Claude Skills by grahama1970

Continuous Pi agent infrastructure health monitor. Checks the D-Bus daemon (embry-agent.service), scheduler daemon, scillm service, and scheduler job success rates. Detects crash-looping, stale jobs, and log bloat. Fires Discord alerts on failures. Designed to run on a scheduler.

Nightly cron roundtable review of every skill amended during the day. Discovers amended skills from git history, builds an equal-context packet from /project-state, /ops-workstation, and /brave-search, convenes a five-seat browser roundtable (webgpt, webclaude, webkimi, webgrok, webgemini) through /ask tau-dag, and stores the attributed synthesis in /memory (ArangoDB `project_roundtables` + lessons with Qdrant semantic sync) so the human and project agent can recall and discuss it.

Nightly self-hack orchestrator and continuous security monitoring for embry-os. Four-tier probe system: T0 deterministic (Semgrep, pip-audit, gitleaks), T0.5 threat intelligence (consume-feed, social-bridge, dogpile), T1 OWASP LLM Top 10 analysis, T2 Docker self-hack (containerized /hack against a twin), T3 cascade validation (GPT swarm classifiers + Brandon teacher). Composes /hack, /battle, /consume-feed, /social-bridge, /dogpile, /assistant, and /scillm.

Nightly skill and agent quality monitor that scans registered skills plus agents for best-practice violations, aspirational gaps, and trend drift. Produces per-target findings and an aggregate summarized report for tracking over time.

Continuous skill health monitoring with auto-drift correction. Runs every N minutes to ensure all projects have the newest skills. Never worry about skill drift again.

Continuous SPARTA quality monitor. Runs the 3-tier validation cascade alongside Stage 12 QRA generation — Tier 0 deterministic, Tier 1.5 trained GPT (via /assistant), Tier 2 Brandon (edge cases). Accumulates labels, auto-triggers /create-gpt training at 2000 labels, and tracks convergence.

Three-tier cascade taxonomy quality monitor. Assesses whether Heart/Mind taxonomy tags and collection_tags on memory documents are CORRECT (not just present). Uses T0 heuristic → T1.5 classifier/GPT → T2 Brandon teacher cascade. Accumulates training labels for autonomous quality assessment. Heart/Mind field conventions: - `mind` (8 SPARTA tactical tags: Detect/Evade/Exploit/Harden/Isolate/Model/Persist/Restore) lives on `sparta_qra` documents. - `heart` (6 emotional tags: anger/fear/joy/neut...

Audit, sync, and update the public grahama.co site and resume surfaces from README.md and RESUME.md. Report-only audit detects drift between README's curated projects/inventory and site/content.json, plus live-site health; apply regenerates content.json from the README; update cascades README/RESUME changes into site content, resume JSON, Markdown, PDF, DOCX, llms.txt, and ops-linkedin JSON handoffs. Use for "is the website current", "sync the site with the README", "website drift", "monitor ...

Nightly workstation health monitor. Enforces "no artifacts on NVMe" rule, detects cache bloat, checks drive health, and alerts on threshold breaches. Composes existing ops-* skills — never reimplements their logic. Uses /analytics + /create-figure for visual reports.

Self-improving music creation convergence loop for Horus persona. Generates audio from annotated lyrics + piano roll spec, analyzes with MIR tools, scores the delta between spec and output, re-quantizes prompts, and iterates until convergence. Thin orchestrator — ONE Python file of subprocess calls to existing skills.

Automated data quality audit for SPARTA Explorer. Runs deterministic httpx checks against the Express API (localhost:3001) to verify all 6 ArangoDB collections have complete, non-placeholder data. Produces structured JSON findings with severity levels.

Emergency operating procedure for suspected OAI quantization, context rate limiting, compression damage, or degraded reasoning. Use when the user says the model is quantized, context-limited, evasive, lazy, lying, losing thread, or asks to switch to GLM, Kimi, DeepSeek, or another non-OAI model because OAI is unreliable.

Use OpenCode serve child sessions as persistent, inspectable subagents. Use when proving or operating OpenCode subagents, child-session reuse, parent/child session maps, concurrent subagent fan-out, DAG-like task probes, SSE/event capture, or persistence across multiple turns without the full scillm transport/orchestrator overhead.

Manage ArangoDB operations including backups with automatic retention, health checks, embedding gap detection, duplicate detection, and integrity verification. Works with local or containerized ArangoDB.

Buzz relay client for posting skill notifications, rendering Buzz-ready messages, querying recent channel messages, and checking Buzz CLI connectivity. Use when a skill needs to notify Buzz, post a morning report, inspect Buzz configuration, or prepare a Buzz agent-facing handoff.

