AI agent
Agent long-term memory
Put "who you are, habits, constraints" into `.kition/agent.md` — loaded into every chat.
Why this exists
The agent does not know who you are, what you're building, or how you like things written. Restarting every chat from scratch gets old — you should not retype "I write B2B SaaS copy; keep it direct and short" each time.
.kition/agent.md is that memory. Kition injects it into the system prompt at the start of every session. Edit once, applies everywhere.
A solid example
# Agent context
## Who I am
- Freelance writer focused on B2B SaaS topics.
- Primary audience: technical buyers in North America.
- I dislike marketing jargon ("synergy", "leverage", "unlock").
## Voice
- Direct, short sentences. Active voice.
- No emoji unless I ask.
- Avoid Oxford comma in body copy; keep in lists.
## Workflow
- Drafts live under `drafts/`. Finals under `published/`.
- Always run `scripts/check-claims.sh <file>` after writing claims-heavy content.
- Cite sources inline with footnote style: `[^1]`.
## Tools
- Prefer `web_fetch` over `web_search` when I name a domain.
- Never `Bash rm` without confirming — ask first.
## Things to never do
- Suggest emoji.
- Use the word "leverage" as a verb.What it does
At each chat start, the agent loads this as system context — saves you from repeating "I am…, please…" every time. It influences every turn, including how plan mode phrases its plans.
In team vaults this file is tracked by git — effectively a team "AI working agreement". New teammates clone the repo and inherit the same conventions.
Layered memory
~/.kition/agent.md— global, for "who I am, my voice"<vault>/.kition/agent.md— vault-level, for "this project's conventions"- A subdirectory
agent.md— stacks when you work in that subtree - Load order: global → vault → subdir, later wins on conflicts
What not to put in it
Don't dump your whole knowledge base here — every chat pays tokens for it. A few KB is fine, tens of KB slow the first response and waste money. Reference specific knowledge with @ on demand, or mount an MCP server.
Don't put credentials here — by design this file goes to the provider. Tokens and passwords belong in Keychain.