Create long-form technical blog content for tyingshoelaces.com. Narrative-driven, deeply personal, and technically precise.

Prerequisites

Load the Session Playbook first. Verify your session and profile before starting.

{ "command": "core playbooks show cli_session" }

IMPORTANT: NEVER start, stop, or restart services. The API is already running.

{ "command": "admin session show" }

{ "command": "admin agents list --enabled" }

If services are not running, ask the user to start them. Do not attempt to start them yourself.

MCP Tools

Tool Purpose
systemprompt_cli Execute CLI commands
systemprompt_help Get command documentation and playbooks

CRITICAL: Context Management

New content = New context. ALWAYS. See the Contexts Playbook for full context management commands.

WRONG: Reusing blog-general context for multiple posts RIGHT: blog-mcp-deep-dive, blog-ai-production-gap, blog-rag-architecture (one context per post)

Execution Model: One Tool Per Message

CRITICAL: Each step is a SEPARATE message. NEVER combine steps.

  1. Socratic Dialogue (no tools) - Agent interrogates your content goal
  2. Goal Confirmation - Agent confirms the plan
  3. Research - Agent calls research_content, waits for response
  4. Create - Agent calls content_create in next message
  5. Response - Agent summarises result
Step Action Separate Message?
Plan Socratic Dialogue YES - refine goals with agent
Research Call research_content YES - send message, wait for response
Create Call content_create YES - send message, wait for response
Image Generate featured image YES - send message, wait for response

RIGHT: "Research X" -> wait for response -> "Now create the post using research_id ABC" WRONG: Combining research and creation in one message or tool plan

Step 1: Create a FRESH Context (REQUIRED FIRST STEP)

Every new blog post MUST start with a fresh context. Name it descriptively for what you're creating.

{ "command": "core contexts list" }

{ "command": "core contexts new --name \"blog-[topic-slug]\"" }

Examples of good context names:

{ "command": "core contexts new --name \"blog-mcp-deep-dive\"" }
{ "command": "core contexts new --name \"blog-ai-production-gap\"" }
{ "command": "core contexts new --name \"blog-rag-architecture-lessons\"" }

ONLY continue an existing context if:

  • You're resuming work on the SAME blog post
  • You haven't created any other content since
  • The context was created specifically for this post
// MCP: systemprompt_cli - ONLY if continuing the SAME blog post
{ "command": "core contexts use blog-mcp-deep-dive" }

Once a context is active, all agents message commands automatically use it.

IMPORTANT: Due to a known issue, context may not persist automatically between messages. Always pass --context-id explicitly.

Step 2: Review Performance and Choose a Topic

Before creating new content, review what's performing well.

2.1 Review Content Performance

{ "command": "analytics content stats --since 30d" }

{ "command": "analytics content top --limit 10" }

2.2 Choose a Topic

Select a topic that:

  1. Aligns with your expertise (AI, production systems, Rust, DevOps)
  2. Is timely or provides a unique perspective
  3. You have a contrarian or unique angle on

Step 3: SOCRATIC DIALOGUE - Convince the Gatekeeper (MANDATORY)

CRITICAL: The blog agent is a GATEKEEPER. It will NOT research or create content until you have convinced it with deep, articulated reasoning about the content goal.

The agent uses the Socratic method to interrogate your content idea. Your job is to provide clear, specific, well-reasoned answers that demonstrate you've thought deeply about the value this content will provide.

3.1 Initiate the Dialogue

{ "command": "admin agents message blog -m \"I want to write a blog post about [topic]. Let's define the goal before researching.\" --blocking" }

The agent will ask probing questions. DO NOT skip this step or give vague answers.

3.2 Be Prepared to Answer These Questions

On Thesis:

  • What specific claim are you making that others would disagree with?
  • If you summarised this in one sentence, what would it be?
  • What's the 'so what' - why should anyone care?

On Evidence:

  • What production experience do you have that backs this?
  • What data, metrics, or concrete examples support this?
  • What failures taught you this lesson?

On Contrarian Angle:

  • What does the conventional wisdom say, and why is it wrong?
  • Who will this post make uncomfortable, and why?
  • What's the uncomfortable truth you're revealing?

On Reader Value:

  • What will a senior engineer do differently after reading this?
  • What decision will a CTO reconsider?
  • What misconception are you correcting?

