Blog Content Pipeline — Quick Reference

Reference card for the 4-tier content classification system, 4 skills, and 3 agents that power startaitools.com publishing.

Version: 3.0.0 | Author: Jeremy Longshore | Last Updated: 2026-04-09

The 4-Tier System

Tier	Name	Trigger	Length	Quality Gates
1	Field Note	Auto-classified	80-140 lines	Hugo build
2	Technical Deep-Dive	Auto-classified	150-250 lines	Hugo build + consistency audit
3	Case Study	Auto-classified	300-500 lines	Hugo build + consistency + fact-check
4	Distinguished Paper	Manual (`/blog-research-article`)	1200-1800 words	Full quality gate chain + user review

Expected distribution: 60-70% Tier 1, 25-35% Tier 2, 5-10% Tier 3

Key rule: Every day's work is covered regardless of tier. Tier determines treatment (depth, structure, quality gates), not scope.

Skills

`/blog-backfill [START END]`

What it does: Generates one blog post per day from git activity across all repos. Auto-classifies each day into Tier 1-3. Records all decisions in methodology tracking. Publishes to startaitools.com, dual-publishes to tonsofskills.com, queues cross-posts, sends social bundles.

Arguments:

/blog-backfill 2026-03-01 2026-03-15   # Specific date range
/blog-backfill                          # Auto-detect: last post date → yesterday
/blog-backfill weekly                   # Generate weekly recap
/blog-backfill monthly                  # Generate monthly retrospective

Pipeline (per day):

Gather material (git, PRs, beads, transcripts, email signals)
Classify → blog-classifier agent scores 6 dimensions, returns tier
Record decision → methodology/decisions.jsonl
Write post → tier-specific structure and length
Quality gates → tier-specific (consistency audit for T2+, fact-check for T3)
Publish → Hugo build, commit, push to master, Netlify auto-deploys
Cross-post → Tier 2+ only (Dev.to +24h, Hashnode +24h, Medium +48h)
Social bundle → Tier 2+ gets X thread + LinkedIn + Substack

Automation (RemoteTrigger cron):

Daily: 17 6 * * * (6:17 AM CT)
Weekly recap: 42 17 * * 5 (Friday 5:42 PM CT)
Monthly retro: 23 8 1 * * (1st of month 8:23 AM CT)

Key files:

~/.claude/skills/blog-backfill/
├── SKILL.md                              # Main orchestrator (v3.0.0)
├── agents/
│   ├── blog-classifier.md                # Tier classification agent
│   ├── blog-consistency-checker.md       # Thesis/tone/contradiction audit
│   └── blog-fact-checker.md              # External claim verification
├── methodology/
│   ├── decisions.jsonl                   # Classification decision log
│   ├── feedback.jsonl                    # Post-publication assessments
│   ├── patterns.jsonl                    # Emergent classification rules
│   └── rebuild-index.sql                 # SQLite analytics schema
├── references/
│   ├── content-tier-classification.md    # Tier defs, dimensions, anti-inflation
│   ├── classify-day.md                   # Classifier prompt + calibration examples
│   ├── write-post.md                     # Tier-specific writer instructions
│   ├── cadence-system.md                 # Weekly/monthly templates + cron specs
│   ├── content-strategy.md              # Research-backed distribution strategy
│   ├── gather-material.md               # Source collection + email guardrails
│   ├── final-verification.md            # Post-publish checklist
│   ├── publish-verify.md                # Hugo build + push commands
│   ├── crosspost-queue.md               # Staggered cross-post schema
│   └── social-bundle.md                 # X/LinkedIn/Substack format
└── scripts/                              # Shell scripts for cross-posting

`/blog-research-article [slug-or-url]`

What it does: Tier 4 Distinguished Paper workflow. Interactive, research-grade writing methodology. Starts from a user-provided article link, researches the landscape, presents angles, discusses with user, then runs an 8-step methodology grounded in classical rhetoric and composition theory.

