Mesh Pipeline Runner

Agent-driven article generation pipeline with full traceability.

Pipeline Stages

[1] DISCOVERY    → Broad search → Pick trending topic
[2] RESEARCH     → Deep search  → Topic brief + angle
[3] LDR          → Deep research → Synthesis + facts
[4] WRITER       → LLM generate → Article draft (~850 words)
[5] EDITOR       → Quality review → Edited draft
[6] EIC          → Word count gate → WordPress + Discord

Usage

# Run full pipeline
python3 traced_agent.py --seed "AI news 2025" --type news

# Output: /tmp/pipeline_runs/run_YYYYMMDD_HHMMSS/

Artifacts (Traceability)

Each stage saves input + output:

run_20260304_073045/
├── 01_discovery/
│   ├── 01_input_seed.txt
│   ├── 02_search_results.json
│   └── 03_output_topic.json
├── 02_research/
│   ├── 01_input_topic.json
│   ├── 02_deep_search.json
│   └── 03_output_brief.json
├── 03_ldr/
│   ├── 01_input_brief.json
│   └── 02_output_enriched.json
├── 04_writer/
│   ├── 01_input_enriched.json
│   └── 02_output_draft.md
├── 05_editor/
│   ├── 01_input_draft.md
│   └── 02_output_edited.md
├── 06_eic/
│   ├── 01_input_edited.json
│   ├── 02_decision.json
│   └── 03_output_scheduled.json
└── pipeline.log

Resume / Debug

Inspect any stage:

# See what topic was discovered
cat /tmp/pipeline_runs/run_*/01_discovery/03_output_topic.json

# See why EIC rejected
cat /tmp/pipeline_runs/run_*/06_eic/02_decision.json

Environment

Required:

• BRAVE_API_KEY - for web search
• OPENAI_API_KEY - for LLM generation
• LDR server at ldr:5000
• WordPress credentials in ezine/config/

Content Type Rules

Type	Word Target	Min Gate	48h Rule
News	850	800	Must be <48h old
Analysis	1600	1500	Any age
Opinion	1050	1000	Any age

Output

On success:

• WordPress post scheduled (future date)
• Discord notification to #Zine
• Full article saved as markdown

On failure:

• Decision json with rejection reason
• Draft saved for inspection
• Logs show where it failed