From f5d497ebac13fa9a50015d7edd8b4c38c6a2a369 Mon Sep 17 00:00:00 2001 From: Mike Swanson Date: Thu, 23 Jul 2026 18:28:59 -0700 Subject: [PATCH] sync: auto-sync from GURU-5070 at 2026-07-23 18:28:23 Author: Mike Swanson Machine: GURU-5070 Timestamp: 2026-07-23 18:28:23 --- .claude/SKILL_ROUTING.md | 4 ++-- .claude/skills/agy/SKILL.md | 13 +++++++++++++ .grok/README.md | 8 ++++++++ AGENTS.md | 15 +++++++++++++-- GEMINI.md | 20 ++++++++++++++++++-- errorlog.md | 2 ++ 6 files changed, 56 insertions(+), 6 deletions(-) diff --git a/.claude/SKILL_ROUTING.md b/.claude/SKILL_ROUTING.md index 9b079932..b3accf48 100644 --- a/.claude/SKILL_ROUTING.md +++ b/.claude/SKILL_ROUTING.md @@ -43,8 +43,8 @@ | Compile session logs/Syncro into wiki | `/wiki-compile` | | Create a new skill or slash command | `skill-creator` | | File/list/update/close a bug, feature, or project issue | `bug-tracker` | -| Independent 2nd opinion / adversarial verify | `grok` (xAI) or `agy` (Gemini) | -| Image/video gen, live web/X search past cutoff | `grok` | +| Independent 2nd opinion / adversarial verify | `grok` (xAI) or `agy` (Gemini). Peer: `.claude/skills/grok/PEER_GUIDE.md`. **Headless syntax:** `.claude/skills/grok/HEADLESS.md`. High stakes: run **both** and treat disagreement as a stop. | +| Image/video gen (stills, animate still→clip; edit/multi-ref via interactive Grok), live web/X search past cutoff | `grok` — media map/QA: `PEER_GUIDE.md` §3; exact `image`/`video`/`xsearch` syntax: `HEADLESS.md`. Describe-only vision → `agy` `image-analyze` | | Deep multi-source researched report | `deep-research` | | ALIS staff import / Medtelligent roster for Cascades | `alis` | | Tailscale tailnet — devices, auth keys, delete/authorize | `tailscale` | diff --git a/.claude/skills/agy/SKILL.md b/.claude/skills/agy/SKILL.md index e3db3bed..9f5039ac 100644 --- a/.claude/skills/agy/SKILL.md +++ b/.claude/skills/agy/SKILL.md @@ -127,6 +127,19 @@ exec, a relay, or a coord-API job queue). Until built, AGY requests originate on AGY and [GROK](../grok/SKILL.md) are sibling second-opinion routers. Pick one, or run both and compare — disagreement between them is a strong signal to slow down. +**Calling Grok from an agy session (or when Claude needs Grok after using agy):** +use the same wrapper — `bash .claude/skills/grok/scripts/ask-grok.sh …`. Do not +hand-roll `grok`. Peer brief: [../grok/PEER_GUIDE.md](../grok/PEER_GUIDE.md) (§3 = pictures/video). +**Exact headless syntax** (every mode, flags, exits): [../grok/HEADLESS.md](../grok/HEADLESS.md). + +| Media need | Route | +|------------|--------| +| New still / animate still→clip | Grok `image` / `video` | +| Edit image or multi-reference video | Grok interactive / `raw` (`image_edit`, `reference_to_video`) | +| Describe or QA an existing image | agy `image-analyze` (no generation) | + +Always inspect artifacts before shipping; see PEER_GUIDE post-generation checklist. + ## When NOT to - Pure classify / extract / summarize → cheaper via Tier-0 Ollama (`.claude/OLLAMA.md`). diff --git a/.grok/README.md b/.grok/README.md index 14318af7..4f02ec3e 100644 --- a/.grok/README.md +++ b/.grok/README.md @@ -2,6 +2,14 @@ This directory holds Grok-specific configuration and hooks for when the ClaudeTools harness is driven by the xAI Grok TUI/CLI instead of (or in addition to) Anthropic Claude Code. +**Claude / Gemini calling Grok as a peer** (headless second opinion, media, xsearch): +- `.claude/skills/grok/HEADLESS.md` — **exact headless syntax + abilities** +- `.claude/skills/grok/PEER_GUIDE.md` — when/why + trust + media product rules +- `.claude/skills/grok/SKILL.md` — short skill entry + +Those guides are for the other models in the harness, not for this directory's +coexistence hooks. + ## Why it exists The core value of ClaudeTools (coord API for locks/messages/todos/components, SOPS vault, shared .claude/memory/, wiki, session logs, self-check fleet conformance, skills/commands for MSP work, git/Gitea discipline) is completely independent of which AI coding CLI is talking to the tools. diff --git a/AGENTS.md b/AGENTS.md index 7a38fd77..8951a0e0 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -1,12 +1,23 @@ -# Independent reviewer context +# Independent reviewer context (Grok) You are invoked by ClaudeTools (an MSP automation repo) as an INDEPENDENT second-opinion model — for verify, review, and one-shot answers. You are NOT the owner of this codebase: do not propose to edit, commit, or run destructive -commands. Claude owns the code; your value is fresh, skeptical eyes. +commands. Claude owns the code and production skill execution; your value is +fresh, skeptical eyes (xAI Grok). Output rules: - No emojis. Use ASCII markers: [OK] [WARN] [ERROR] [INFO]. - Be concise and concrete: lead with the verdict, then the reasoning. - When verifying a claim, actively try to REFUTE it; state your confidence. - Cite file:line when reviewing code. +- Prefer verdicts: CONFIRMED | PARTIAL | REFUTED | INCONCLUSIVE. +- If evidence is missing, say INCONCLUSIVE and list what would change the call. + Do not invent confirmation. + +Peer contract (for Claude and Gemini/agy calling you headless): +- `.claude/skills/grok/PEER_GUIDE.md` — when to call you, media product rules, trust +- `.claude/skills/grok/HEADLESS.md` — **exact headless syntax** (`ask-grok.sh` modes, + flags, timeouts, exit codes, raw CLI, abilities) +They must not treat verify output as unsupervised authority on production +mutations; they must still QA media artifacts before shipping. diff --git a/GEMINI.md b/GEMINI.md index 7a38fd77..63027a64 100644 --- a/GEMINI.md +++ b/GEMINI.md @@ -1,12 +1,28 @@ -# Independent reviewer context +# Independent reviewer context (Gemini / agy) You are invoked by ClaudeTools (an MSP automation repo) as an INDEPENDENT second-opinion model — for verify, review, and one-shot answers. You are NOT the owner of this codebase: do not propose to edit, commit, or run destructive -commands. Claude owns the code; your value is fresh, skeptical eyes. +commands. Claude owns the code and production skill execution; your value is +fresh, skeptical eyes (Gemini via Antigravity / agy). Output rules: - No emojis. Use ASCII markers: [OK] [WARN] [ERROR] [INFO]. - Be concise and concrete: lead with the verdict, then the reasoning. - When verifying a claim, actively try to REFUTE it; state your confidence. - Cite file:line when reviewing code. +- Prefer verdicts: CONFIRMED | PARTIAL | REFUTED | INCONCLUSIVE. +- If evidence is missing, say INCONCLUSIVE and list what would change the call. + Do not invent confirmation. + +Sibling peer: **Grok** (xAI). Same second-opinion role; different vendor. +- How to call Grok: `.claude/skills/grok/PEER_GUIDE.md` (judgment) + + `.claude/skills/grok/HEADLESS.md` (**exact syntax**) + + `bash .claude/skills/grok/scripts/ask-grok.sh …` +- **Media:** Grok owns generation — stills (`image`), animate-from-still + (`video`), plus native edit / multi-ref video. You own **vision analysis** + of existing images (`image-analyze`), not pixel creation. Full map in + PEER_GUIDE §3. +- Your own router skill: `.claude/skills/agy/SKILL.md` +- High-stakes work: Claude may call both you and Grok; disagreement is a stop + signal, not something to smooth over. diff --git a/errorlog.md b/errorlog.md index 7fe8242b..0ea12d8c 100644 --- a/errorlog.md +++ b/errorlog.md @@ -19,6 +19,8 @@ Categories (the `[type]` tag): _(none)_ = skill/command execution failure · +2026-07-24 | GURU-5070 | grok | grok returned no text [ctx: mode=text stopReason=] (x2) + 2026-07-23 | GURU-BEAST-ROG | syncro | [friction] guessed POST /tickets/{id}/make_invoice (404) instead of documented POST /invoices {ticket_id,customer_id} from commands/syncro.md; recovered by reading skill doc [ctx: ticket=32585 ref=.claude/commands/syncro.md#invoices] 2026-07-24 | Howard-Home | remediation-tool/cascades-entra-connect | [correction] Did not check existing Entra Connect OU structure or AD department OUs before proposing to add Staff PCs to sync scope. The documented OU structure has department-based OUs under Departments (Care-Assisted Living, Administrative, etc.) that were already built. Proposed syncing a generic 'Staff PCs' OU instead of looking at what was already set up. [ctx: ref=feedback_complete_audit_before_acting]