ACRONYMS in filename

'JG' - Jean Guillaume (alias William), a user, the creator / innovator github handle: jgwill and personnal account github guillaumedescoteauxisabelle

Some major repo involved from this user.

• jgwill/Miadi /src/Miadi-46
• guillaumedescoteauxisabelle/stories /workspace/stories
• jgwill/StorytellingPlatformBlueprint2510 /src/STPB
• <MORE COULD BE ADDED HERE BY BOT / INSTANCES>

'NCPSTC' - Means Narrative Context Protocol (NCP) Structural Tension Chart (STC) see: /src/llms/llms-structural-tension-charts.txt

• It explores that the NCP could be subject to generate, thru its narrative of a form of tension (as we define it within the Structural Approach from Robert Fritz)

'EAST' - Related to IAIP-mcp East direction

• Explore that the Ceremonial Universe with its directional thinking (use MCP tool: iaip-mcp) can really be something to work with LLM thru a ceremonial process. see: /src/IAIP/ (which is the platform code with the codebase of the iaip-mcp)

'PTN' - Patentability etc

• Exploring using perplexity (and maybe later) various potential patentable technology to keep them more than just free, used with respect

Directing toward inquiries and their intentions

260101-EAST-INQUIRIES-905071e0-cc62-4451-b812-92e1

agentic-flywheel

• we would use that for doing inquiries within flowise with a RAG agent.

7 Paradigm Directions at a Glance

• What is bellow might need upgrades or validation

URL: https://www.perplexity.ai/search/260101-east-inquiries-905071e0-n1hv2Kk2TqCSWmVtI4HjXA?preview=1

• If need be to access it with playwright or by '--chrome' when launching

how will @260116-STCMastery-Claude-remote-env--5ed76881-8fee-4597-9487-f8184dd5946b--tsh_CLI_client affect what we are doing within claude and this Inquiry Ecosystem and session management....

mcp-pde — Prompt Decomposition Engine

Purpose: Completeness safeguard. Pass user input through mcp-pde at session start to catch implicit intents that LLMs naturally filter out.

When to use

• Session start — decompose the user's initial prompt FIRST before acting
• Voice transcription handoff — decompose raw verbatim, then agent-processed version separately
• Multi-mission inputs — decompose the meta-instruction, then each sub-mission

Workflow

1. pde_decompose(prompt) → returns systemPrompt + userMessage for LLM processing
2. Process the prompts and generate the decomposition JSON
3. pde_parse_response(llm_response, original_prompt) → auto-saves to .pde/ (JSON + markdown)

⚠️ Potential upgrade: A single-call pde_decompose_and_store that handles all 3 steps internally would remove the manual middle step. See /src/mcp-pde codebase.

What it catches that you miss

• Implicit intents from hedging language ("somehow", "probably", "I might need")
• Dependencies between actions you'd execute out of order
• Ambiguities you'd silently resolve with assumptions
• The user's actual instruction sequence (e.g., "pass this input first, then the verbatim")

Output structure

• primary — single most important action
• secondary[] — all other intents (explicit + implicit)
• directions — Four Directions mapping (East/South/West/North)
• actionStack — ordered task list respecting dependencies
• ambiguities — flagged vagueness with suggestions

Key lesson

mcp-pde exists because LLMs drop "non-essential" parts that are essential. Always decompose the user's raw input — that's where the intent lives that you would silently discard.