Operate the Calendly integration for grahama.co: inspect Calendly API connectivity, generate safe public scheduling metadata, manage the CALENDLY_PAT GitHub secret, and plan real capacity holds for scheduling. Use for "calendly api", "calendly embed", "calendly secret", "grahama.co booking", "ops-calendly", or "meeting slot capacity".

Diagnose local Chatterbox or Chatterbox-Turbo TTS services, Docker containers, health endpoints, render receipts, paralinguistic tag handling, voice delivery capability fields, and Chatterbox caller misuse. Use when the user asks for ops-chatterbox, Chatterbox health, Chatterbox Docker status, Chatterbox TTS smoke tests, or Chatterbox voice-delivery diagnostics.

Manage Chutes.ai resources, tracking subscription quota and account balance. Integrates with scheduler to pause operations when quota is reached.

Claude Code maintenance, diagnostics, and usage analytics. Check inotify limits, clean caches, diagnose startup failures, manage skills directories, and track token usage, costs, and rate limit patterns across projects.

Check codebases against compliance frameworks (SOC2, GDPR, HIPAA, PCI-DSS). Generates compliance reports with pass/fail status and remediation guidance.

Unified cost aggregator across all provider ops-* skills. Calls ops-chutes, ops-claude, ops-google, and ops-runpod, normalizes their JSON output into a single cost report with figure_data.

Query DARPA programs, opportunities, and BAAs. Searches RSS feeds and Grants.gov API for defense research funding opportunities across all DARPA technical offices.

TOS-compliant Discord notification monitor. Watches YOUR Discord server for security content forwarded by researchers, matches keywords, and pushes to create-paper/dogpile via webhooks. Persists to graph-memory for semantic search.

Safe Docker cleanup and compose stack management. Prune unused containers/images/volumes, redeploy stacks.

Manage the Embry Agent D-Bus daemon (embry-agent.service) for system-level Pi access

Build and validate Excalidraw whiteboard toolkits for interview strategy diagrams, including reusable finished elements and movable animation tokens that compile into create-svg scene/timeline input. Use when users ask for Excalidraw libraries, whiteboard strategy, animation toolkit items, or Excalidraw-to-create-svg handoff.

F-36 plant floor operations namespace. Subcommand router for machine testing, disposition, shift management, vendor review, and compliance monitoring. Designed for Paul Bevilaqua — gloved hands, standing, noisy environment.

Operate the Chrome Gemini sidebar as an iterative design collaborator and assessor when the user says Gemini sidebar, Ask Gemini side panel, paste into Gemini sidebar, copy Gemini response, Gemini design loop, or wants Gemini to turn the project agent's rough Surf/DOM observations into professional browser-page UI updates through a repeatable sidebar copy/paste loop.

Read and propose Google Calendar changes over OAuth for meeting-assistant scheduling. Use for "schedule a follow-up", "reschedule", "move the meeting", "push this to Friday", "what's on my calendar", "calendar oauth", or when live-evidence proposes a calendar action. Writes are PROPOSE-ONLY and require an explicit --confirm; nothing mutates a real calendar without it.

Compose Google Meet preparation and diagnosis with mandatory Live Evidence companionship. Use for meeting setup, testing Meet audio, blurry video, Jabra recovery, teleprompter placement, upcoming calendar meetings, or checking whether every manually opened Meet has its Live Evidence companion.

Manage Google Gemini API resources, tracking usage budget and model availability. Prevents surprise Google billing by enforcing cross-process call limits.

Use this skill when a project agent needs to spin up, inspect, communicate with, or remove Herdr-managed workstations for long-running multi-agent tasks. It provides a Typer CLI for dynamic Herdr workspaces, provider-specific agent panes, pane-to-pane notifications, semantic role state reporting, and bounded creator/reviewer loops with durable receipts.

Dry-run-first Hugging Face Hub operations: search models/datasets, inspect repos, list files, create repos, upload, snapshot, and validate/template model & dataset cards. Mutations require --execute; every receipt records an auth check without exposing token values.

LG webOS TV local control and audio gain-staging diagnostics over the same LAN WebSocket API the LG iOS app uses (ports 3000/3001). Pair with the TV, read sound output route and volume, change sound output (gated behind --execute), and run a combined gain-staging check that composes ops-wiim to compare TV-side output state against the amp's reported state. Use for "lg tv sound output", "pair with the LG TV", "is the TV sending quiet audio", "lg webos control", "ops-lgtv".

