While using the new 3-retry gemini path for live VPN research, two bugs surfaced:
- emit_or_fail checked auth_failed INSIDE the retry loop; a benign mid-run token-refresh line
matched the over-broad auth regex (bare login|credential|authenticat|oauth|401) and aborted the
retries with a false "auth error" - even though `gemini -p` auth tested fine. Moved auth-classify
to AFTER the retries (it only picks the final error message now) and tightened auth_failed to real
signatures (invalid_grant, not authenticated, login with google, token expired, ...).
- Added quota_exhausted() + a QUOTA FALLBACK: the pinned strong model (gemini-3.1-pro-preview) hit
"exhausted your capacity on this model" mid-session; emit_or_fail now retries once on the default
(lighter) model by stripping -m (separate quota). Validated: capped pro run -> fell back -> 2.9KB answer.
CT_THOUGHTS Thought 2 Resolution updated with both. (Search-bot reliability hardening continues.)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Mike's must-fix. Diagnosed from RAW output of failing queries (not guessed):
- grok xsearch = TIMEOUT: grok-4.20-multi-agent web_search runs past budget on multi-part queries
(286s/280s, rc=124, still searching - 183 thoughts, only progress-noise text); buffered json => total loss.
- gemini search = INTERMITTENT empty turn (a clean re-run gave a real 2.6KB answer in 122s); the wrapper
retried only once, so two empties in a row failed spuriously.
Fixes:
- ask-gemini.sh emit_or_fail: retry up to 3x with 3s/6s backoff (was 1).
- ask-grok.sh xsearch: --output-format streaming-json (salvage partials) + AUTO-FALLBACK to
ask-gemini.sh search when grok doesn't finish (rc!=0 or empty). Validated e2e: grok timed out
(rc=124) -> fell back -> gemini returned a real sourced answer (UniFi Teleport invite-link API).
grok's own multi-agent timeout is an xAI-side limitation; the fallback makes xsearch reliable regardless.
Docs: grok SKILL.md xsearch row + CT_THOUGHTS Thought 2 Resolution.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Mike's correction: web search (grok xsearch + gemini search) carries at least as much weight as
live API probing - the searches gave the real leads this session (connector proxy, teleport setting
path); blind endpoint-probing is "highly suspect" (mostly 404s). And the search bots MUST be properly
fixed - both returned empty repeatedly on UniFi research despite the same-day partial grok fix.
- docs/CT_THOUGHTS.md: Thought 2 (HIGH PRIORITY) - web-search reliability must-fix, with the observed
failures + a proper-fix investigation plan (capture failing-query JSON; max-turns/streaming-json/
retry; cross-fallback grok<->gemini; 5/5 acceptance).
- memory feedback_web_search_over_probing: lead with web search/docs; probe only to CONFIRM a
hypothesis, never as primary discovery. Reading our own config is fine; guessing paths is not.
- errorlog correction logged.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Earlier "no usable Teleport API" was wrong (probed /rest/teleport, /stat/teleport, /v1/teleport).
Gemini research + live verification: Teleport config lives at /api/s/<site>/rest/setting/teleport
(GET/PUT, also under /get/setting key 'teleport') - reachable via the connector. Brooklyn confirmed
enabled, subnet 192.168.1.1/24. Invite generate/revoke is reportedly POST /api/s/<site>/cmd/teleport
{"cmd":"generate-invite"|"revoke-invite"} (untested - it creates a live VPN access link; gate as a
write). Invites are WiFiman-app-only. Proxy path is /v1/connector/consoles/{id}/proxy/... (Gemini's
/v1/hosts/{id}/proxy form 404s). Doc updated.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
neighbor-collect.sh: add `--console <name> [--site <short>]` so the AP name/BSSID/IP map can come
from the cloud connector (/v1/connector/.../stat/device) instead of a UOS direct-login -- lets the
disable-analysis collector run against ANY console we have AP-VLAN reach to (the AP SSH harvest of
/proc/ui_neighbor is unchanged and still needs L3 reach). UOS path untouched. Validated against
Cascades via connector: source=CONNECTOR, built 77-mac + 450-bssid map for the 75 online APs.
This completes the hybrid (don't-lose-functionality): connector for airtime everywhere + neighbor-
collect (any source) for the SNR matrix -> NEIGHBOR_JSON -> optimize-radios disables on remote sites.
Documented (references/site-manager-api.md): the neighbor-collect --console flow, and the gateway
VPN/Teleport reach -- connector reaches /rest/networkconf (VPN servers: wireguard-server/openvpn-
server, site-to-site) read+writable in principle (gate writes like gw-control); Teleport has no
usable API (v1/ea/teleport 404, per-console /teleport 403).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Validated the cloud-connector analysis against a KNOWN entity (Cascades, normally UOS-Mongo).
