AgentMemory Context Payload

AgentMemory context payload reference

Note

Status: implemented Plan: PLAN.md (V4 #22) task 1 — canonical context payload + self-describing vault copy. Source of truth: templates/agentmemory-context.md — pasteable body starts at line 19 (# Using my Agent Memory); the lines above are operator instructions, not part of the paste.

The canonical "how-to-use-AgentMemory" payload at templates/agentmemory-context.md — the single source of truth the operator pastes into each agent surface (Claude.ai custom instructions, a Gemini Gem's system instructions, the Antigravity rule). A self-describing copy lives at <vault>/_meta/how-to-use-agentmemory.md so an agent that reaches the vault finds its own usage instructions. Read-only (v1): agents read + query the vault; they never write — capture = suggest an entry for the operator to paste into Obsidian by hand.

⚡ Quick Reference

Question	Answer
Where is the canonical payload?	templates/agentmemory-context.md (agentm) — the source of truth (DC-5).
Where is the self-describing copy?	<vault>/_meta/how-to-use-agentmemory.md (written outside the repo, not in git).
Which surfaces consume it?	Claude.ai / ChatGPT, Gemini, Antigravity — see Use AgentMemory in any agent surface. Claude Code instead receives it via SessionStart/UserPromptSubmit hooks (no paste needed).
Is it host-specific?	No — host-agnostic; no Claude-Code-specific assumptions.
Read or write?	Read-only (v1, DC-2): surfaces read + query; never write. Capture = suggest a paste-ready entry.
What do I actually paste?	The body from line 19 onward (# Using my Agent Memory); the leading HTML comment is operator-only instructions.

Payload sections

The required sections of the payload, in order, as they appear in templates/agentmemory-context.md.

Section (heading in template)	Covers
Intro (# Using my Agent Memory)	Names the vault (AgentMemory/, a GDrive-synced Obsidian vault); states read-the-vault-before-own-memory and the read-only stance up front.
Where the vault is, on your surface	Per-surface path resolution: Claude.ai / ChatGPT → GDrive connector + pinned AgentMemory/; Gemini → native Workspace/Drive access; Antigravity → installer-configured path; Claude Code / local → MEMORY_VAULT_PATH.
Folder map — what's where	personal-private/_always-load/ (global conventions, read first), projects/<slug>/ (_index.md / decisions/ / open-questions/ / _harness/), _idea-incubator/<slug>/, _inbox/, _meta/.
How to read it (priority order)	1) always-load first → 2) project context (_index.md + decisions/) → 3) query by topic; vault wins over the model's general knowledge.
Reading entries correctly	Markdown + YAML frontmatter; core trio kind + status + created; kebab-case slugs/tags; status: active vs superseded; follow [[wikilinks]].
Your boundary — READ-ONLY	Read + query freely, never write; capture = suggest a paste-ready entry and name its home; GDrive sync means you see last-synced state (flag, don't guess).

Related

• Use AgentMemory in any agent — the setup recipe for every surface (Claude.ai · Gemini · ChatGPT · Antigravity).