katas-session-resume-fork

Decides between resume, fork, and fresh session patterns for Claude Code context continuity.

Updated Jun 28, 2026
One-click install
npx skills add https://github.com/JaviMontano/claude-plugins --skill katas-session-resume-fork-javimontano
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: katas-session-resume-fork
Source: https://github.com/JaviMontano/claude-plugins/tree/main/plugins/claude-native-toolkit/skills/katas-session-resume-fork
Command: npx skills add https://github.com/JaviMontano/claude-plugins --skill katas-session-resume-fork-javimontano

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Resuming an agent session after the codebase changed leads to hallucinations based on stale tool results, while pasting full old transcripts inflates context with resolved noise. This Skill provides explicit criteria for choosing between session resume, fork, and fresh sessions with a typed summary. ## Core Features & Use Cases - Decision Framework: Classifies the situation into resume (context still valid), fork (parallel branches from a shared baseline), or fresh session (world changed, tool results stale). - Staleness Detection: Identifies refactors, migrations, deploys, or mass edits that invalidate prior tool results and rule out resume. - Typed Summary Injection: Builds a curated summary from the structured scratchpad (Kata 18) instead of replaying raw transcripts. - Use Case: After a large refactor rewrote half the codebase, the Skill rejects claude --resume and starts a fresh session with a typed summary of validated findings, reloading current sources. ## Quick Start Ask the assistant to decide whether to resume, fork, or start a fresh session for continuing yesterday's codebase audit after a major refactor.

Frequently Asked Questions about katas-session-resume-fork

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

FAQPage Schema
How do I resume a previous Claude Code session?▼

Use claude --resume <session-name> when the prior context is still valid and nothing changed in the project. If a refactor, migration, or deploy happened since the last session, resume is unsafe because the tool results describe a state that no longer exists.

When should I fork a Claude Code session instead of resuming?▼

Fork when you want to explore two independent approaches from the same baseline without cross-contamination. Use claude --fork <baseline> --new-name approach-A to create named, isolated branches with zero interference between them.

What are stale tool results in agent sessions?▼

Stale tool results are outputs from prior tool calls that describe files or state changed by a later refactor, migration, or mass edit. Resuming on them causes the model to hallucinate about code as it used to be, so a fresh session with a typed summary is required.

Why not paste the full old transcript into a new session?▼

Injecting a raw transcript inflates context with already-resolved turns that compete for attention and degrade the prefix cache. A curated typed summary from a structured scratchpad preserves durable findings without the noise.

What should a typed session summary contain?▼

A typed summary uses a stable structure: objective, findings, decisions, file states, and next steps. It is sourced from the structured scratchpad and injected into the system prompt, with current sources reloaded so tool results reflect the present state.