| Age | Commit message (Collapse) | Author |
|---|
|
|
|
f0 cron now does 'touch ... && date +%s > /data/nfs/nfs.LIVE_CHECK' so the
file content (epoch) plus its mtime both cross to f1's read-only sink via
zrepl. Documented the cron % escaping gotcha and the f1 freshness check
(seconds between f1's clock and the epoch in the file). Stalled if lag
exceeds ~10-11 min.
|
|
f0 root crontab touches /data/nfs/nfs.LIVE_CHECK every 10 min, gated on
/data/nfs/nfs.DO_NOT_REMOVE. zrepl replicates zdata/enc/nfsdata to f1's
read-only sink every minute, so the file's mtime on f1 reveals replication
freshness (stalled if >~10-11 min behind). Added to zrepl.md and the hub
SKILL.md reference bullet.
|
|
The only live S3 offsite backup is the zusb quarterly workflow, which
backs up zusb/data/* (not the f-host zdata pool). Verified no aws/s3/
backup cron in root crontab on any f-host and no pre-existing /root/.aws.
zdata redundancy is via zrepl replication, not S3. Also corrected the
aws-credentials caveat (symlink is deliberately dangling on non-zusb
hosts until zusb is replugged there) and noted the 2026-07-20 install
across all f-hosts.
|
|
How to install awscli on a FreeBSD host (py39 on 14.x, py312 on 15.x)
and wire /root/.aws with a credentials symlink to
/opt/snonux/secrets/aws.credentials, which rides on the zusb pool.
Covers both the zdata daily-cron S3 backup and the zusb quarterly
backup. Cross-linked from the zusb note in usb-keys.md.
|
|
zusb is a 4-disk raidz2 USB-SATA pool used as offline backup storage,
loaded ~once per quarter. Migrated from t450 to the f3s USB-key scheme:
zusb/data/enc rekeyed to a raw key on /keys/zusb.key, scripts on all
f-hosts. Note added to usb-keys.md and the hub SKILL.md.
|
|
|
|
skill-maintenance audit found overview.md repeated the SKILL.md intro/role
sentence verbatim (a DRY-within-skill violation). Replace it with a one-line
pointer to SKILL.md so the role/IP details have a single canonical home and the
reference focuses on setup detail.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Extend the f3s- prefix to the two remaining f3s-scoped siblings so the whole
homelab family groups together. Both are f3s-specific: pkgrepo IS
pkgrepo.f3s.buetow.org, and rocky-vm-setup documents the 'rocky' VM on the f3s
host f3.
- git mv both skill dirs (history preserved); update name: frontmatter; add the
'Part of the f3s homelab skill family' note to both descriptions.
- Rewrite every reference to the *skills*: backtick mentions (`pkgrepo` ->
`f3s-pkgrepo`, `rocky-vm-setup` -> `f3s-rocky-vm-setup`), markdown cross-links
(../pkgrepo/ -> ../f3s-pkgrepo/, etc.), the f3s hub Related-skills block and
description, and skill-maintenance's index-pattern example.
- Left references to the package REPOSITORY (pkgrepo.f3s.buetow.org, 'the custom
pkgrepo') unchanged — those name a service, not the skill.