3.3 Example Dialogue

WRONG (Vague - Agent Will Reject):

{ "command": "admin agents message blog -m \"Write about AI agents\" --blocking" }
// Agent will push back: "What specific claim? What evidence? What's the contrarian angle?"

RIGHT (Trend-Informed + Specific - Agent Will Accept):

{ "command": "admin agents message blog -m \"I found a trending topic I want to write about:\n\nTREND: 'Microsoft pauses Claude Code rollout' is trending on Reddit (84 upvotes, 44 comments)\nSOURCE: https://reddit.com/r/ClaudeAI/comments/...\n\nTHESIS: The Claude Code vs Copilot battle reveals a fundamental truth - terminal-native AI tools beat IDE plugins because they work WITH developer workflows, not against them.\n\nCONTRARIAN ANGLE: Everyone thinks AI coding assistants are about autocomplete. Wrong. The real battle is agentic (Claude Code) vs autocomplete (Copilot). Microsoft pausing Claude Code isn't about competition - it's about realising their paradigm is losing.\n\nEVIDENCE: I've used both extensively in production. Claude Code's agentic approach reduced my context-switching by 70%. Copilot's autocomplete interrupts flow. The terminal is where real work happens.\n\nREADER OUTCOME: Senior engineers will reconsider their IDE-centric workflow. CTOs will understand why terminal-native tools are winning. Teams will evaluate AI tools based on workflow integration, not feature lists.\n\nLet's refine this before researching.\" --blocking" }

3.4 Gate Check - Agent Must Confirm

The agent will NOT proceed to research until it can articulate:

  • The thesis in one sentence
  • The contrarian angle
  • The production evidence
  • The reader takeaway
  • The hook direction

Wait for the agent to confirm the goal:

CONFIRMED CONTENT GOAL:
- Thesis: [one sentence]
- Contrarian angle: [what conventional wisdom we're challenging]
- Evidence base: [production experience/data we'll draw from]
- Reader outcome: [what they'll think/do differently]
- Hook direction: [opening that grabs]

Only then will the agent proceed to research.

Step 4: Research (Separate Message)

Once the agent confirms the goal, tell it to proceed with research. The agent will call research_content and return a research_id.

{ "command": "admin agents message blog -m \"Proceed with research.\" --blocking --timeout 120" }

The agent will:

  • Research the topic (minimum 5 grounding reference links)
  • Return a research_id artifact
  • Wait for next instruction

Save the research_id from the response for Step 5.

Step 5: Create Content (Separate Message)

After research completes, tell the agent to create the content using the research_id.

{ "command": "admin agents message blog -m \"Create the blog post using research. Use these specifics:\n- slug: 'agent-mesh-architecture-production'\n- keywords: ['AI agents', 'agent mesh', 'production AI', 'microservices', 'orchestration']\n- Include the 3 specific failure cases we discussed\n- Add code examples showing the God Agent vs Mesh Agent patterns\n- Reference the 90% error reduction metric\n- British English, 3500+ words minimum\" --blocking --timeout 300" }

The agent will:

  • Create 3500-5000 words of technical content
  • Use British English (realise, optimise, colour)
  • Include clear structure with scannable headers
  • Back claims with production evidence
  • Include code examples and technical details
  • Honour the thesis and angle agreed in the dialogue

Ask the agent to generate a featured image for the post.

{ "command": "admin agents message blog -m \"Generate a featured image for this blog post. It should represent [the core technical concept] visually.\" --blocking --timeout 60" }

Step 7: Publish and Verify

Publish the content to make it live on the site.

IMPORTANT: Image optimization MUST run before publish. The blog detail template expects .webp images, but content is created with .png. Without optimization, featured images will 404.

See the Jobs Playbook for job management commands.

