stc

Tracks long chat sessions with numbered message blocks and a session debts file.

Updated Nov 21, 2025
One-click install
npx skills add https://github.com/alatyshau/duet --skill stc-alatyshau
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: stc
Source: https://github.com/alatyshau/duet/tree/main/packages/skills/modes/stc
Command: npx skills add https://github.com/alatyshau/duet --skill stc-alatyshau

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Long IDE chat sessions with an AI assistant become noisy: tool calls, inline narration, and intermediate text bury the messages actually addressed to the user, and open questions get silently dropped. This Skill enforces a strict per-turn communication protocol so the user can navigate a long chat without scrolling and never loses track of open items. ## Core Features & Use Cases - Numbered message blocks: Every assistant turn ends with a ## 📨 К тебе #N block containing a routing intro (context, open debts, diff, calibration) and a next-best-action closer, with session-wide monotonic numbering that survives context compaction. - Session debts tracking: A session_debts.md file mirrors all open questions and commitments, synchronized every turn via Diff lines, so nothing agreed upon is lost. - One question per post: The single most blocking question is placed in the За тобой line of the closer, preventing stacked questions from going unanswered. - Use Case: During a multi-hour refactoring session in an IDE chat, the user can jump directly to each numbered block, see exactly what decision is on them, and resume after compaction by reading the debts file. ## Quick Start Load the stc skill and ask the assistant to track this session with numbered message blocks and a session_debts.md file.

Frequently Asked Questions about stc

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I keep track of open questions in a long AI chat session?

Use a session tracking protocol where every assistant turn ends with a numbered message block and all open items are written to a session_debts.md file. The file is synchronized each turn via diff lines, so it always reflects the current state of open questions.

How to resume an AI chat after context compaction without losing numbering?

Open the session_debts.md file, find the maximum turn heading, and set the next turn number to that value plus one. If the file does not exist, the session starts at turn one. This keeps numbering monotonic across compaction.

Where is the session_debts.md file stored?

Location depends on session type: project-management sessions use projects/WIP_<name>/session_debts.md, saga executors use <node>/work/session_debts.md, vizir sessions use <saga>/session_debts.md, and ad-hoc sessions use the current working directory.

Why should an AI assistant ask only one question per message?

Stacked questions typically get only one answer while the rest hang silently and create noise. Placing the single most blocking question in the closer and deferring the rest to the debts file keeps every turn answerable at a glance.

What are the limitations of numbered chat block tracking?

The protocol adds overhead to every turn and requires discipline to keep the debts file synchronized with the chat. It is designed for long working sessions; short one-off questions gain little from the full block structure.