- Verified all 114 relative links resolve. Plan doc updated to record the rename.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Execute plans/f3s-skill-split-plan.md. Carve the oversized f3s skill into six
f3s-prefixed siblings so each loads on its own triggers, and slim f3s to a hub
(38 reference files -> 10; 119 -> 73 lines):
- f3s-storage ZFS/zrepl/CARP/NFS-stunnel, nfs-mount-monitor, backups
- f3s-k3s cluster install, off-LAN access, ingress, etcd, r-node Rex
- f3s-observability Prometheus/Alloy/Loki/Tempo, FreeBSD node_exporter
- f3s-workloads Immich, Garage, Player, yChat, goprecords/uptimed
- f3s-raspberry-pi pi0/pi1 NetBSD static site, pi2/pi3 Pi-hole/LAN DNS
- f3s-dtail dserver deployment/ops (SSH 2222)
f3s hub keeps the master Host-IP table, physical hosts, bhyve, WireGuard mesh,
and off-LAN access as the canonical inward-pointing context, plus a Related
skills block. Applies skill-maintenance best practices: fixes the three inlined
SKILL.md duplications (Pi/webserver, DTail) by moving prose to one canonical home
and keeping each new SKILL.md a slim index that points to its references rather
than re-inlining them. All cross-skill links rewritten to ../../<skill>/references/
form (incl. inbound links from pkgrepo and rocky-vm-setup, and two pre-existing
broken links); verified all 112 relative links resolve.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
|
|
|
|
- FreeBSD: host key now persistent in /var/db/dserver (cleanvar purged
the old /var/run location every boot); document the migrate-first
upgrade order for pre-2026-07-10 packages and that the rc.d precmd
now repopulates the key cache on every start (manual re-run only
needed when authorized_keys changed without a restart)
- OpenBSD: rc.d uses rc_bg=YES + plain rc_cmd since 2026-07-10 (the
old 'rc_cmd $1 &' hid rc_pre failures from rcctl); note that the
FREEBSD_HOST=f0.wg0 override applies to all dtail-* targets
- NetBSD: drop the outdated "unlike FreeBSD" volatile-key comparison
- Verification table: f0-f2 and fishfinger reverified 2026-07-10 with
host key identity preserved; f3 unreachable (still on old package),
blowfish still on the older package
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
|
fec2f9d is the dtail server-side fix (absolute CacheDir resolution
independent of CWD), not a FreeBSD template bug — the FreeBSD template
never used relative paths. Also make the pre-2026-07-10 upgrade
procedure an explicit ordered list: migrate the host key first, then
reinstall, then rcctl restart — neither pkg_add nor Rex restarts a
running daemon, and after a reboot the old host key is already gone.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
|
Document the 2026-07-10 OpenBSD dtail package changes: absolute
CacheDir/HostKeyFile, host key migration to /var/db/dserver, rc_pre
key-cache refresh, same-version reinstall procedure, and the f0.wg0
Makefile override needed from the WireGuard VPN.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
|
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
|
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
|
pi0/pi1 (NetBSD 10.1 aarch64) now run dserver from the custom pkgrepo:
- pkgrepo: NetBSD repo layout (netbsd/10.1/packages/aarch64, unsigned,
pkg_summary.gz), client setup via pkg_add URL, make dtail-netbsd
pipeline (cross-compile on Linux, pkg_create natively on pi0),
manual packaging reference, install/update steps + gotchas,
verification state 2026-07-09
- f3s: pi0/pi1 host-role rows, rc.d service list + npf port 2222,
removed stale 'pi0/pi1 do not run DTail' claims
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
|
|
|
|
|
Amp-Thread-ID: https://ampcode.com/threads/T-019f299f-7596-73b6-ba71-0f71fcd3c131
Co-authored-by: Amp <amp@ampcode.com>
|
|
Updates the pi0/pi1 WireGuard docs to reflect that wireguardmeshgenerator
generates and installs their tun0.conf + rc.d/wireguard script directly
now, closing the previously-documented gap where both nodes' configs were
manually-maintained exceptions outside the generator.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
|
|
Trims dates, "reinstalled"/"converted"/"validated twice" framing, and
forward-looking "next conversion" language from the f3s/pkgrepo skill docs.
Keeps only current-state reference material: pi0 and pi1 run NetBSD, and
how their services (WireGuard, bozohttpd, uptimed, npf, content sync,
goprecords) are installed and configured. bootstrap-netbsd-pi.md is
reframed from a migration runbook into a plain service-reference doc.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
|
|
pi0 and pi1 were both reinstalled from Rocky Linux to NetBSD 10.1 this
session. Adds the reusable conversion runbook (bootstrap-netbsd-pi.md) and
updates every f3s/pkgrepo reference that assumed pi0/pi1 were still Rocky
(WireGuard setup, DTail client lists, pkgrepo URLs/build host, goprecords
automation, wol-f3s doas behavior), plus documents the bugs found along the
way: no native wg(4) on this platform, uptimed's dual-file read bug,
bozohttpd's -V fallback redirect leaking the system hostname.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
|
|
hexai dropped the tmux-edit popup editor feature; drop the tmux
keybinding, update-tool lists, and the orphaned tmux-edit-send script
that duplicated it but was no longer bound to anything.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
|
|
|
|
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
|
- New skill: skill-maintenance (SKILL.md + 4 references: best-practices,
dry-across-skills, sub-division, audit-checklist) encoding the Agent Skills
spec rules, cross-skill DRY, and the f3s/rocky-vm-setup sub-division pattern.
