What are phase codes used for?

Phase codes (like F1 for Discovery or B4 for Root Cause Analysis) provide a precise resume point for the AI, ensuring it doesn't repeat work or skip steps after recovery.

What is context compaction in Claude Code?

Context compaction occurs when the conversation length exceeds the model's window, causing it to summarize or drop earlier details. This skill prevents the loss of critical planning data during that process.

Where are the session states stored?

Session states and working data are stored in the finesse-plans/ directory, ensuring your planning progress is saved locally within your project repository.

Can I use this skill for code implementation?

No, this skill is explicitly designed for 'planning-only' modes to ensure a thorough architectural plan is finalized before any source code is modified.

How does this skill prevent data loss?

It mandates writing key findings and session metadata to a working markdown file with a specific YAML schema, allowing the AI to perfectly restore its state if it loses memory.

Context Compaction Handling

Name: Context Compaction Handling
Author: jonathanung

byjonathanung

•

Productivity & Workflow

Manages state persistence and recovery for long-running AI planning sessions to prevent data loss during context window compaction.

The Context Compaction Handling skill for Finesse provides a robust framework for maintaining continuity during complex, multi-phase software planning. By enforcing a standardized YAML frontmatter schema and mandatory working file updates, it ensures that critical codebase findings, architectural decisions, and session states survive when Claude Code reaches its context limit. This skill is essential for developers undertaking large-scale refactors, feature designs, or research tasks that require high-fidelity state recovery and strict adherence to a planning-only workflow.

Key Features

014 GitHub stars

02Context pressure monitoring with token budget and cost estimation fields

03Standardized YAML frontmatter for persistent session state tracking

04Phase-specific recovery codes for Feature, Bug Fix, Refactor, and Research workflows

05Strict planning-only guardrails to prevent accidental code implementation

06Automatic state restoration from specialized working files in finesse-plans/

Use Cases

01Maintaining continuity during multi-step architectural planning that exceeds context limits

02Recovering session progress and metadata after AI context compaction occurs

03Standardizing the hand-off between codebase exploration and plan construction phases

Key Features

014 GitHub stars

02Context pressure monitoring with token budget and cost estimation fields

03Standardized YAML frontmatter for persistent session state tracking

04Phase-specific recovery codes for Feature, Bug Fix, Refactor, and Research workflows

05Strict planning-only guardrails to prevent accidental code implementation

06Automatic state restoration from specialized working files in finesse-plans/

Use Cases

01Maintaining continuity during multi-step architectural planning that exceeds context limits

02Recovering session progress and metadata after AI context compaction occurs

03Standardizing the hand-off between codebase exploration and plan construction phases