4 replies

Where is the beat place to initiate session rules

✅Solved

if I need to add the prompts as below, the best place to add it is to agents.md - am I correct?

SESSION INITIALIZATION RULE:

On every session start:
1. Load ONLY these files:
- SOUL.md
- USER.md
- IDENTITY.md
- memory/YYYY-MM-DD.md (if it exists)

2. DO NOT auto-load:
- MEMORY.md
- Session history
- Prior messages
- Previous tool outputs

3. When user asks about prior context:
- Use memory_search() on demand
- Pull only the relevant snippet with memory_get()
- Don't load the whole file

4. Update memory/YYYY-MM-DD.md at end of session with:
- What you worked on
- Decisions made
- Leads generated
- Blockers
- Next steps

Solution

Yep — AGENTS.md
AGENTS.md
is the right place for that kind of “operating rules / how to run the session” instruction set.

A couple important nuances in OpenClaw though:

- You can’t enforce “load ONLY these files” just by writing it in
AGENTS.md
AGENTS.md
.
OpenClaw itself injects a fixed bootstrap set into the system prompt each run (by default:

AGENTS.md

AGENTS.md

SOUL.md

SOUL.md

TOOLS.md

TOOLS.md

IDENTITY.md

IDENTITY.md

USER.md

USER.md

HEARTBEAT.md

HEARTBEAT.md

, plus

BOOTSTRAP.md

BOOTSTRAP.md

on brand-new workspaces).
You can verify what’s actually injected with /context list
/context list
.
Docs: https://docs.openclaw.ai/concepts/context

- MEMORY.md
MEMORY.md
is not auto-injected by default (unless you customized things). So your “DO NOT auto-load MEMORY.md” rule is mostly already aligned with how it works; you’re basically telling the agent “don’t go read it unless asked / unless needed”.

- For your “update

memory/YYYY-MM-DD.md

memory/YYYY-MM-DD.md

at end of session” goal: OpenClaw already supports workflows/hooks that help with this:
- The automatic memory flush before compaction (config-driven)
- The bundled session-memory
session-memory
hook (writes a memory note on

/new

/new

)
Docs: https://docs.openclaw.ai/concepts/memory and https://docs.openclaw.ai/hooks

If you truly want to change which files are injected (so it literally only injects your chosen subset), that’s more advanced: it’s done via a hook on