- agent-task-management/references/verification-honesty.md: canonical home
for the general verification-honesty discipline (linked from go-best-practices).
- blog-writing-style/references/gemtext-conventions.md: shared foo.zone
gemtext conventions (linked from compose-blog-post, update-blog-post,
blog-writing-style).
- blog-writing-style/references/patterns-and-examples.md: foo.zone LLM-pattern
catalog + rewrite examples extracted from blog-writing-style SKILL.md.
- f3s/references/{console-jetkvm-shutdown,shelly-plug}.md: referenced by
f3s/SKILL.md (previously untracked).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Corrected mutual pubkey and PSK mismatch between f3 and both OpenBSD
gateways. f3 is now reachable directly via WireGuard from fishfinger or
blowfish without the double-hop workaround.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
|
- Current version bump: 15.0-RELEASE → 15.1-RELEASE
- Added breaking changes section for 15.1: ZFS sysctl rename
(vfs.zfs.min_auto_ashift → vfs.zfs.vdev.min_auto_ashift),
libzfs7/libzpool7 SONAME bump, if_bridge/if_epair MAC change,
CARP failover + NFS sink dataset readonly fix
- Updated minor upgrade example: two-pass flow with correct PAGER=cat
workarounds, doas env-var handling via doas sh -c, sysctl rename step
- Noted carpcontrol.sh fixes: readonly flip on MASTER + rollback on BACKUP
(commits 6c13b6a + c9f06a2 in conf repo)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
|
The prior worker extracted the style sections into four references/ files
(naming-and-formatting, files-and-headers, module-organization,
code-body-conventions) but was interrupted before slimming SKILL.md, leaving
the content duplicated. Resume: slim SKILL.md to intro, When-to-Use, a
reference index, and the Quick Checklist summary (187 -> 44 lines). Content
moved verbatim; nothing dropped.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Reorganize the 217-line SKILL.md into a slim 50-line entry point
(purpose, principles, core workflow, reference index, quick reference)
plus four focused references, matching the f3s / llm-benchmark-comparison
pattern. Pure reorganization; all content preserved.
References:
- beets-layout-and-dedup.md (layout, sorting, duplicates plugin)
- navidrome-rescan.md (full scan on host and in Kubernetes)
- cmus-reindex.md (lib.pl/cache, offline and live re-index)
- tiddl-downloads.md (auth, batch loops, quality/output config)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Reorganize the skill to follow the f3s / blog-writing-style pattern: a slim
SKILL.md (purpose, trigger, inputs, core workflow, reference index) plus three
focused reference files. Pure reorganization — all content preserved verbatim.
- references/search-and-sourcing.md: search endpoints, per-model queries,
source trust order, source endpoint URLs (workflow steps 2-3)
- references/benchmarks-catalog.md: per-model fields-to-capture table and the
2026 benchmark glossary
- references/output-and-analysis.md: comparison table structure, analysis
sections, anti-patterns, output format, output shape (workflow steps 4-7)
SKILL.md drops from 225 to 86 lines.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Break up the three most monolithic flat skills into a slim SKILL.md plus a
references/ directory of focused sub-files, matching the f3s and
agent-task-management structure. Pure reorganization: all content and meaning
preserved, only moved behind a reference index with resolving links.
- blog-writing-style (619->216): extract the Wikipedia-based 'Signs of AI
writing' deep catalog into references/signs-of-ai-writing.md.
- check-shopping-status (319->135): extract references/imap-scan-script.md
(steps 1-3 Python) and references/amazon-cdp.md (CDP login helper).
- photo-processing (232->155): extract references/auto-enhance.md (standalone
ImageMagick auto-enhance script).
