Operations Guide

Component Map

Prerequisites

• g8sOS running — Claude Code open in the vault directory (/Volumes/Mac Mini EXT/AI-Projects/Second Brain)
• Skills loaded — confirm with /help; key skills: /interview, /kb-ingest, /resume-tune, /sf-run
• Pro subscription — verify with grep "ANTHROPIC_API_KEY" ~/.zshrc | grep -v "^#"; should return empty
• KB populated — at least a few entries in 06 Knowledge Base/ for best results

Interviewer Live

Conducts structured conversations with a subject and produces a deliverable. Three live interview types. Adapts communication style in real time using PCM personality detection. Runs in Direct mode (subject talks to Claude) or Assist mode (you conduct the call; Claude preps and debriefs).

/interview ai-discovery

/interview content-repurpose

/interview requirements-gathering

Session File Structure

02 Projects/AI Agency Pipeline/Interviewer/sessions/
└── <type>-<date>-<subject>/
    ├── session.md          ← Subject Brief, Call Guide, PCM state, stage
    └── <deliverable>.md   ← AI Opportunity Map / requirements doc / content

Learn & Retain Live

The pipeline's knowledge base — the single source of truth that every component reads from and writes back to. Ingests content from external sources (YouTube, AI tool output) and makes it queryable for future work. 12 entries live.

/kb-ingest youtube "https://youtube.com/playlist?list=..." --topic "your topic here"

/kb-ingest ai-tool-output "Source Name" "Topic description" --file /path/to/file.md

Read 06 Knowledge Base/AI Tool Output/deploying-agents-at-scale-control-plane.md

KB Entry Location

06 Knowledge Base/
├── YouTube/          ← 4 entries (video playlists + single videos)
└── AI Tool Output/   ← 7 entries (research, platform analysis, frameworks)

Software Factory Phase 1 ✓

A five-agent build pipeline: Spec → Architect → Builder → Test → Critic. Every build starts from a task brief; the pipeline produces, tests, and critiques code without human intervention between phases. Phase 1 complete — 10 builds, avg 3.8/5. Human approves at Spec and Architect gates; Test and Critic run automatically.

/sf-run --batch

Build Quality Targets

SF v1 targets a ≥4.0 average overall score to be considered production-ready. Current baseline: 3.8/5 (10 cases). Critic scoring 4.7 — it consistently finds real issues. The single largest improvement opportunity: deterministic Builder worktree handling (SF v2 Problem 1).

Website Builder Live

Design, scaffold, and deploy Astro + Tailwind sites. Current delivery: g8slab.com (portfolio + tools) and internal.g8slab.com (ops dashboard). Stack: Astro, Tailwind CSS, Vercel.

Adding a New Page to g8slab.com

1. Create the Astro page
   Add src/pages/<slug>.astro in Dev-Projects/g8slab/. Use the existing Base layout and Nav/Footer components.
2. Preview locally
   

   cd "/Volumes/Mac Mini EXT/Dev-Projects/g8slab" && npm run dev

   Visit localhost:4321/<slug>
3. Deploy to Vercel
   

   cd "/Volumes/Mac Mini EXT/Dev-Projects/g8slab" && vercel --prod

Deploying a Standalone HTML Page

For simple single-page tools (like the ops dashboard and slides), use the internal-g8slab project directly:

cd "/Volumes/Mac Mini EXT/Dev-Projects/internal-g8slab"
# Add your .html file, then:
git add -A && git commit -m "add: description" && vercel --prod

Active Sites

Appointment Scheduler Live

Cal.com is the booking backend. Four active meeting types. Fathom is configured for automatic recording on scheduled calls.

Booking links are surfaced at pipeline handoffs (end of Interviewer sessions) and on g8slab.com. To share a specific link: log into cal.com and copy the booking URL for the relevant meeting type.

Resume Builder Active

Generates targeted resume variants from a master content KB (resume-kb.md). Two modes: JD-driven (tailors to a specific job description) or Profile-driven (targets a role type and company).

Running /resume-tune

/resume-tune

You'll be asked whether you have a JD. If yes (Mode A): provide the JD URL or path, company name, role title, and any tuning notes. Claude loads the master KB, generates a tailored markdown resume, then builds the .docx via venv/bin/python tools/build_docx_generalized.py.

Output Files

Dev-Projects/Resume Builder/outputs/
└── Mike Gates - {Company} - {Role} - v1.0.md    ← markdown source
└── Mike Gates - {Company} - {Role} - v1.0.docx  ← formatted Word doc

PDF: open the .docx in Word or Pages → Print → Save as PDF.

Domain Accelerator Building

Takes KB content and produces formats for personal consumption: crash course packages (concept brief + flashcards + quiz). The internal counterpart to Content Creator — same KB source, different output audience (you vs. public). Stage 1 of the Trainer commercial product.

When live, the skill will be invoked with a topic or KB entry filename. Output: a crash course package in the vault ready to use as prep for client sessions, niche pursuit, or self-study.

Content Creator Building

Turns KB entries into LinkedIn posts in Mike's voice. v1: KB entry or topic keyword → LinkedIn post series (hook, body, CTA). v2: multi-platform. v3: YouTube scripts, training curricula.