Arguments:

/blog-research-article software-supply-chain-security                    # Slug only — asks for source
/blog-research-article https://arxiv.org/abs/2025.12345                  # URL only — derives slug
/blog-research-article supply-chain https://arxiv.org/abs/2025.12345     # Both

Workflow:

Phase 0: Input & Research
  ├── Fetch article, research landscape, find related work
  └── Prepare research brief

Phase 1: Angle Selection (interactive)
  ├── Present 3-5 candidate angles with Toulmin structure
  ├── Back-and-forth discussion
  └── Lock thesis statement

Phase 2: 8-Step Methodology
  1. Invention — freewriting, 5W1H, evidence gathering
  2. Audience & Framing — PAPAS framework, Toulmin argument map
  3. Arrangement — outline with placed evidence
  4. Drafting — full first draft (1200-1800 words)
  5. Revision — structural and accessibility review
  5b. Quality Gate — fact-check + consistency audit (parallel)
  6. Editing — polish, SEO, front matter
  7. Platform Adaptation — 6 platform versions
  8. Publication — full pipeline (same as blog-backfill)

Quality bar: "Would someone cite this?"

Theoretical foundations: Cicero's 5 canons, Flower & Hayes cognitive model, Toulmin argumentation, Elbow freewriting, Murray process of discovery, ACM 6-criteria evaluation, PAPAS method.

`/blog-calibrate [YYYY-MM]`

What it does: Monthly calibration report from the methodology tracking system. Analyzes classification decisions and feedback to detect tier inflation, measure calibration accuracy, and surface patterns.

Arguments:

/blog-calibrate 2026-04    # Calibrate April 2026
/blog-calibrate            # Calibrate previous month

What it measures:

Tier distribution — is Tier 1 still >60%? Is Tier 3 <15%?
Brier score — stated confidence vs actual accuracy (lower = better, target <0.15)
Decision quality matrix — Annie Duke 2x2 (good/bad decision × good/bad outcome)
Dimension analysis — which dimensions drive tier upgrades? Any systematic bias?
Anti-inflation flags — how often triggered, any gaps?
Pattern learning — emergent rules from 30+ decisions

Requires: At least 5 entries in decisions.jsonl. Better with feedback.jsonl data (run /blog-feedback on past posts).

`/blog-feedback <slug> [--correct|--wrong TIER] [--notes 'text']`

What it does: Post-publication assessment for classified posts. Records whether the tier was right, captures engagement data, feeds the calibration loop.

Arguments:

/blog-feedback knowledge-os-bootstrap --correct                           # Quick: correct
/blog-feedback wcag-color-audit --wrong 2 --notes "More teaching value"   # Wrong tier
/blog-feedback plugin-registry --correct --notes "Got 2 backlinks"        # Correct with data
/blog-feedback plugin-registry                                            # Interactive mode
/blog-feedback --batch 2026-04-01 2026-04-07                              # Batch review

How it feeds the system:

/blog-feedback → feedback.jsonl → /blog-calibrate → Brier score + patterns

Agents

`blog-classifier`

When invoked: Automatically during /blog-backfill Phase 2, step 2.

Input: All gathered source material for a day (git logs, PRs, beads, transcripts, email).

Output: JSON classification with tier, confidence (0-1), 6 dimension scores (0-5 each), thesis candidate, rhetorical structure, anti-inflation flags.

6 Scoring Dimensions:

Dim	Name	What it measures
NOV	Novelty	New to the field vs routine application
ARC	Architectural Significance	Config tweak vs new paradigm
NAR	Narrative Richness	Linear execution vs dramatic arc
TCH	Teaching Potential	Project-specific vs changes mental models
SCP	Scope	Single file vs greenfield multi-system
RPR	Reproducibility	Project-specific vs packageable pattern

Classification gates:

Tier 3: max(NOV, TCH) >= 4 AND NAR >= 4 AND 3+ dims >= 3
Tier 2: max(NOV, TCH, NAR) >= 3 AND 2+ dims >= 3
Tier 1: Everything else

7 anti-inflation rules: Volume ≠ quality. Busy ≠ distinguished. First-time-for-me ≠ novel. "Year from now" test. Tier 3 is rare. Default down. High scope alone doesn't escalate.

`blog-consistency-checker`

When invoked: Quality gate for Tier 2+ posts. Also used by /blog-research-article step 5b.

What it checks:

Thesis drift (introduction promise vs conclusion delivery)
Contradictory claims within the post
Tone shifts (casual → formal → casual)
Code-narrative alignment (does the code match what the text describes?)
Structural coherence

Output: Audit report with issues categorized as CONTRADICTION, DRIFT, TONE, or CODE-NARRATIVE. Each issue gets HIGH/MEDIUM/LOW severity.

