kubernetes/RUNBOOK-home-assistant-image-rollback.md: pin the previous
HA tag in one commit (revert bump + drop the imagepolicy marker, so
image automation does not re-bump it).
Docs only. No manifest changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
claude-bot
