context persistence

The Obsidian second-brain is a concrete implementation of context persistence across sessions—files remain available across multiple AI interactions, enabling compounding.

Article demonstrates context persistence achieving knowledge compounding across projects

MCP server enables context to persist across sessions—Claude can query current workout stats without user re-explaining state each time.

Demonstrates how MCP configuration enables context to persist across multiple devices and sessions

MCP is a protocol explicitly designed to maintain tool/data context across sessions rather than losing this information on conversation restart

Experience libraries are a specific implementation of context persistence—maintaining structured context across multiple agent interactions to enable learning and compounding intelligence

The HTML report pattern is a concrete implementation of context persistence—creating an artifact that preserves understanding across sessions so intelligence doesn't reset.

Proprietary code persisting on vendor servers without deletion policy demonstrates uncontrolled context lifetime

Shows that context persistence requires infrastructure investment (trace storage layer). Can't compound intelligence without persistent storage.

MCP's 'context-aware workflows' explicitly enable context persistence across multiple tool interactions, which is a core context engineering requirement

Checkpointing and state persistence allow context to survive across agent steps and cycles—this is persistence infrastructure.

By making coding guidelines available as MCP Resources, the update directly enables context to persist across sessions without re-explanation.

GitHub for agents implies need to persist and version agent context/behavior across team iterations and deployments

MCP enables tool context to persist and compound across sessions by formalizing the tool interface as a protocol rather than embedding it in prompts

CLAUDE.md persists context/instructions across sessions for the Apple Support app integration

CLAUDE.md persists across tasks; author observes improvement across multiple tasks ('a few tasks in it feels tighter'), demonstrating context compounding.

Artifacts stored in repos solve the session-reset problem by making context durable and discoverable across time and projects

Markdown files persisted across sessions enable context to compound—a direct solution to the persistence problem.

Routines are a concrete implementation of context persistence—the configured prompt, repo, and connectors persist across execution cycles rather than being reset.

The ~/.claude/output-styles configuration file is a direct implementation of persisting context (user preferences) across sessions rather than re-explaining in every conversation.

Event triggers require agents to maintain execution context across multiple async events, directly supporting the thesis that intelligence should compound rather than reset.

The paper's distinction between external scaffolding (stateless per session) and internal initiative (state-bearing across sessions) is a formalization of why context persistence matters. GIC architec

Unified MCP pattern creates persistent context layer independent of model/provider choices, enabling knowledge compounding

Explicitly addresses the problem of context resets by creating persistent folder structure that survives across sessions and team members

Managing context across sessions and preventing degradation implies preservation strategies are taught in course.

Claude Code Skills/Memory/Files system enables context carrying across development phases—compounding intelligence through session management

MCP servers function as a persistence layer for context about system capabilities and integrations. Once defined, this context is reused across agents rather than rebuilt.

The 5M-word second brain is a concrete implementation of persisting context across sessions to enable compounding intelligence.

Challenges traditional stateful context persistence by exploring stateless alternatives with dynamic auth

Artifacts now persist and are shared across users, extending context lifetime beyond individual sessions

The question implicitly depends on understanding whether MCP better enables context preservation across sessions compared to skills/APIs

LangGraph's 'built-in persistence' and CrewAI's 'memory sharing' are concrete persistence mechanisms that preserve context

The SPEC.md pattern is a concrete implementation of preserving intelligence across sessions. Modified specs compound knowledge rather than resetting it.

FOMAT pattern reveals context is NOT persisting across prompt iterations. Engineers caught in loops where intelligence resets.

All four strategies enable context to compound across sessions rather than reset; Compress + Select + Isolate preserve working context

MCP servers maintain persistent connections to external data sources, allowing context to be retrieved consistently across sessions

Implements persistent context across multiple coding contexts/multitasking scenarios, not just single-conversation persistence.

Well-structured specs that are 'executable' remain consistent across iterations; loose specs decay. This enables compound intelligence across sessions.

Author's real problem: how to persist context and decisions across sessions (using scheduled tasks + reminders), but without clarity on what gets persisted where

System prompts are context that must persist/survive process boundaries; tmp file approach is a persistence mechanism

Skills enable knowledge preservation across multiple agent invocations, reducing need to re-explain tool usage in each session

The article claims intelligence doesn't last; context engineering argues that preserving intelligence IS possible and critical. These are opposing views of what's feasible.