// MCP: systemprompt_cli - Run image optimization FIRST (converts PNG to WebP)
{ "command": "infra jobs run blog_image_optimization" }

// MCP: systemprompt_cli - Then run the publish job (prerenders, updates sitemap)
{ "command": "infra jobs run publish_content" }

// MCP: systemprompt_cli - Sync content from database to filesystem (REQUIRED for new content)
{ "command": "cloud sync local content --direction to-disk --source blog -y" }

// MCP: systemprompt_cli - Verify the content is accessible and published
{ "command": "core content verify mcp-architecture-deep-dive --source blog" }

// MCP: systemprompt_cli - Show the content details
{ "command": "core content show mcp-architecture-deep-dive --source blog" }

MCP Tools Used by Agent

The Blog agent has access to these MCP tools (you don't call these directly):

Tool Purpose
research_content Research topic using web search, returns research_id
content_create Create content from research, saves to database
generate_image Generate AI image for featured image
analytics_query Query performance data
content_store Store ideas and suggestions

Content Requirements

Blog posts have strict requirements:

Requirement Value
Length 2000-4000 words
Grounding links Minimum 5 reference links
Keywords Required for SEO
Featured image Required (placeholder used if not generated)
Language British English

Voice and Style

  • Lead with a compelling hook that challenges conventional wisdom
  • Use personal experience and production data to back claims
  • Technical depth with accessible explanations
  • Contrarian takes that provoke thought
  • Clear structure with scannable headers
  • Sardonic humour about industry hype

Example Session

// Step 1: Create context
{ "command": "core contexts new --name \"blog-claude-code-vs-copilot\"" }

// Step 2: Socratic dialogue with topic
{ "command": "admin agents message blog -m \"I want to write about:\n\nTOPIC: 'Terminal-native AI tools vs IDE plugins'\nTHESIS: Terminal-native AI tools beat IDE plugins because they work WITH developer workflows.\nCONTRARIAN ANGLE: The real battle is agentic vs autocomplete, not Claude vs Copilot.\nEVIDENCE: I've used both - Claude Code reduced context-switching 70%.\nREADER OUTCOME: Engineers reconsider IDE-centric workflows.\n\nLet's refine before creating.\" --blocking" }
// Agent responds with questions or confirms goal

// Step 4: Research (after goal confirmed - SEPARATE MESSAGE)
{ "command": "admin agents message blog -m \"Proceed with research.\" --blocking --timeout 120" }
// Agent calls research_content, returns research_id

// Step 5: Create (SEPARATE MESSAGE using research_id)
{ "command": "admin agents message blog -m \"Create the blog post using research. Use these specifics:\n- slug: 'terminal-native-ai-tools-winning'\n- keywords: ['Claude Code', 'Copilot', 'AI coding', 'terminal', 'agentic AI']\n- Include code examples comparing workflows\n- Reference the Reddit trend\n- British English, 3500+ words minimum\" --blocking --timeout 300" }
// Agent calls content_create with research_id

// Step 6: Image
{ "command": "admin agents message blog -m \"Generate a featured image showing terminal vs IDE, simple vs complex\" --blocking --timeout 60" }

// Step 7: Optimize images, Publish, and Sync to filesystem
{ "command": "infra jobs run blog_image_optimization" }
{ "command": "infra jobs run publish_content" }
{ "command": "cloud sync local content --direction to-disk --source blog -y" }
{ "command": "core content show terminal-native-ai-tools-winning --source blog" }

Blog Post Structure

Typical structure for technical blog posts:

# [Title - Hook that challenges]

[Opening paragraph - personal stake, why this matters]

## The Conventional Wisdom

[What everyone believes/says]

## What Actually Happens

[The reality from production experience]

## Why This Matters

[Technical deep-dive with code examples]

## The Data

[Production metrics, statistics, evidence]

## What To Do About It

[Actionable insights]

## Conclusion

[Provocative closing thought]

Step 8: Validate Filesystem and Refine Content (FINAL QUALITY GATE)

After publishing, validate the content has been synced to the filesystem and perform a final quality review.

8.1 Validate Filesystem Sync

IMPORTANT: Content created by agents lives in the database. You MUST sync to filesystem before validating:

// MCP: systemprompt_cli - Sync database content to filesystem
{ "command": "cloud sync local content --direction to-disk --source blog -y" }

// MCP: systemprompt_cli - Verify sync status
{ "command": "core content status --source blog" }

Expected frontmatter fields:

---
title: "Blog Post Title"
description: "Brief description"
author: "Edward"
slug: "ai-demo-production-gap"
keywords: "keyword1, keyword2, keyword3, keyword4"
image: "/files/images/blog/..."
kind: "blog"
public: true
tags: []
published_at: "2026-01-17"
---