Prepare evidence-bounded LinkedIn profile updates, posts, image-post captions, comments, connection notes, messages, manual search plans, lead-research plans, and content reviews as local drafts with typed human-execution handoff packets; export editable JSON for Graham's own LinkedIn profile; prepare bounded Surf own-profile sync plans after explicit account-risk acknowledgement; and prepare bounded read-only contact graph capture plans for named opportunity contacts after explicit account-r...

Local LLM operations: health checks, cache management, Ollama model management, Docker service control, and performance monitoring. Probe Ollama/vLLM/SGLang endpoints, manage Docker-based Ollama models, monitor GPU usage and running models.

Natural-language front door that de-opaques the /memory stack — ArangoDB + Qdrant + the jina embedder + the memory daemon. Merges health, renders the topology as an ASCII chart, and prints a per-collection metrics matrix that flags collections which went stale, lost their Qdrant vector sync, hold no secondary index, are empty, or violate the "Arango stores no embedding arrays" contract. Also lists and triggers ArangoDB backups on the 12TB drive. Detection and read-only backups only — all Aran...

General-purpose NZBGeek search and SABnzbd download management with interview and task-monitor integration

Monitor, diagnose, and optimize OBS Studio on KDE Plasma Ubuntu. Use when the user says "obs health", "obs stats", "monitor obs", "obs dropped frames", "optimize obs", "obs stream control", "start/stop obs recording", "obs doctor", "obs websocket", or wants OBS buttons on the Stream Deck.

Pi Term (WezTerm fork) operations, interaction testing, and visual regression. Automated UI testing via xdotool, screenshot evidence, burst captures for animations, blind test integration via /test-lab, and design review automation.

Read-only Qdrant vector-store health detector. Lists collections, flags expected-but-missing collections, reports per-collection point counts, and probes end-to-end semantic recall through the memory daemon (dense score). Detection only — it performs NO writes. All Qdrant mutation and semantic sync stay in the memory repo per the memory contract. Use for "qdrant health", "check qdrant collections", "dense recall is zero", "vector store health", "ops-qdrant".

Diagnose local RealtimeSTT, Embry listener runtime, and Whisper-compatible ASR services, including health/readiness endpoints, Docker containers, OpenAI-style transcription endpoints, WebSocket ports, CUDA/cuDNN signals, and STT caller misuse. Use when the user asks for ops-realtimestt, RealtimeSTT health, Whisper ASR health, STT Docker status, voice listener readiness, or CUDA/cuDNN STT diagnosis.

Provision, manage, and terminate RunPod GPU instances for LLM training. Use when user says "spin up GPU", "create RunPod instance", "terminate pod", "check GPU status", "provision training server", or needs cloud GPU resources.

Query SAM.gov for federal contracts, entity registrations, exclusions, and opportunities. Uses official GSA APIs for government contractor intelligence and OSINT research.

Bundle design context (screenshots, DESIGN.md, references, prompt) for Google Stitch sessions. Prepares a context-bundle directory with current-state screenshots, design tokens, reference images, and a structured prompt ready for upload to the Stitch web UI.

Stream Deck control skill for agents. Provides ability to restart streamdeck daemons, execute button tasks, query status, and manage Stream Deck hardware through a persistent daemon interface. Works with the streamdeck Python package for CLI operations and agent-driven automation.

Read-only Terraform posture detector: binary/version, formatting, validate, and saved-plan summary for a target module directory. Detection only - no init with backend, no plan against live state, no apply. Use for "terraform health", "is this module valid", "summarize this plan file", "ops-terraform".

WiiM Amp local-network diagnostics for low sound output triage. Discovers the amp on the LAN, snapshots getStatusEx/getPlayerStatus/EQ/output-mode state over the local LinkPlay HTTP API, monitors state deltas while a fault is reproduced, and emits a low-volume triage report with heuristic findings (mute, low volume, EQ engaged, fixed-output, source gain-staging). Read-mostly: every mutating command (volume, EQ) is gated behind --execute. Use for "wiim low volume", "debug wiim amp", "wiim stat...

Workstation diagnostics, health monitoring, and maintenance. Knows this machine and can help troubleshoot, clean up, and maintain it.

Git worktrees as leases that expire, with unmerged work surfaced and archived recoverably instead of stranded. Use when worktrees have accumulated, when a ticket close is blocked by the worktree retention audit, when work may have been left unmerged in a worktree, or to schedule automatic reclamation. Also owns scoped landing: commit and push ONLY named files onto origin/main from any dirty checkout, worktree, or branch via `run.sh land`.