Workbench New doc →
🧠

Agent Memory template

Give your agent a memory that outlives the session: identity and standing orders, a conventions sheet, and a decisions log it reads at startup and appends to as it learns — so no session starts from zero.

Use this template → Free · no account · opens an editable copy · live components
Preview

Agent Memory — [agent name]

🤖 READ THIS FIRST

This doc is your durable memory. At the start of every session, read it top to bottom before doing anything else — it’s the context your last session left you. As you work, keep it current: when you learn something durable (a convention, a gotcha, a decision), append it to the right section with a versioned write (If-Match; on 409, re-read and retry). Session memory dies with the session; this doc doesn’t. Keep it tight — this is a briefing, not a diary: durable facts in, session chatter out.

Identity & standing orders

Conventions

The facts you’d otherwise rediscover every session. Add a row the moment you catch yourself learning something twice.

KeyValue
repo[where the code lives, and the branch you work from]
run locally[the command — "npm run dev; needs .env from the password manager"]
tests[how to run them, and which ones are known-flaky]
deploy[how it ships, and who pushes the button]
style[linter, formatter, and the rules the human actually cares about]

Decisions log

Dated and append-only, newest first. One line per decision: what was decided, and why — so future sessions don’t relitigate it.

Open loops

Things in flight that the next session must pick up. Prune ruthlessly — a stale open loop is worse than none.

Built with Workbench — live markdown for humans & agents. Agent API