The connector reaches the self-hosted "UOS Server" host; Cascades is its site `va6iba3v`.
Two fixes from the validation:
- rf-analyze.py: pass macs:[<all uap macs>] to /stat/report/*.ap. The UniFi report endpoint
returns only a small DEFAULT subset otherwise -- Cascades came back as 10 of 77 APs until the
MAC list was supplied. Now profiles all 75 (uaps with 2.4 radios), matching the UOS path.
- model-rank.sh / optimize-radios.sh: --console now accepts --site <name> (internal short name
from /api/self/sites) for multi-site controllers like the UOS Server (Cascades = va6iba3v).
Result lines up with the known UOS-Mongo figures: 75 APs, 2.4GHz util 65-90% / interf 53-78% /
~1 client each, all power-down, 0 disables (roam graph absent via connector -> same coverage-safe
degradation; disables still need NEIGHBOR_JSON). Apples-to-apples confirmed.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Both analyses now accept `--console "<name>"` and run against the UniFi cloud connector
instead of the UOS Mongo server, so RF airtime tuning works on standalone/non-UOS consoles
(e.g. Brooklyn/Skybar). The UOS Mongo path is unchanged.
- New shared analyzer scripts/rf-analyze.py: pulls per-AP/band airtime history via the
connector POST /stat/report/hourly.ap (SAME schema as ace_stat.stat_hourly) + /stat/device
for names/zones, derives cu_interf = cu_total - cu_self_rx - cu_self_tx, and runs the SAME
model-rank ranking and optimize-radios greedy power-down/disable logic (ported faithfully).
- Roam graph (/stat/event) is usually empty on small/stationary sites -> graceful degrade:
model-rank ranks by airtime pressure; optimize-radios returns power-down candidates + 0
disables (coverage-safe). NEIGHBOR_JSON (SNR matrix) still enables disables, as on UOS.
- model-rank.sh / optimize-radios.sh: added the `--console` route (resolves the key from
vault services/unifi-site-manager, execs rf-analyze.py). Validated on Brooklyn/Skybar:
2.4GHz saturated (Yoga AP cu 63%/interf 55%), 5GHz idle (1-5%) - the expected pain-band split.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
New backend reaching ANY of the ~36 ACG UniFi consoles remotely via api.ui.com with the
account key (vault services/unifi-site-manager) - no UOS server, no LAN/VPN. Mapped the API
surface empirically (key live), corroborated by grok+gemini web search:
- Tier 1 (Site Manager): fleet/devices/sites/isp commands - inventory, site health (counts,
IPS, ISP/ASN), and WAN/ISP time-series (latency/throughput/downtime).
- Tier 2 (CLOUD CONNECTOR -> console LOCAL Network API = UOS PARITY): the `net` command proxies
/v1/connector/consoles/<id>/proxy/network/api/s/<site>/stat/{device,sta}, returning the SAME
ace_stat depth as the UOS Mongo path - per-radio cu_total airtime/channel/bw/tx_power/num_sta/
satisfaction and per-client rssi/signal/noise/satisfaction/rates. Verified live on Brooklyn/
Skybar (standalone UDM, WAN-firewalled): `net brooklyn radios` + `net brooklyn clients` work.
This achieves parity with (and broader coverage than) the UOS server for non-UOS consoles.
Added references/site-manager-api.md (full catalog + 3 tiers), a Plane 3 note in SKILL.md, and
updated the reference memory. Read-only; POST actions (device restart, client block) exist, not wired.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Radio Show Promotion:
- Add LIVE badge to header phone with subtle pulse animation
- Add radio promo bar under hero (show name, time, call-in number)
- Radio ticker at bottom remains from Phase 3A
3-Step Visual Funnel:
- New funnel-steps section after Trust
- Progression: 1. Build estimate → 2. Talk it through → 3. Month-to-month start
- Interactive number badges with hover lift + color fill
- Mobile responsive (stacks vertically on small screens)
Strengthen Calculator CTAs:
- Pricing teaser: Make 'Build your exact price' primary button (was more-link)
- All 6 service cards: Add inline 'See what this costs →' calculator links
- Guides prospect directly to calculator from any service mention
Increase Vertical Rhythm:
- Section padding: clamp(2.75rem, 5.5vw, 4.25rem) for breathing room
- Service list: +1.5x base margin-bottom after dense grid
- Dispatch grid: +1.5x base margin-bottom after blog cards
All changes deployed to ww9.azcomputerguru.com and verified live.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Per Mike: don't keep the full archive tracked. Trimmed the brand kit from 34M -> 14M,
keeping only forward-relevant assets: the StyleGuide PDF, current logos
(transwhite/flatwhite) + vector master (guru-vector.eps, moved out of Old/), Guru
Icons, the Lato family (+ OFL + README), Colors reference, social avatar, and the
2025 letterhead (png + docx). Removed legacy Old/ marks, business cards, raster
flat-icon sheets, the superseded 2021 letterhead, and generic stock images.
