Repository navigation
Conversation
The omp role installed omp but never pointed it at the fleet's auth broker (`omp auth-broker`, running on punjab), so a rebuilt machine had an empty ~/.omp/agent/config.yml and no provider credentials until someone ran /login by hand. The role now sets auth.broker.url from group_vars and, when Bitwarden is unlocked, writes auth.broker.token from the item omp-auth-broker-token. A locked vault leaves an existing token alone. The config file is set to 0600 and the role reports `omp auth-broker status`. just seed-omp-broker (run on punjab) creates the Bitwarden item from `omp auth-broker token`, which prints the token without rotating it. The broker's own systemd unit on punjab is still hand-made and unmanaged.
…dead device cleanup
Read the broker token over Tailscale SSH when seeding Bitwarden, and preserve the vault when the server is unreachable. Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
|
Read-only laptop smoke: archived commit ad4f8c6, ran native |
|
CI also found an existing server-image blocker: |
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
|
Replacement Punjab is active at 100.103.187.46 under tailnet name punjab. Final home sync and representative hashes passed; Kiro Crew removed. Corrected DHCPv6 on the IPv4-only masquerade interface; full VM start and subsequent guest reboot returned authenticated SSH, broker, Atuin daemon and osquery reporting/polling. Real native just apply-nosecrets passed. Relevant full Python modules: 206 passed; full AWS mock-provider module: 7 passed; VM manifest server dry-run passed. Earlier broader local bootstrap tests had three failures caused by installed Bitwarden CLI shadowing stubs; not represented as a clean full-suite pass. Atuin cloud login remains a pre-existing local/vault key mismatch; history preserved. runs-on stack is DELETE_COMPLETE with both S3 buckets/versions retained. EC2 final shutdown/EIP release runs from the replacement and records a root-only receipt; retained root volume/snapshots/S3 still incur storage charges. |
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
|
Final cutover evidence: replacement is now the session host (100.103.187.46, Ubuntu generic kernel). Its punjab-aws-cutover.service journal records successful completion at 2026-10-28 19:27:15 UTC after observing EC2 stopped, releasing its EIP and retaining recovery storage. AWS logout was intentional. Root-only completion receipt preserves explicit journal provenance; a later check-only authentication failure is retained separately. Current broker, Atuin daemon, update timer, osquery daemon and fleet poll checks pass. Kiro Crew remains masked/inactive with launcher absent. Matrix, auth, School, Reef and Hub each return HTTPS 200. All current-host staging/active-EC2 claims were updated. Retained EBS/snapshots/S3 still incur storage charges. |
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
|
Atuin recovery completed with owner-approved clean cloud replacement. Preserved all 23,133 readable cloud, 512 Punjab and 10 Kerala history UUIDs plus 105 additional raw-shell entries; 23,761 visible entries include one real daemon capture/cloud round-trip verification command. SQLite integrity and every active encrypted record verify. Independent fresh-client download matches the full history UUID set. Punjab, Karnataka and Kerala sync using the existing verified Bitwarden key; prior Punjab/Kerala keys are hidden recovery fields. Original server ciphertext, including 245 records under 11 unavailable keys, is retained locally and in decrypt-tested/checksum-verified age archives on Spindle; Himachal remained offline/untouched. Installed official atuin-sync on Kerala. Fixed the unit to trust atuin status exit status rather than the obsolete [Sync] heading. Full affected Atuin/package-check modules: 20 passed; shellcheck passed. Broader platform module run had one unrelated existing Termux boot-script migration failure; Kerala apk-tools3.0.8 also rejects the package unit existing --no-upgrade option, so its direct package replacement is not claimed as a full APK-unit validation. No secrets/history content committed. |
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
|
