Maintain Project Context Across Sessions

Provides continuity between Claude Code sessions: captures, compresses, and restores context automatically - topics, decisions, tasks, errors, modified files, and technical discoveries. On session end, save finds the most recent session JSONL, analyzes every message, tool call, and result, generates a structured summary (session-NNN.md), updates a consolidated ACTIVE_CONTEXT.md (capped at 150 lines), syncs it into MEMORY.md - which is loaded into every session's system prompt automatically - and indexes everything for full-text search. On a new session, load generates a briefing covering active projects, pending tasks by priority, blockers, recent decisions, conventions, and a summary of recent sessions. A status command gives a quick few-line summary, search "<query>" runs a SQLite FTS5 full-text search across all sessions, and maintain archives old sessions, compresses the archive, resyncs MEMORY.md, and rebuilds the search index.

python C:\Users\renat\skills\context-agent\scripts\context_manager.py save

Use it when the user asks to save context, wants a briefing for the next session, a session summary, or session-to-session continuity. Not a fit for tasks outside context management, or as a substitute for a more specific tool.

Input: an ending or ongoing Claude Code session's JSONL transcript, or a search query against session history. Output: a per-session summary file, an updated ACTIVE_CONTEXT.md and MEMORY.md, a load-time briefing (projects, prioritized tasks, blockers, decisions, conventions), and full-text search results across all captured sessions.

Built as a set of Python scripts (session_parser.py, session_summary.py, active_context.py, project_registry.py, compressor.py, search.py, context_loader.py, context_manager.py) backed by a SQLite FTS5 database, with its ACTIVE_CONTEXT.md synced into the Claude Code project's MEMORY.md so every new session's system prompt already reflects current project state without manual action. Complements the context-guardian skill.

Developers running many Claude Code sessions on the same project who want automatic session-to-session continuity instead of manually re-explaining state each time.

Each session capture records nine categories of information: topics discussed, technical and architectural decisions, completed tasks, pending tasks with priority, modified files, technical discoveries, resolved errors and their fixes, open questions without answers, and usage metrics like tokens consumed and tool calls made. Its own stated best practices call for providing clear, specific project context and reviewing suggestions before applying them to production code, and its common pitfalls warn against using it outside its domain applying its recommendations without understanding the specific project context, or supplying too little context for the analysis to be accurate.