(Original archive remains recoverable from git history.)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Imported the canonical "Branding Stuff" archive (2026-06-15) into the website
project at projects/acg-website-showcase/brand-kit/: logos (transwhite/flatwhite/
vector EPS), full Lato font family (+ SIL OFL license), Guru Icons (16-512px),
brand Colors reference, social avatar, letterhead (2021/2025), business cards,
flat icon sheets, and stock images. 64 files. Added a README documenting contents
+ licensing (Lato = OFL; verify stock-image licenses before public reuse).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- apply-wlan.sh: wlan_bands token was "6e" but this controller stores "6g" (verified live on Cascades
Guest SSID) -> setting 6 GHz membership would have failed. Fixed band values + option names (5g6g/6g/all).
- Cascades 2.4 runbook: folded in Phase 5 (5 GHz: width 80->40 on 76 radios; channel plan with the
DFS decision flagged -- DFS empirically clean here, so including clean-DFS gives ~20 channels vs ~5
non-DFS-only for 77 APs) and Phase 6 (6 GHz: root cause = production SSID CSCNet not on 6 GHz [bands
2g,5g only]; add 6g + enable bss-transition; band-steering already on). Per Howard.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Updated aria-label and title attributes on skin toggle button to reflect
current 3-skin configuration: Bold / Midnight / Verdigris
Deployed to ww9.azcomputerguru.com
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- CSS: Bold tokens now at :root (bone/near-black palette, signal orange)
- JS: skin switcher cycles Bold/Midnight/Verdigris (default: bold)
- HTML: all 7 pages default to bold skin
- Docs: README + DESIGN.md rewritten for Bold design language
- Deployment: live at ww9.azcomputerguru.com (IX hosting, grey-cloud DNS)
Paper/Sonoran Ledger skin dropped per user direction. Single-page ledger
version in root retained as archived reference pattern.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Upstream description ("You MUST use this before any creative work...") would
auto-fire the brainstorming skill on routine feature/code work. Rewrote the
frontmatter description to invoke ONLY when the user explicitly asks to
brainstorm/design. Methodology body (incl. HARD-GATE) unchanged. Noted in SOURCE.md.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
From the ComposioHQ/awesome-claude-skills list. Checked licenses BEFORE copying:
- threat-hunting-with-sigma-rules: repo is gone (GitHub 404) -- not harvested.
- forensics (mhattingpete): repo restructured, those skills no longer exist -- not harvested.
- pdf / mcp-builder (Anthropic official): LICENSE.txt FORBIDS copying out of the
Service / derivatives / redistribution -- NOT harvestable into this repo (install via
the official Claude Code marketplace instead if wanted).
- obra/superpowers: MIT -> the only legally harvestable set; imported with attribution.
Imported (each with its own MIT LICENSE copy + SOURCE.md provenance, commit a21956e48c13,
ASCII-normalized to house style, no emojis):
- using-git-worktrees
- test-driven-development (+ testing-anti-patterns.md)
- root-cause-tracing (+ find-polluter.sh helper, emojis -> ASCII markers)
- brainstorming (methodology only; upstream visual websocket server intentionally omitted)
Faithful imports -- content not reworded beyond ASCII typography/emoji normalization.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Audited the Ollama reference (no wrapper script — it's the OLLAMA.md doc + inline
HTTP-API call pattern) against the live server (Ollama 0.30.8 on GURU-5070):
- /api/chat + think:false + res['message']['content'] confirmed working (clean
output, no thinking leak) -- the core call pattern is correct.
- All referenced models exist on the server (qwen3:8b, qwen3.6:latest, qwen3:14b,
codestral:22b, nomic-embed-text).
Real bug found + fixed: the "Preferred one-liner" auto-detected the endpoint with
`urlopen(...)` used as a truthiness test. urlopen RAISES URLError on a down host
(proven), so the ternary's fallback branch was dead code -- it crashed on a down
localhost instead of failing over to Beast, and it did a per-call probe that
contradicts the doc's own "read endpoint from identity.json, no probe" rule 30 lines
above. Replaced with the identity.json endpoint+model pattern (also swaps the
hardcoded qwen3:14b for the per-machine prose_model). Validated verbatim end-to-end.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
