session-start

Loads active project context from markdown artefacts to orient a new Claude Code session.

Updated May 25, 2026
One-click install
npx skills add https://github.com/DATA-AI-XYZ/Tandem --skill session-start-data-ai-xyz
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: session-start
Source: https://github.com/DATA-AI-XYZ/Tandem/tree/main/skills/session-start
Command: npx skills add https://github.com/DATA-AI-XYZ/Tandem --skill session-start-data-ai-xyz

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When returning to a project after time away, the reasoning behind decisions and the state of in-flight work is scattered across chat history and files. This Skill rebuilds that context at session start by reading the project's maintained markdown artefacts, so you resume from an accurate picture instead of reconstructing it manually. ## Core Features & Use Cases - Active WIP orientation: Reads ACTIVE.md and scans stories to list in-progress, blocked, and stale in-review work with acceptance-criteria state. - Decision and monitor history: Surfaces the five most recent ADRs and the latest MONITOR revision line, treating the monitor as the authoritative source over fan-out scans. - Autopilot run detection: Probes checkpoints for unfinished or stale paused autopilot runs and reports them read-only, without altering any artefact. - Use Case: You open Claude Code on Monday and ask "where did we leave off" — the Skill returns a compact orientation block naming the active story, one blocked item, recent ADRs, and a single suggested next action. ## Quick Start Ask the assistant to run the session-start orientation and summarize what is in progress, blocked, and what to do next.

Frequently Asked Questions about session-start

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

FAQPage Schema
How do I restore project context at the start of a Claude Code session?

Invoke the session-start skill or ask "where did we leave off". It reads ACTIVE.md, the MONITOR revision history, recent ADRs, and story statuses, then outputs a short orientation block with a suggested next step.

What files does session-start read to build the orientation?

It reads ACTIVE.md from the resolved active folder, MONITOR.md from the monitor folder, the last five ADR-*.md files, and STORY-*.md files with in-progress, blocked, or stale in-review status. Folder locations are resolved through a path map rather than hardcoded.

Does session-start work with a non-standard folder layout?

Yes. It resolves folders through pm-paths.js, which maps logical roles like stories, decisions, and monitor to either the canonical scaffold or a flattened variant. The orientation notes the resolved layout in one line at the top.

Why does session-start trust the MONITOR over a story folder scan?

A broad fan-out scan can silently undercount or sample a subset of stories, while the MONITOR is the maintained source of truth. If the two disagree and the monitor was updated within seven days, the monitor wins and the discrepancy is flagged.

Can session-start resume or modify a paused autopilot run?

No. It only reports unfinished or stale paused runs by reading checkpoint output, naming the run ID, reason, age, and checkpoint path. Acting on a paused run is a deliberate operator decision made in a separate session.