8.2 Review Against Original Aim

{ "command": "core content show ai-demo-production-gap --source blog" }

Evaluation checklist:

  • Title is max 8 words, no colons or em-dashes
  • Title is personal and specific (not generic "Best Practices" style)
  • Length is 3,500-5,000 words
  • Minimum 5 inline citations with full URLs and descriptive anchor text
  • No domain-only citations like [medium.com] or [reddit.com]
  • British English throughout (realise, optimise, colour)
  • Structure follows: Prelude -> Problem -> Journey -> Lesson -> Conclusion
  • 2-3 code examples included (if technical topic)
  • No AI anti-patterns ("I discovered that...", "Fascinatingly...")
  • No fake engagement questions
  • Contrarian hook that challenges conventional wisdom

CRITICAL: No Fabrication Check

  • All "I built", "we implemented", "my team" claims are backed by cited sources
  • Metrics and statistics have citations (not invented)
  • Personal anecdotes reference research, not fabricated experiences
  • Code examples are from research sources or generic illustrations (not fake "production" code)

If fabrication is detected: Content must be regenerated with proper research citations. The agent MUST NOT invent experiences.

8.3 Edit to World-Class Quality

If the content needs refinement, edit directly in the filesystem, then re-ingest:

{ "command": "core content ingest services/content/blog --source blog --override" }

{ "command": "core content show ai-demo-production-gap --source blog" }

Refinement focus areas:

  • Strengthen the prelude - must hook within first 2-3 sentences
  • Ensure personal story feels authentic, with specific details (dates, numbers, names)
  • Check each section earns its place - remove any padding
  • Verify code examples are real, not contrived
  • Ensure citations use descriptive anchor text with full URLs
  • Remove any hollow conclusions or generic takeaways
  • Tighten prose - every paragraph must advance the narrative
  • Check the lesson connects to bigger themes beyond the immediate topic
  • Ensure the return to the opening in the conclusion lands

8.4 Final Verification

{ "command": "core content verify ai-demo-production-gap --source blog --base-url https://tyingshoelaces.com" }

Content is ONLY complete when:

  1. Exists in database (core content show succeeds)
  2. Exists on filesystem (services/content/blog/[slug]/index.md exists)
  3. Frontmatter is valid YAML with all required fields
  4. Word count is 3,500-5,000 words
  5. Contains minimum 5 inline citations with full URLs
  6. Title follows rules (max 8 words, no colons/em-dashes, personal)
  7. Matches the contrarian hook and narrative arc agreed in planning
  8. NO FABRICATION - All experiences/metrics backed by research citations
  9. Published and accessible on site

Troubleshooting

Issue Solution
Agent not responding Check with { "command": "infra services status" }
Content not created Verify with { "command": "core content search \"[slug]\"" }. Agent may claim success falsely.
Agent says "created" but content doesn't exist Create a NEW context and try again. Context contamination is the most common cause.
Research artifact not found Create a NEW context. The old context has conflicting artifacts.
Agent uses wrong research_id Create a NEW context. Multiple research artifacts are confusing the agent.
Agent hallucinates Content ID Create a NEW context. Agent is referencing artifacts from previous sessions.
"Minimum 5 grounding links" error Research phase incomplete; ask agent to research more
"Keywords cannot be empty" error Include keywords array in content_create message
"Expected presentation_card artifact" error Create a NEW context. Research artifact type mismatch from old session.
Content not created with requested slug Create a NEW context. Agent reused prior context artifacts.
Content not publishing Run { "command": "infra jobs run publish_content" }
Featured image 404 on blog page Run { "command": "infra jobs run blog_image_optimization" } BEFORE publish_content. Blog template expects .webp but images are created as .png.
Image works on homepage but 404 on blog detail Same as above - homepage uses .png, blog detail expects .webp. Run image optimization.
Missing featured image Generate with generate_image or use default placeholder
Content too short Explicitly state "MUST be 3000+ words minimum"
Agent auto-researches instead of planning Add "Do NOT call research_content yet" to message
American English instead of British Remind agent: "Use British English (realise, optimise)"
No code examples included Explicitly request: "Include 2-3 code examples showing..."

See the Session Playbook and Contexts Playbook for context contamination symptoms and solutions. The fix is always: create a NEW context and start fresh.