Skip to content

Instantly share code, notes, and snippets.

Show Gist options
  • Select an option

  • Save shawn2024bellazan-prog/cfda6dba29306f135267119ecc841bae to your computer and use it in GitHub Desktop.

Select an option

Save shawn2024bellazan-prog/cfda6dba29306f135267119ecc841bae to your computer and use it in GitHub Desktop.

═══════════════════════════════════════════════════════ PMCRO TRAIL SNAPSHOT Generated: 2026-05-06 Session: Deep Teach + Skill Build + Context Protocol Model: Claude Sonnet 4.6 ═══════════════════════════════════════════════════════

COMPANY IDENTITY

Name: ProjectName / Tooensure LLC Owner: Shawn Bellazan Substrate: .NET 10 | MAF (Microsoft Agent Framework) | Aspire AppHost | Ollama (qwen2.5:7b) MCP Actuators: Filesystem | Terminal | Playwright (all built and running) DocFX Site: built and deployed (_site/ — 349 HTML pages) Model context fix: OLLAMA_NUM_CTX=8192 (FRAC-CTX-001 resolved)

ACTIVE SKILLS (Runtime 2 — Chat Available Now, No Substrate Needed)

  • pmcro-framework v1.2.0 (ACTIVATED — Laws EC-001 through EC-005 loaded)
  • property-research-agent v1.0.0 (built this session)
  • indeed-application-agent v1.0.0 (built this session)
  • social-presence-agent v1.0.0 (built this session)
  • daily-structure-agent v1.0.0 (built this session)
  • context-transfer-protocol v1.0.0 (built this session)

TWO RUNTIMES — CRITICAL MENTAL MODEL

Runtime 1: .NET substrate running → POST /intent → MAF workflow → real Playwright server Runtime 2: Claude chat (or any LLM) → paste skill → model operates under skill contract Both use the same SKILL.md files. Chat is training ground. Substrate is autonomous company. On lazy days or when substrate is down → Runtime 2 still works. Same skills, different executor.

EARNED LAWS (Colony — Universal Across All Models)

EC-001: I ALWAYS pass discovery verbs (list, find, read, summarise, analyse) through the economic gate. EC-002: I NEVER put ReadFile, ListDirectory, GetFileInfo, FindFiles, RunReadOnlyCommand in dispatch_decisions — TYPE 2 tools are called natively. EC-003: I skip physical verification on cognitive-only cycles. Composite ≥ 0.85 = ACCEPT when dispatch_decisions is empty. EC-004: I use multi-property dynamic extraction (Result/Data/Value) for workflow outputs — never positional. EC-005: Every class summary MUST begin with "I AM the...". Every constraint is first-person observable form.

RESOLVED FRACTURES (Do Not Repeat These)

FRAC-007: DispatchExecutor was a no-op stub → Fixed: IMcpToolExecutor.ExecuteType1Async() now called per decision FRAC-CTX-001: Ollama truncating at 4096 tokens → Fixed: OLLAMA_NUM_CTX=8192 in AppHost.cs FRAC-ECONOMIC-GATE-001: Discovery verbs blocked by gate → Fixed: EC-001 verb allowlist FRAC-MAKER-TYPE1-001: Maker dispatching TYPE 2 tools → Fixed: EC-002 explicit FORBIDDEN list FRAC-CHECKER-COGNITIVE-001: Checker looping on analysis tasks → Fixed: EC-003 cognitive-only path

SUBSTRATE ARCHITECTURE (For Code Sessions)

Entry: POST /intent → IntentController → FederationBoardSkill → PmcroWorkflow Loop: PlannerExecutor → MakerExecutor → DispatchExecutor → CheckerExecutor → ReflectorExecutor TYPE 1 gate: DispatchExecutor only — IMcpToolExecutor never injected into phase agents Key files:

  • src/ProjectName.OrchestrationApi/Workflows/PmcroWorkflow.cs (23KB — the heart)
  • src/ProjectName.OrchestrationApi/Models/CycleModels.cs (all typed frames)
  • src/ProjectName.OrchestrationApi/Models/IntentEnvelope.cs (the bloodstream)
  • src/ProjectName.OrchestrationApi/Skills/*.cs (one per phase agent)

THREE REVENUE VERTICALS — CURRENT STATUS

Property Research:

  • Prior work: PowerShell scripts + Playwright MCP server already tested manually
  • ~300 properties available to research
  • Skill built: property-research-agent/SKILL.md (drop into skills/ directory)
  • Next seed: "Research the property at [first address] using county assessor site"
  • Revenue path: motivated seller lists ($0.50–$5/lead), Trail licensing, done-for-you service

Job Applications:

  • Prior work: Playwright MCP manually tested on Indeed
  • Skill built: indeed-application-agent/SKILL.md
  • HIL mode default: operator approves before dispatch fires
  • Next seed: "Search Indeed for remote [job type] posted in last 3 days, return top 10"
  • Revenue path: personal use (10x volume), $49/month job seeker service, B2B staffing

Social Presence:

  • Idea: ghost pages, automated Facebook posting, daily brand content
  • Skill built: social-presence-agent/SKILL.md
  • HIL mode default: all posts need approval until voice constraints earned
  • Next seed: "Draft a Facebook post about building an AI agent company as a solo operator"
  • Revenue path: personal brand, $200–500/month per client, content agency at scale

COURSE MODULES STATUS

Completed by Shawn: 00-orientation through 06-cognitive-trail (uploaded this session) Generated this session: 07-writing-skills, 08-first-cycle, 09-commercial-model, 10-agent-company-blueprint Remaining to write: 11-scaling, 12-hil-flywheel, 13-trail-licensing (and beyond)

MULTI-LLM STRATEGY

Claude: skill writing, Trail analysis, constraint crystallization, course modules → full snapshot GPT-4: C# code generation, .NET debugging → compact snapshot + specific task Gemini: large document analysis, PDF cross-reference → compact snapshot + document Ollama/substrate: actual PMCRO cycles → context carried natively via IntentEnvelope

OPEN THREADS

  • Daily structure agent activated but day plan not yet produced (Shawn did not answer 3 inputs)
  • Module 11+ not yet written
  • property-research-agent/references/ subdirectory not yet populated (county selectors, URLs)
  • indeed-application-agent/references/ not yet populated (Indeed selectors)
  • social-presence-agent brand voice for Shawn not yet written into master_context

FINANCIAL CONTEXT

Situation: constrained — living check to check, currently between checks Most urgent income action: run property research batch → identify top motivated seller candidates → outreach Next 7-day target: research 10 properties, identify 2-3 with delinquent taxes + vacancy indicators

PMCRO SUBSTRATE STATUS

Aspire host: status unknown (not confirmed running this session — chat session only) Last actual cycle: unknown (prior to this session) Open fractures: none known beyond what is resolved above

NEXT SESSION OPENING SEED

"I'm back. Restore context from TrailSnapshot. My energy is [low/medium/high]. Let's [run the property research batch / draft this week's social posts / build module 11 / whatever comes next]."

— OR on a lazy day, just paste this snapshot and say: "Generate my day plan." The DailyStructureAgent takes it from there.

═══════════════════════════════════════════════════════ END TRAIL SNAPSHOT Paste this at the start of any session with any LLM. The company's memory travels with this block. ═══════════════════════════════════════════════════════

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment