Releases·Tool Release
Solving Agentic Amnesia: Dev builds Craft, an open-source Claude Code plugin using the file system as a state machine for context
A developer built Craft, an open-source Claude Code Plugin addressing "agentic amnesia" — the context-window overflow that makes agents hallucinate or forget earlier architectural decisions. Craft shifts context management from prompts to the file system, treating prompt rules as guidance the model can drop under context pressure.
11 engagement1 sourceMon, Sep 7, 2026, 05:28 PM
AI Overview
- Craft (MIT-licensed, drobins25/craft) uses a local .craft folder as a deterministic state machine where the LLM reads and writes learnings, design tokens, and locked decisions
- The LLM has read-only access to the codebase by default and must propose an approved "story" or "fix" to write source code
- The "Room" protocol has the LLM review its on-disk past learnings before each session instead of keeping full history in the prompt
- Author argues prompt rules are just requests the model can starve out under context pressure, so management shifts to the file system
Entities
Related
Products·building
AgentJournal: open-source memory system for AI coding agents on long tasks
AgentJournal is an open-source memory management system for AI agents, designed for long coding tasks. It keeps the original prompt pinned, logs actions to a journal, and clears working context between tasks. This addresses failures where context compaction loses rules and context editing or Context Language Models consume excessive tokens without completing tasks.
AgentJournal is an open-source memory management system for AI agents, designed for long coding tasks. It keeps the original prompt pinned, logs actions to a journal, and clears working context between tasks. This addresses failures where context compaction loses rules and context editing or Context Language Models consume excessive tokens without completing tasks.
5 engagement1 sourceredditSun, Oct 11, 2026, 09:21 AM
Releases·Tool Release
Dev builds deterministic state plugin to fix long-horizon task drift in local models after context compaction
A developer shared an open-source plugin that counters goal drift in local agent setups after context-window compaction. It hooks into the agent loop and maintains a structured state file outside the chat buffer, re-anchoring the model to deterministic milestones after each compaction. The tool targets a common failure mode — hallucinated task completion and loops on long tasks — and is available on GitHub for adaptation.
A developer shared an open-source plugin that counters goal drift in local agent setups after context-window compaction. It hooks into the agent loop and maintains a structured state file outside the chat buffer, re-anchoring the model to deterministic milestones after each compaction. The tool targets a common failure mode — hallucinated task completion and loops on long tasks — and is available on GitHub for adaptation.
11 engagement1 sourceredditSat, Oct 3, 2026, 11:17 AM
Products·launched
LEO: Open-source framework to fix coding agent context drift
LEO is an open-source framework that addresses architectural amnesia in coding agents by providing a structured 127-file project organization. It uses LLMs to maintain context and enforce constraints, helping developers working on large codebases avoid context drift and improve agent reliability.
LEO is an open-source framework that addresses architectural amnesia in coding agents by providing a structured 127-file project organization. It uses LLMs to maintain context and enforce constraints, helping developers working on large codebases avoid context drift and improve agent reliability.
1 engagement1 sourceredditTue, Aug 25, 2026, 02:15 PM
Products·launched
MemContinuum — long-term decision memory for Claude Code projects
A long-term memory system that records what was decided on a project and why, addressing forgetfulness in LLM coding sessions and blind subagent dispatch in Claude Code. The builder found existing memory tools insufficient, built his own for Claude Code workflows, and released it publicly.
A long-term memory system that records what was decided on a project and why, addressing forgetfulness in LLM coding sessions and blind subagent dispatch in Claude Code. The builder found existing memory tools insufficient, built his own for Claude Code workflows, and released it publicly.
1 engagement1 sourceredditTue, Sep 8, 2026, 01:15 AM
Social·Community
Stop overengineering agent harnesses from scratch. How Project Athena manages memory, sub-agents, and verification in Claude Code
A Reddit poster argues developers should stop overengineering agent harnesses from scratch, describing how their Project Athena setup manages memory, sub-agents, and verification on top of Claude Code. The post responds to recent community threads about an 883-commit harness project abandoned to burnout, sub-agents damaging codebases, and Claude Code token burn. Its core claim: the LLM should be a replaceable worker, not the system owning planning, durable state, and verification.
A Reddit poster argues developers should stop overengineering agent harnesses from scratch, describing how their Project Athena setup manages memory, sub-agents, and verification on top of Claude Code. The post responds to recent community threads about an 883-commit harness project abandoned to burnout, sub-agents damaging codebases, and Claude Code token burn. Its core claim: the LLM should be a replaceable worker, not the system owning planning, durable state, and verification.
2 engagement1 sourceredditSun, Sep 6, 2026, 11:24 AM