2026-10-29 live AWS shutdown re-verification: account181185361136 has no running workloads in the enumerated services across all17 enabled regions. Regional inventory:332 successful reads;8 endpoint errors resolved as unsupported service/region pairs via AWS public SSM availability registry. Another68 AutoScaling/SpotRequest/SpotFleet/EC2Fleet checks all returned empty. Direct final check: Punjab EC2i-085690e02ca98c95a stopped, no us-east-1 EIPs. All3 S3 buckets,250GiB gp3 root volume and5 snapshots retained; both DLM policies disabled. Independent live Kiro subscription API confirms Power and ProPlus both CANCELLED, expiring2026-10-31T23:59:59.999Z; remaining October Power billing is not an active renewal. No AWS mutations were necessary because shutdown/cancellation state was already achieved. Migrated Punjab passed strict-host-key SSH and a fresh omp request explicitly using anthropic/claude-opus-5-5. Private verification receipt retained under ~/backups/aws-shutdown-verification-20261029T013010Z/verification.json. Retained recovery storage is still billable; no zero-bill claim. No tests run for this documentation-only change; evidence is live AWS APIs and actual omp execution. |
Preserve historical cluster contexts and vault notes, scope API firewall access to the tailnet, and document verified client rollout. Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
|
Karnataka K3s documented and live client access rolled out to Punjab, Karnataka, Kerala and Spindle. Each default context is karnataka-k3s and passed authenticated TLS /readyz + Ready-node listing; existing context/cluster/user entries compared and preserved, configs0600. Dedicated Bitwarden note kubeconfig-karnataka-k3s server-refreshed; historical notes untouched. Added all-active-platform unit eligibility, package prerequisites, atomic context merge, safe historical seeder default, and explicit cached-only mode for production hosts. Reproduced API failure before; scoped runtime/permanent firewalld rule restored6443 overtailnet without reload/restart. Full firewall helper still returnsnonzero at pre-existing absent runtime pod-egress policy; not force-activated. 78 kube/check/seeder/firewall tests passed, shellcheck error-level passed, docs built with CI-pinned mdBook0.4.44 and rendered page visually checked. Mumbai/Himachal/Dilli/Kanpur unreachable, Termux8022 refused; repository rollout prepared but no live configuration claimed there. BiharSSH22 refused/unverified and retired VPSes untouched. AWS final deletion still waits for complete off-AWS fleet-backup preservation. |
Record owner-approved deletion of historical backups without a new archive, preserve live Karnataka/Spindle services, and remove stale infrastructure recreation and CI entrypoints. Co-authored-by: Claude <noreply@anthropic.com> Signed-off-by: James Reilly <5840441+hanthor@users.noreply.github.com>
|
Final AWS retirement completed2026-10-29 after explicit owner authorization that old Hive/Spindle backups are unnecessary. Canceled archival transfer and removed only its newlycreated partial payloads. Deleted3S3buckets, all1430versions+413delete markers(185.26GB), oldPunjabEC2,250GiBroot,5snapshots,migrationVPC/network,PunjabSG,2DLMpolicies,5budgets,3IAMusers/credentials,3customroles/profiles andcustompolicy. Final102regional APIchecks across17regions empty; buckets/users/customroles/customerpolicies empty; each5budgetnames returnsNotFoundException. KarnatakaReady,PunjabVMRunning,andfresh explicitAnthropicomp request passed afterdeletion. Removed executableAWSrootmodule/helper/recipes/providerupdate/CIwiring; Gitdcbbfbc retains historicalsource. Removed ONLY obsolete tofu requiredbranchcheck; retained lint/python/tests/justfile/broker bindings/reviewgate. justlint,Justfilerecipesmoke,jq andCI-pinnedmdBookbuild passed; K3s78test result remains aspreviouslyreported. Existinglocalmigrationdata,K3sconfigs,Atuinarchives andliveSpindle/Punjabuntouched. No new completearchive claimed. AWSaccountdefaults/service-linkedroles remain; KiroalreadycanceledexpiresOct31,andfinal/delayedchargespossible. Privatecompletionreceipt: ~/backups/aws-final-retirement-20261029T015927Z/completion.json. |
- flatpak: new flatpark_flatpaks list, installed system-wide from the flatpark remote; system_flatpaks stays Flathub-only. - beeper-login unit: Beeper Desktop only supports emailed one-time codes, so it starts sign-in via the local Desktop API, reads the code over IMAP from the Bitwarden item 'beeper', then unlocks E2EE with the Recovery Code. Skipped with --no-secrets, a locked vault, non-desktops or skip_beeper.
retry.usageReservePct 10 -> 0 and omp-route's Claude candidate reserve 10 -> 0 (OMP_ROUTE_CLAUDE_RESERVE_PCT), so sessions stay on Claude until its window is actually spent. Other omp-route candidates keep 10%. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: James Reilly <jreilly1821@gmail.com>
Changes
Verification