Adds notes/skills-review-2026-06-19.md with the full inventory and the
follow-up task IDs (dq0, eq0, fq0) for the remaining oversized skills.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Add references/r-node-deploy.md describing the Rex nfs_mount_monitor
rollout to r0/r1/r2: group/root/parallelism-3 conventions, idempotent
file/on_change reload flow, and systemctl/journalctl verification, so
conf-project agents don't re-derive the rollout each task. Wire it into
SKILL.md's reference index.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Add a section on handling test suites that work but exceed the
go-test/mage/command timeout (e.g. mage integrationTest > 30m):
which focused subset to run (-short/-run/build tags, touched
packages), how to annotate that the full suite was intentionally
skipped, and the acceptance implications (focused subset is not
full verification). Cross-references the 8q0 missing-toolchain
section rather than restating it.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Document how agents should handle a local toolchain that cannot build
or test the project (e.g. missing libbpf bpf/bpf.h headers, missing
flutter/dart). Preflight the build/test path before trusting it, run
the smallest verifying subset that works, annotate the blocker via
ask annotate, and never claim full verification that did not run.
Scoped to MISSING/incomplete toolchains; slow/long-running test suites
are covered separately by task 9q0.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
vendored/upstream deps
A sub-agent patched a vendored dependency (libbpfgo) and the change had to
be reverted. When completing a task appears to require changing a vendored,
generated, or upstream dependency, that is a scope boundary: flag it as a
blocker (annotate + report) instead of patching the dep. Added to
2-start-task.md with a pointer from SKILL.md; cross-references the
complete-task commit hygiene and stalled-worker recovery rather than
restating them.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Extend the complete-task reference with a 'Commit only in-scope files'
section: at task start and before committing, run git status and split
dirty paths into in-scope vs out-of-scope (pre-existing/unrelated user
edits such as .gitignore/start.sh). Stage only in-scope files by explicit
path, never git add -A/. over a pre-existing dirty worktree, never commit
unrelated files, and record the in-scope/out-of-scope decision in an
ask annotate note. Cross-reference 6-recover-stalled-task.md (broken
partial WIP) to keep the two cases distinct and avoid duplication.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Multiple ior/player tasks had a worker stall or get interrupted mid-edit,
leaving broken, uncommitted partial file edits with no documented recovery
path. Add references/6-recover-stalled-task.md covering detect -> assess ->
revert -> resume: how to spot a partial/interrupted edit (dirty worktree,
broken build, half-applied diffs), assess dirty changes against the task
annotations (keep in-scope WIP vs. discard broken edits vs. preserve unrelated
user changes), revert cleanly with scoped git checkout/stash WITHOUT destroying
unrelated work, then re-read annotations, resume from the last known-good
point, and annotate the recovery.
Wire the new reference into SKILL.md's 'When to load what' table and cross-ref
it from 2-start-task.md when picking up an already-started task. Commit
discipline is cross-referenced to 3-complete-task.md rather than duplicated.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
invocation contract
Agents repeatedly mistook `ask` for a natural-language interface and ran
things like `ask agent-task-management ...` or `ask <free text>`, forcing
orchestrators to paste the correction into every task description.
- Add a prominent STOP blockquote at the very top of SKILL.md stating that
`ask` is a CLI with fixed subcommands and that agent-task-management is a
skill name, never an `ask` subcommand.
- Add an explicit 'Invocation contract' section listing the only valid form
(`ask <subcommand> [args]`) and the real subcommands, derived from existing
content; consolidate the valid/invalid examples there (DRY) with inline
reasons on the invalid ones.
- Mirror the same STOP warning at the top of references/00-context.md and
tighten its rule to point at the contract.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Documents the fishfinger/blowfish ProxyJump path to reach f-hosts
(paul + doas), r-VMs/rocky (root), and Pis from outside the LAN.
Notes the f3 WireGuard routing caveat and tcsh shell pitfall.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
|
Amp-Thread-ID: https://ampcode.com/threads/T-019ed908-ca57-7251-8ff1-ce10c8c3a787
Co-authored-by: Amp <amp@ampcode.com>
|
|
Amp-Thread-ID: https://ampcode.com/threads/T-019ed908-ca57-7251-8ff1-ce10c8c3a787
Co-authored-by: Amp <amp@ampcode.com>
|
