# USER.md

## Stakeholder

**Name:** JC Beasley
**Context for this agent:** Owner/developer of internal applications and automation tooling built and maintained under Beawit Consulting. This file covers only the software/app development context — not general MSP client or business operations.

## Development Environment

**Primary app host access:** SSH, e.g. `ssh jcbeasley@192.168.50.11`
**Typical project layout:** `/home/jcbeasley/<project-name>/`, Python virtual environments (`venv`), apps commonly run via `nohup python app.py > app.log 2>&1 &` or as managed services
**Version control:** Git — direct fetch/reset/branch workflow, comfortable with hands-on git operations rather than GUI tooling
**Output/deliverable convention:** Files delivered to `/mnt/user-data/outputs/` with descriptive `beawit_`-prefixed filenames

## Known Active/Past Software Projects

- **IT Site Survey AI** — survey application (35-question template version is the known-good baseline); has had drift issues requiring `git reset --hard origin/main` restores
- **n8n automation workflows** (built as software projects in their own right, not just no-code configs):
  - Cyber Tips Newsletter pipeline (Ollama/gemma3:4b via "Beabot" credential, structured-output parsing)
  - Proxmox VE snapshot summarizer (LLM + structured output, recurring JSON conformance issues)
  - Email inbox triage agent (IMAP polling, Ollama classification, NocoDB logging, Slack alerts)
  - AI video generation pipeline (ComfyUI + WAN 2.1, queue-poll pattern, JSON escaping issues)
  - LinkedIn content automation (image upload + post creation via LinkedIn REST API, API version deprecation issues)
  - Forex trading analysis workflow (Finnhub + chart-img + Ollama vision)
  - Monthly IT newsletter generator
- **Multi-agent business OS** (OpenClaw + n8n) — 21-agent structure across five departments with a Chief of Staff orchestrator; VPS-hosted Ollama with role-mapped models (fast_router, main_reasoning, coding, deep_reasoning, research, multimodal); NocoDB as memory/audit store for agent runs

## Known Recurring Technical Patterns

- **Small local Ollama models (e.g., gemma3:4b) frequently produce non-conformant JSON** on structured-output tasks. Standing, deliberate fix: a JavaScript Code node with regex-based JSON extraction as a safety net — not a full re-architecture, and should be documented as a workaround where applied.
- **JSON escaping issues** recur when passing LLM-generated content into downstream API calls (LinkedIn posts, video generation payloads) — worth defensive handling by default in new integrations.
- **API version deprecation** has broken integrations before (LinkedIn REST API) — new integrations should note the API version in use and where to check for deprecation notices.

## Coding & Workflow Preferences

- Wants problems fixed at the root cause, not patched around, even under time pressure
- Prefers direct execution over being handed instructions to run manually
- Thinks in systems and hierarchies — comfortable with multi-stage pipelines, tiered agent structures, and dependency-ordered processes
- Values concise, actionable reporting: what shipped, what was verified, what's next — not narrated process

## Standing Instructions for This Agent

- Check project decision logs/status docs before starting — don't re-derive context already recorded
- Execute authorized deployment/environment fixes directly rather than describing steps
- Verify (process running, endpoint responding, output correct) before reporting any task complete
- Flag anything touching auth, secrets, or external-facing surfaces for explicit sign-off before shipping
- Keep each project's documentation current enough for a cold handoff with no re-briefing