The hermes-agent installs its CLI toolchain at runtime into the shared
/opt/data/tools Longhorn volume, but every download hardcoded arm64. On
the amd64 node titan-22 that left configure-agent-clients failing with
"Missing optional dependency @openai/codex-linux-x64" and the operator
toolchain fetching arm64 binaries, so the pod churned.
Detect the running node's arch (uname -m; fail closed on anything but
aarch64/x86_64) and resolve every asset per-arch:
- install-agent-tools init script (agent-deployment.yaml): ttyd and
kubectl download the arch-correct asset with the arch-correct sha256
(real ttyd 1.7.7 x86_64 and kubectl v1.33.3 amd64 checksums added; the
arm64 ones kept). The npm CLI stamp is now arch-specific
(.cli-versions-<vers>-${arch}) so a fresh arch re-runs npm install and
pulls its own native optional deps; npm keeps both arches' packages.
- install_agent_tools.sh: flux/helm/kustomize/jq/yq/gh/vault/sops/age/
k9s/terraform/go URLs, tarball subdirs (helm linux-${arch}, gh dir),
and checksums are all arch-resolved with both arches pinned. Stamps
and the Go tree are arch-specific, and an active-arch marker forces a
republish of the single-arch ${bin} binaries when the pod moves
between arches on the shared volume. Single fetch/verify helper kept.
Tests updated to assert the arch-aware form (both arches' Go checksums,
${dl_arch} templating) instead of the arm64-only literal.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BvMSXH8VH2tMWXanb8SJdf
A transcription that landed while the private Whisper service was
restarting (an image roll) crashed hermes_stt_client.py with a raw
urllib ConnectionRefused traceback that got dumped into the
conversation. The client now retries the request with backoff (up to 5
attempts, ~10s - long enough to ride an STT pod restart) and, on a
persistent outage, exits with one concise line ('speech transcription
unavailable...') instead of a stack trace. Delivered via the coordinator
ConfigMap; the next reconcile picks it up.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BvMSXH8VH2tMWXanb8SJdf
Schemas, examples and a flag registry for the twelve HUX cards, a dependency-free
validator, the governance rules (memory ledger, autonomy matrix, friendly modes
mapped to real Switchyard routes, privacy defaults, suggestion gating, release
state machine) and the contract doc UI work codes against.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RNPhwu2bsaRNg3DETSAZoM
Port the original #27 detected-language pipeline onto the verified PR #39 prerequisite while preserving the current-main conversation instrument and host continuity changes.
Keep voice selection server-side with no user selector or client voice field. Reuse 207c16ab only for its stricter exact-code trust boundary, omitting malformed or absent language so Piper defaults to Amy.
Capacity/auth/quota failover for cli-auto previously called select_route
with a hardcoded manual lane (cli-{alternate}-{effort}), so the retry
boundary was classified as switchyard-manual instead of going through
Jetson automatic classification, making the routing evidence misleading.
Now the retry calls select_route(context, "cli-auto", exclude_provider=...)
so the boundary stays automatically classified with an explicit
failed-provider exclusion. If the classifier reclassifies to a lower
effort than the original route, the lane re-pins the chosen provider at
the original effort floor so a capacity failure never silently downgrades
a high/xhigh task. Manual lanes (assignee != cli-auto) remain unchanged
and still fail closed without switching providers.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The Vault CSI secrets-store projects /vault/secrets/node-ssh-public-key
as a symlink chain (file -> ..data/file -> ..<timestamp>/file), which
the O_NOFOLLOW hardened reader refuses with ELOOP, crash-looping the
hermes-node-ssh-access key reconciler on most nodes.
Resolve the projection safely instead of relaxing the reader: realpath
the secret strictly (loops and dangling links become a clean refusal),
refuse any resolution that escapes the read-only mount, then apply the
unchanged O_NOFOLLOW regular-file read to the resolved target. Host
account databases and authorized_keys keep the strict no-symlink read.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fix reviewer BLOCK: the stateless poll loop re-planned every terminal card each
tick, so SHIP and the two cycle-limit escalations - whose source cards stay in
the done state - re-fired their comment/block on every pass (~2880/day/chain).
- Add a persistent emission Ledger (/opt/data/supervisor/emitted.json): SHIP and
escalate perform their action only when the (action, target) key is absent,
then record it, so each fires at most once and survives a pod restart. Spawns
remain self-limiting via the existing dedup scan.
- Cycle-limit escalations now target the SOURCE review/repair card (not the impl
root), so the card also leaves the done state and is skipped next tick even if
block_task round-trips imperfectly.
- Harden the _spawn TypeError fallback: re-run the dedup scan before retrying so
a post-insert TypeError can never double-insert.
- Add across-ticks idempotency tests (10x supervise_once -> exactly one
comment/flag/block) plus ledger persistence/corruption and spawn-guard tests.
Both modules stay 100% line+branch, <500 LOC. Stacks on the merge train
(base 5f27e50c).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
