Skip to main content

Minds

A mind card is a harness card that contributes to the agent's cognitive layer — the beliefs, persona, memory, and hook policies that shape how the AI thinks and behaves — rather than (or in addition to) contributing skills and MCP servers.

What a mind card contributes​

Content typeFileEffect
BeliefsBELIEF.mdFactual assertions composed into the system prompt
PersonaPERSONA.mdBehavioral and stylistic identity layer
Memory (L4)l4/<entry>/Short-term project memory (markdown format)
Memory (L5)l5/<entry>/Mid-term accumulated knowledge (JSONL or mixed)
Memory (L6)l6/<entry>/Long-term durable facts (structured)
Hook policieshooks/<name>/policy.tsRuntime tool intercept and observation policies

A mind card can contribute any combination of these. A card that contributes only skills and MCP servers is a regular harness card, not a mind card, though the distinction is informal — the same card format supports both.

The active mind stack​

The active mind stack is an ordered list of mind card names that drwn write projects into downstream tool state. It is declared in .agents/drwn/config.json under activeMinds:

{
  "version": 1,
  "cards": ["@team/base", "@team/frontend"],
  "activeMinds": ["@team/base", "@team/frontend"]
}

Order matters: beliefs and persona content from earlier cards in the stack take precedence during composition.

Default behavior​

When activeMinds is absent from project config, all installed mind cards are active. This default ensures installed minds are immediately useful without requiring explicit activation. To pin a subset or control order, use drwn mind use.

Managing the active stack​

# Inspect installed minds and which are active
drwn mind list

# Activate an explicit ordered stack
drwn mind use @team/base @team/frontend

# Remove all active-stack declarations (returns to all-active default)
drwn mind clear

Changes to the active stack take effect on the next drwn write.

Installation​

Mind cards are installed the same way as regular harness cards:

drwn add @team/base
drwn install
drwn mind use @team/base
drwn write

Authoring a mind card​

Mind content is scaffolded via drwn card source subcommands:

drwn card source add-belief @your-handle/mind engineering --visibility public
drwn card source add-persona @your-handle/mind voice --visibility internal
drwn card source add-memory @your-handle/mind context --layer l4 --visibility private --format md
drwn card source add-hook @your-handle/mind audit-tool-calls

See Beliefs, Personas, and Memory and Hook Policies for the content model.

If a mind card declares hook policies, consuming projects must grant trust before the policies are active at runtime:

drwn card trust @team/base --hooks

drwn doctor reports a hookIssues entry for any installed mind card that declares hooks but has not been trusted. See the Diagnostics Model for details.

See also​