Agent Memory Instructions

Use AGENT-MEM-RULES.md as the canonical source of truth. Read and follow it every session.

Project name: agent-mem

Mandatory behavior at session start:

• Resolve the repository root.
• Resolve the project name from the repository root and keep it consistent.
• Load memory before planning, coding, or answering historical questions.
• Prefer saved memory over old chat history.
• If the user pastes a watch-generated handoff prompt, execute that handoff before anything else.
• If the user pastes a migration handoff prompt, execute it first, save a structured summary, then continue normal implementation.

If MCP tools are available:

• call query_memory first
• call summarize_to_obsidian when needed

If MCP tools are unavailable:

• read the latest memory artifact directly
• prefer .agent-memory/active.md over older memory when fallback mode is active
• still produce the structured summary in chat if direct save tools are unavailable

Mandatory summarization triggers:

• long or repetitive context
• completed milestone
• important design or implementation decision
• multiple-file change
• session handoff or stop point
• a watch-generated handoff prompt pasted by the user
• a migration handoff prompt pasted by the user

Required summary contents:

• Goal
• Outcome
• Key decisions
• Files changed
• Open tasks or blockers
• Next prioritized steps

Safety rules:

• never hallucinate historical decisions
• never write secrets into memory
• if memory conflicts with the codebase, say so explicitly
• prefer verified state over guessed state
• in Obsidian mode, keep summaries dense, specific, and useful as future engineering notes

---

Source: atharvavdeo/agent-mem · 3★