Recommendations: PASS (0 high), REVISE (1+ high or 3+ medium), BLOCK (2+ high contradictions).

`blog-fact-checker`

When invoked: Quality gate for Tier 3 posts. Also used by /blog-research-article step 5b.

What it checks:

Version numbers and release dates
API behavior claims
Performance claims
Attribution and citations
Comparative claims

Priority levels: HIGH (central to argument), MEDIUM (supporting credibility), LOW (incidental).

Verdicts: VERIFIED, INACCURATE, OUTDATED, OVERSTATED, UNVERIFIABLE.

Recommendations: PASS (0 high issues), REVISE (1+ high overstated/outdated), BLOCK (1+ high inaccurate).

Methodology Tracking

All classification decisions are recorded in append-only JSONL files. SQLite provides a derived queryable index.

methodology/
├── decisions.jsonl    # One record per classified post (source of truth)
├── feedback.jsonl     # Post-publication assessments (via /blog-feedback)
├── patterns.jsonl     # Emergent rules from calibration reviews
└── rebuild-index.sql  # SQLite DDL: 3 tables, 7 indexes, 7 analytics views

Decision record schema:

{
  "date": "2026-04-06",
  "slug": "knowledge-os-bootstrap",
  "tier": 2,
  "confidence": 0.85,
  "dimensions": { "novelty": 3, "arc": 4, "nar": 3, "tch": 4, "scp": 5, "rpr": 3 },
  "reasoning": "Why this tier, not the adjacent one",
  "thesis_candidate": "One-sentence thesis",
  "rhetorical_structure": "problem-solution",
  "source_signals": { "git": "strong", "prs": "strong", "session": "moderate", "beads": "absent", "email": "absent" },
  "cadence_type": "daily"
}

Analytics views (SQLite):

v_tier_distribution — monthly tier percentages
v_confidence_by_tier — avg/min/max confidence per tier
v_calibration — stated confidence vs actual accuracy
v_brier_score — overall calibration score
v_anti_inflation_frequency — most common anti-inflation flags
v_top_teaching — highest TCH-scored posts
v_decision_quality — Annie Duke 2x2 quadrants

Email Signal Sources

Three tool options for lightweight email scanning (allowlist-only, date-bounded, <30s per day):

Tool	Type	Best for
IntentMail (`intent-mail/`)	Local SQLite+FTS5	Fastest, privacy-first, no network after sync
Gmail MCP	Claude.ai native	Works without local setup
gogcli (steipete)	CLI, JSON-first	Script-friendly, multiple accounts

Future: msgvault (Wes McKinney) — DuckDB+Parquet for offline email analytics at scale.

Content Distribution by Tier

Tier	Platforms	Social
1	startaitools + tonsofskills	Optional X post
2	All platforms	X thread + LinkedIn
3	All platforms	X thread + LinkedIn + outreach
4	All platforms	Full social campaign
Weekly	startaitools + tonsofskills	LinkedIn post
Monthly	All platforms	X thread + LinkedIn

Stagger: Canonical Day 0 → Dev.to/Hashnode +24-48h → Medium +48-72h

Quick Commands

# Daily backfill (auto-classify + publish)
/blog-backfill

# Backfill specific range
/blog-backfill 2026-04-01 2026-04-08

# Weekly recap
/blog-backfill weekly

# Monthly retrospective
/blog-backfill monthly

# Distinguished Paper from article
/blog-research-article https://example.com/paper

# Assess a past post's classification
/blog-feedback knowledge-os-bootstrap --correct

# Batch assess a week
/blog-feedback --batch 2026-04-01 2026-04-07

# Monthly calibration report
/blog-calibrate 2026-04

# Rebuild analytics index
sqlite3 ~/.claude/skills/blog-backfill/methodology/index.db < ~/.claude/skills/blog-backfill/methodology/rebuild-index.sql

jeremylongshore/blog-content-pipeline-reference.md

Select an option

No results found

Select an option

No results found

Blog Content Pipeline — Quick Reference

The 4-Tier System

Skills

`/blog-backfill [START END]`

`/blog-research-article [slug-or-url]`

`/blog-calibrate [YYYY-MM]`

`/blog-feedback <slug> [--correct|--wrong TIER] [--notes 'text']`

Agents

`blog-classifier`

`blog-consistency-checker`

`blog-fact-checker`

Methodology Tracking

Email Signal Sources

Content Distribution by Tier

Quick Commands