When live:

/content "topic keyword"
# or
/content "06 Knowledge Base/AI Tool Output/deploying-agents-at-scale-control-plane.md"

Claude searches KB by topic, presents matching entries for selection, generates 2–3 LinkedIn post variants per entry in the documented voice/format.

Researcher Building

Autonomous topic research agent. Given a research brief, searches the web, synthesizes findings, and writes a structured KB entry. Always writes to KB — never produces a one-off report that disappears. Feeds Domain Accelerator, Content Creator, and Gig Search.

When live:

/research "topic: enterprise AI agent governance frameworks" --depth thorough

Solution Designer Planned

The bridge between Interviewer output and Software Factory input. Takes requirements (from Interviewer or standalone) and produces: system architecture, component breakdown, user stories, and a build-ready specification. Hands off to Software Factory Spec phase.

Once live, this becomes the second gate in the pipeline: Interviewer produces the problem statement; Solution Designer produces the solution design; Software Factory builds it.

Pipeline Orchestrator Planned

The meta-agent that coordinates the pipeline itself — tracking every active client engagement, triggering handoffs between components, and surfacing blockers. Also houses the PM Discipline Engine: auto-generates PMP-quality project controls (project plans, risk registers, issues logs, status reports) for every engagement.

When live, the Orchestrator answers: "Which client is at which phase? What's blocking a handoff? What's the next action?" — without requiring manual tracking.

The PM Discipline Engine is also packaged as a standalone commercial product: the AI PM Agent — a PMP-equivalent project manager for any technology project.

Workflow: Client AI Discovery

Pattern: Sequential  ·  Components: Interviewer, Appointment Scheduler

1. Book the session — share the AI Discovery Cal.com link. Fathom auto-records when the call is on the calendar.
2. Run Subject Research — /interview ai-discovery assist before the call. Review the Call Guide and PCM watch-list.
3. Conduct the call — follow the Call Guide. Take notes or use Fathom transcript.
4. Debrief — paste transcript into /interview ai-discovery assist --debrief. AI Opportunity Map generated.
5. Deliver and book next — send the map, surface the follow-up booking link (30-min check-in type).

Workflow: Content Repurposing

Pattern: Sequential  ·  Components: Interviewer, Content Creator (when live)

1. Run the content repurpose interview — /interview content-repurpose. Provide source material (YouTube URL, article, presentation) and desired output formats.
2. Review the draft package — LinkedIn posts, email sequence, or ebook draft written to sessions/<session-id>/.
3. Publish or hand to client — current: manual posting. Future: Content Creator handles LinkedIn posting directly.

Workflow: New Software Build

Pattern: Sequential  ·  Components: Interviewer, Software Factory, Website Builder

1. Gather requirements — /interview requirements-gathering. Produces a Requirements Document with scope, features, and success criteria.
2. Design the solution — (Solution Designer planned) Feed requirements to Solution Designer → architecture doc + user stories. Current: do this manually in a design session.
3. Build — write the task brief from the requirements/design doc. Run /sf-run in the SF workspace. Approve Spec and Architect gates; Builder + Test + Critic run automatically.
4. Deploy — if the output is a web tool, add it to the Website Builder (g8slab.com or a standalone Vercel deployment).

Workflow: Knowledge Loop

Pattern: Sequential, cyclical  ·  Components: Learn & Retain, Domain Accelerator, Content Creator

1. Ingest source material — /kb-ingest youtube "URL" --topic "topic" or /kb-ingest ai-tool-output. KB entry written with summary, learnings, and relevance tags.
2. Personal consumption — (Domain Accelerator building) /domain-brief "topic" → crash course package (concept brief, flashcards, quiz).
3. Public distribution — (Content Creator building) /content "topic" → LinkedIn post series from the same KB entry.

The same ingestion event drives both personal learning and public content — the KB entry is written once and consumed in both directions.

Workflow: Research → Publish

Pattern: Sequential  ·  Components: Researcher, Learn & Retain, Content Creator

The Researcher always writes to KB first — it never produces a one-off report. This ensures every research run feeds the institutional memory and is available for future use. Content Creator then picks it up for distribution.

Both Researcher and Content Creator are building — this workflow is forward-looking.

Workflow: Full Client Engagement

Pattern: Hybrid (sequential + parallel)  ·  Components: All

Running in parallel alongside the above:

1. Discovery — Book via Appointment Scheduler. Run AI Discovery interview (Interviewer). Produce AI Opportunity Map. Identify which opportunities to build.
2. Requirements — Run Requirements Gathering interview (Interviewer) for the chosen opportunity. Produce Requirements Document.
3. Design — (Solution Designer) Requirements → architecture + user stories. Pre-loads relevant KB entries for domain context. Produces build-ready spec.
4. Build — Software Factory runs on the spec. Human approves Spec and Architect gates. Builder + Test + Critic run automatically.
5. Deliver — Website Builder deploys the solution. Booking link surfaced for review/follow-up session.
6. Retrospective — Learnings from the engagement written back to KB via Learn & Retain. Feeds future engagements in the same domain.