My problem with all of these kinds of things is that memory architecture is the whole point.
If all I needed was common memory space I'd just use symlinks and cloud sync, Obsidian or whatever.
But the context shaping is what makes memory useful in the first place, so just doing one stop shop memory is IMO about as useful as plain old markdown...
Plain old markdown with links is exactly what I use. But I split memory in 3 parts
1. a log of user comments, all of them collected in a chat_log.md file which is used to validate the agent still follows user intent
2. task level memory, a task.md file starts as intent, becomes plan, workbook with inlined results for each gate, and finally remains as documentation; same task.md is passed to judge agents and back; a project can have 100+ tasks logged in, the task is the natural unit of work
3. a project state mind map which tracks current known value of all decisions and intents, it is used to start the agent new sessions and updated after each task
This is it, a memory of user intent, one of agent work, and one of project state. I used this setup in over 100 projects. My harness uses hooks to constrain the agent
1. can't edit code files without an open task
2. the next current gate ("[ ] plan item") is displayed in the post tool use hook as an instruction pointer, auto-advances as current gate gets checked; this keeps the agent straight even in task.md files with 100s of gates
3. can't close a task until all gates are solved
4. a post user message hook appends to chat_log.md, which is cheap
The harness can be refined by optimizing task.md plan structure for various types of tasks, based on reflection. We can do reflection because we keep the 3 memory types.
Memory works for my harness because I don't split the place where memory sits from the place where work gets done and tracked. The plan items are this place. A single gate starts as intent and ends as a log like of work done. The task.md file is edited as it is solved.
One of the first fun usages of LLMS by my non-technical friends back in days of chatgpt 3.5 or so, was impersonations: write this in style of Snoop Dog, create a bedtime story in style of Dr Seuss, explain this like Carl Sagan, etc.
So how come today, even people who do content for a living, not necessarily programmers but writers and "influences", just generate default LLM-style content? I see exactly what triggers you in the github writing linked, and it feels so easy to fix even using AI itself.
It's at the point where I'm like "if you can't even bother to rewrite or mask or ask LLM to give it some personality, why are you asking me to bother reading it" :-/
There have always been plenty of “9-5” programmers who will do the minimum to get by. Even just being on HN is somewhat of a bubble of people more interested in keeping up with tech than the average Joe building websites.
The things you've described don't actually mask the LLM smell to all but the most naive. I see all kinds of "voices" in a variety of LLM-generated articles, and yet they still have numerous tells. You can spray perfume on a smoker but they still smell like a smoker.
We don't need Cursor, we have Curso at home
I sync my Obsidian and agent sessions with RAG and have added a skill to search RAG if the info is not in the context. Works perfectly.
My problem with all of these kinds of things is that memory architecture is the whole point.
If all I needed was common memory space I'd just use symlinks and cloud sync, Obsidian or whatever.
But the context shaping is what makes memory useful in the first place, so just doing one stop shop memory is IMO about as useful as plain old markdown...
Plain old markdown with links is exactly what I use. But I split memory in 3 parts
1. a log of user comments, all of them collected in a chat_log.md file which is used to validate the agent still follows user intent
2. task level memory, a task.md file starts as intent, becomes plan, workbook with inlined results for each gate, and finally remains as documentation; same task.md is passed to judge agents and back; a project can have 100+ tasks logged in, the task is the natural unit of work
3. a project state mind map which tracks current known value of all decisions and intents, it is used to start the agent new sessions and updated after each task
This is it, a memory of user intent, one of agent work, and one of project state. I used this setup in over 100 projects. My harness uses hooks to constrain the agent
1. can't edit code files without an open task
2. the next current gate ("[ ] plan item") is displayed in the post tool use hook as an instruction pointer, auto-advances as current gate gets checked; this keeps the agent straight even in task.md files with 100s of gates
3. can't close a task until all gates are solved
4. a post user message hook appends to chat_log.md, which is cheap
The harness can be refined by optimizing task.md plan structure for various types of tasks, based on reflection. We can do reflection because we keep the 3 memory types.
Memory works for my harness because I don't split the place where memory sits from the place where work gets done and tracked. The plan items are this place. A single gate starts as intent and ends as a log like of work done. The task.md file is edited as it is solved.
https://github.com/horiacristescu/claude-playbook-plugin
Slop alert
One of the first fun usages of LLMS by my non-technical friends back in days of chatgpt 3.5 or so, was impersonations: write this in style of Snoop Dog, create a bedtime story in style of Dr Seuss, explain this like Carl Sagan, etc.
So how come today, even people who do content for a living, not necessarily programmers but writers and "influences", just generate default LLM-style content? I see exactly what triggers you in the github writing linked, and it feels so easy to fix even using AI itself.
It's at the point where I'm like "if you can't even bother to rewrite or mask or ask LLM to give it some personality, why are you asking me to bother reading it" :-/
People are lazy and they don’t know.
I regularly blow the minds of teams I work with by mentioning you can use custom skills, hooks, etc or have the LLM ask you questions.
How do these people keep up with technology? Those are basic skills now.
There have always been plenty of “9-5” programmers who will do the minimum to get by. Even just being on HN is somewhat of a bubble of people more interested in keeping up with tech than the average Joe building websites.
The things you've described don't actually mask the LLM smell to all but the most naive. I see all kinds of "voices" in a variety of LLM-generated articles, and yet they still have numerous tells. You can spray perfume on a smoker but they still smell like a smoker.
[flagged]