Releases·Tool Release
Developer builds MCP-based memory store after coding agents ignored his DECISIONS.md file and re-hit documented dead ends
A developer describes maintaining DECISIONS.md files in every repo, only to find coding agents never read them unprompted, forcing manual re-briefing each session that consumed hours weekly. In response, the developer built a tool — described as a cloud layer over MCP storing decisions, failed attempts, conversations, skills, procedures, and tasks — so every agent reads the same context without reminders. The post doubles as market research, asking whether other developers face the same recall problem and would pay for a solution.
40 engagement1 sourceSun, Sep 27, 2026, 10:21 PM
AI Overview
- A developer's DECISIONS.md file of rejected approaches went unread by agents unless explicitly prompted, costing hours of weekly re-briefing
- An agent hit the same documented CORS problem twice despite the fix being written in the repo's decisions file
- He built a cloud service over MCP that stores decisions, failed attempts, conversations, skills, and tasks so every agent reads the same context without being told
- Re-briefing new AI tools on his whole project before judging them is his main blocker to trying new tools
- He does design thinking via speech-to-text on his phone, but the resulting context stays stranded on his laptop
Entities
Related
Social·Community
A developer dropped per-agent decisions.md files for shared memory over MCP after multi-tool agents desynced
A developer describes abandoning per-tool decisions.md files after multi-agent setups (Claude Code, Codex, n8n workflows, ChatGPT tasks) produced conflicting copies of project state. They switched to a single shared memory accessed over MCP via Vilix AI, so all agents read and write the same state. Caveats: models must be instructed to save, and last-write-wins can shadow careful notes with sloppy ones.
A developer describes abandoning per-tool decisions.md files after multi-agent setups (Claude Code, Codex, n8n workflows, ChatGPT tasks) produced conflicting copies of project state. They switched to a single shared memory accessed over MCP via Vilix AI, so all agents read and write the same state. Caveats: models must be instructed to save, and last-write-wins can shadow careful notes with sloppy ones.
34 engagement1 sourceredditTue, Sep 22, 2026, 11:30 PM
Products·launched
Keep the Why: Git-native memory skill for coding agents
Keep the Why is an open-source Agent Skill that captures the reasoning behind coding decisions and stores it inside the project's Git repository, so coding agents can learn from past mistakes. It uses LLMs to generate and maintain this memory, solving the problem of agents repeating errors without requiring extra infrastructure.
Keep the Why is an open-source Agent Skill that captures the reasoning behind coding decisions and stores it inside the project's Git repository, so coding agents can learn from past mistakes. It uses LLMs to generate and maintain this memory, solving the problem of agents repeating errors without requiring extra infrastructure.
3 engagement1 sourceredditFri, Jul 31, 2026, 07:56 AM
Products·building
SAIPEN: plain-text protocol for AI agent memory continuity
SAIPEN is a small protocol using plain markdown files in a project to record work phase, to-do board, and decision log. Any AI agent that can read files picks up context seamlessly, solving the problem of AI assistants forgetting project state between sessions. Built for developers using AI coding assistants.
SAIPEN is a small protocol using plain markdown files in a project to record work phase, to-do board, and decision log. Any AI agent that can read files picks up context seamlessly, solving the problem of AI assistants forgetting project state between sessions. Built for developers using AI coding assistants.
12 engagement1 sourceredditTue, Jul 21, 2026, 11:52 PM
Social·Community
HN's "plan mode is dead" post sparks debate: coding agents still plan, but invisibly
An HN post titled "plan mode is dead" (~560 points) argues separate plan modes in coding agents are on the way out, because models now explore a repo well enough to plan inside the loop and users skip long generated specs anyway. A Reddit commenter agrees the 300-line specs go unread but argues planning hasn't disappeared — it became invisible, scattered across dozens of undocumented agent decisions. Their fix is a decisions file the agent appends to, one line per decision with the reason, replacing the upfront plan.
An HN post titled "plan mode is dead" (~560 points) argues separate plan modes in coding agents are on the way out, because models now explore a repo well enough to plan inside the loop and users skip long generated specs anyway. A Reddit commenter agrees the 300-line specs go unread but argues planning hasn't disappeared — it became invisible, scattered across dozens of undocumented agent decisions. Their fix is a decisions file the agent appends to, one line per decision with the reason, replacing the upfront plan.
6 engagement1 sourceredditSun, Sep 27, 2026, 07:50 AM
Social·Community
Our agent saved 80,000 characters of lessons. The next agent never read them.
A team added a feature saving reviewer feedback from coding agent tasks so future agents could learn from it. Within 12 hours one project accumulated 20 entries (~80,000 characters), but the next agent never read that field — it only read coding standards and pinned notes. The team disabled automatic appending and now treats rule extraction and connection to tasks as separate work.
A team added a feature saving reviewer feedback from coding agent tasks so future agents could learn from it. Within 12 hours one project accumulated 20 entries (~80,000 characters), but the next agent never read that field — it only read coding standards and pinned notes. The team disabled automatic appending and now treats rule extraction and connection to tasks as separate work.
22 engagement1 sourceredditMon, Sep 28, 2026, 01:21 PM