Maintain persistent observational memory for autonomous subagents via MEMORY.md.
npx skills add https://github.com/fourcolors/luna --skill subagent-memory
Please help me install this Agent Skill. Skill: subagent-memory Source: https://github.com/fourcolors/luna/tree/main/seeds/skills/subagent-memory Command: npx skills add https://github.com/fourcolors/luna --skill subagent-memory
This Skill helps autonomous subagents retain useful context across invocations without letting their memory files become noisy, repetitive, or too large for prompt injection.
Tell the subagent to use subagent-memory so it rereads MEMORY.md, records new observations at task end, and compresses the file before it grows too large.
High-intent search queries and answers about installing and using this skill.
To maintain persistent memory for autonomous subagents, use a single MEMORY.md file per agent to reload historical context at the start of each task and append dated observations at the end.
The best way to compress subagent memory is to run a reflection-based compression pass before the MEMORY.md file approaches the 200-line harness limit, preserving the most important knowledge.
Priority-emoji bullet logging works by appending dated bullets with specific emojis to capture critical lessons, useful facts, and completed milestones in chronological order within MEMORY.md.
Yes, subagent-memory works without external dependencies or components, relying solely on markdown file manipulation to record dated learnings and preserve chronological context across repeated tasks.
Your autonomous subagent loses repository-specific patterns because it lacks persistent observational memory, which can be fixed by rereading MEMORY.md at task start and recording new observations at task end.
You should not use a single markdown file for subagent context retention if your agent's observations exceed the 200-line harness limit without compression, as this degrades prompt injection signal quality.
agusmdevEmit one JSON wide-event log per request per service.
bostonaholicTrace execution paths and document architectural patterns in codebases.
lguibrInstrument applications to generate LangSmith traces for LLM observability.
kriegcloudPersist learnings across sessions in structured markdown memory files.
