メニュー
Operating patterns for Cognify knowledge graph ingestion — model selection, key sourcing, batch strategy, Drive/Typeform extraction, and priority source ranking for the ramzy tenant.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Healthchecks, SSH access, model switching, and systemd debugging for the Appie fleet. Covers Appie-2 (Hetzner Hermes) and common failure modes.
Direct Google Calendar + Gmail + Contacts REST API access via Python. Use when gws CLI is broken (token cache decrypt failure) or google_api.py wrapper is unreliable. Companion to the bundled google-workspace skill.
State-of-the-art text-to-image generation with Stable Diffusion models via HuggingFace Diffusers. Use when generating images from text prompts, performing image-to-image translation, inpainting, or building custom diffusion pipelines.
Use 21st.dev Agent Elements — 25 shadcn/ui components purpose-built for AI agent UIs. Tool cards (Bash, Edit, Search, Plan, Subagent, MCP), chat, input, streaming. Open source, shadcn-registry install, no runtime lock-in.
Use v0 Design Systems 2.0, Bolt Design System Agents, or Lovable to generate production UI from a client's real design system components — not generic Tailwind. Use when building a client website where brand consistency matters.
Umbrella workflow for delegating coding work to local agent CLIs such as Claude Code, Codex, and OpenCode.
| name | cognify-ops |
| description | Operating patterns for Cognify knowledge graph ingestion — model selection, key sourcing, batch strategy, Drive/Typeform extraction, and priority source ranking for the ramzy tenant. |
Default to openai/gpt-4o-mini for entity extraction. DeepSeek V4 Flash hangs on
multi-chunk documents. gpt-4o-mini completes the same document in 25-40s.
Set: COGNIFY_LLM_MODEL=openai/gpt-4o-mini in /root/cognify/.env
Cognify's .env key goes stale. When extraction fails with "No LLM key", source the
working key from Hermes' env. Use Python to write .env, never shell heredocs —
special chars in API keys cause silent corruption.
import subprocess
r = subprocess.run(["grep", "-v", "^#", "/root/.hermes/.env"], capture_output=True, text=True)
for line in r.stdout.split('\n'):
if 'OPENROUTER_API_KEY' in line and '=' in line:
key = line.split('=', 1)[1].strip().strip('"').strip("'")
with open("/root/cognify/.env", "w") as f:
f.write(f"OPENROUTER_API_KEY={key}\n")
f.write("COGNIFY_LLM_BASE=https://openrouter.ai/api/v1\n")
f.write("COGNIFY_LLM_MODEL=openai/gpt-4o-mini\n")
f.write("COGNIFY_LLM_KEYENV=OPENROUTER_API_KEY\n")
f.write("COGNIFY_BACKEND=local\n")
f.write("COGNIFY_EXTRACT_WORKERS=4\n")
break
Combine many small files into batches before ingesting. Use ingest-dir with workers:
cd /root/cognify && set -a && . ./.env && set +a && export COGNIFY_DATA_DIR=/root/.cognify
./.venv/bin/cognify --tenant ramzy ingest-dir /path/to/dir --workers 4
Performance (gpt-4o-mini, 4 workers): ~17K chars/25s, ~35K chars/38s, ~42K chars/42s.
gws drive files export returns 401 even when gws drive files list works.
Workaround: direct Drive API with access token from ~/.config/gws/tokens.json.
Refresh expired tokens: python3 /root/.hermes/scripts/refresh_google_token.py
Contact blocks have nested sub-fields. Recursively parse form definition for ref→label map
before extracting response answers. See references/typeform-extraction.md.
N4LbJCHT) — structured lead intake~/ibrahim-lowticket/*.txt) — client stories for voice