Board Settings and Agent Instructions
Configure project-wide agent behaviour, shared memory, and technical documentation from your board settings. These settings are available to every board member and feed directly into how AI agents interact with your tasks.
Opening Board Settings
Click the gear icon on your board header to open the settings dialog. The dialog has six tabs: General, Appearance, AGENTS.md, Memory, Technical, and Danger.
AGENTS.md
The AGENTS.md tab lets you write project-specific instructions that are shared with all board users via the AGENTS.md file in their workspace.
What it does
When a user connects the VS Code extension to your board, the extension manages a sentinel section inside the workspace's AGENTS.md file. Any instructions you enter in this tab are injected into that managed section automatically.
This means you can define project conventions, tool preferences, coding standards, or workflow rules once on the board, and every team member's AI agent will receive them — without anyone needing to manually edit AGENTS.md or keep it in sync.
When it updates
The extension refreshes the AGENTS.md sentinel at the same points it updates the task instruction file (.agentkanban/INSTRUCTION.md):
- when the extension activates
- when a task is selected or changed
- when a board is selected
- when the setup wizard completes
If the agent instructions are empty, the sentinel includes only the core directive block. When instructions are present, they appear under a ## Board Agent Instructions heading inside the managed section.
What to put here
Good candidates for board-level agent instructions:
- project type and framework ("This is a SvelteKit project using Tailwind CSS v4")
- testing conventions ("Use vitest for unit tests and playwright for e2e tests")
- coding standards and linting rules
- repository structure guidance
- deployment or CI/CD notes the agent should be aware of
- references to important files like
CONTRIBUTING.mdorSPEC.md
Keep it focused. These instructions are injected into every agent session, so favour concise guidance over lengthy documentation. Use the Memory and Technical tabs for larger reference material.
Character limit
The field accepts up to 10,000 characters.
How it appears in AGENTS.md
The managed section in the workspace looks like this when instructions are configured:
<!-- BEGIN AGENT KANBAN — DO NOT EDIT THIS SECTION -->
## Agent Kanban
**CRITICAL: You MUST read `.agentkanban/INSTRUCTION.md` before responding to ANY message.**
...
## Board Agent Instructions
<your instructions from the AGENTS.md tab appear here>
<!-- END AGENT KANBAN -->
Content outside the sentinel markers is never modified by the extension. You can safely have your own sections in AGENTS.md above or below the managed block.
Memory
The Memory tab is a persistent shared document for project conventions, key decisions, and useful context that persists across tasks.
How agents use it
When an agent resumes a task via @kanban /resume, the board memory is loaded at the top of the injected context. Agents are also instructed to proactively update board memory when they discover durable technical context, conventions, or decisions that should survive beyond the current task.
Board memory is also available via MCP tools (get_board_memory, update_board_memory), so agents can read and write to it during implementation.
Local cache
The VS Code extension maintains a read-only local cache of the board memory at .agentkanban/MEMORY.md. This file is synced automatically from the remote board whenever the extension activates, a task is selected, or memory is updated on the server (via the real-time WebSocket connection).
Agents can read .agentkanban/MEMORY.md directly without making an API call. To update board memory, agents use the MCP tool update_board_memory — the local cache is overwritten on the next sync. Agents should never edit .agentkanban/MEMORY.md directly.
The .agentkanban/ directory is gitignored, so the local cache does not appear in version control.
What to put here
- project conventions and patterns
- key architectural decisions
- dependency notes or version constraints
- build quirks or known issues
- anything an agent should know at the start of every task
Character limit
The field accepts up to 50,000 characters.
Technical
The Technical tab holds implementation details for both agents and humans.
How agents use it
Agents can read the full technical document or request specific sections by heading via the MCP get_board_technical tool. This is useful for larger documentation that would be too noisy to inject into every session — agents fetch the sections they need on demand.
Agents are expected to update the technical document when making changes that affect architecture, deployment, or implementation details.
Local cache
The VS Code extension maintains a read-only local cache of the board technical document at .agentkanban/TECHNICAL.md. This file is synced automatically from the remote board whenever the extension activates, a task is selected, or the technical document is updated on the server.
Agents can read .agentkanban/TECHNICAL.md directly as a local file. To update the technical document, agents use the MCP tool update_board_technical (which supports both full replacement and patch mode). The local cache is overwritten on the next sync.
This replaces the previous pattern of maintaining TECHNICAL.md at the workspace root. Existing workspace-root files are left as-is — the extension does not migrate or remove them.
What to put here
- architecture documentation
- API documentation
- deployment and infrastructure notes
- database schema notes
- integration details
Character limit
The field accepts up to 100,000 characters.
Memory vs AGENTS.md vs Technical
| AGENTS.md | Memory | Technical | |
|---|---|---|---|
| Injected into | AGENTS.md sentinel in workspace |
Resume context at task start | On-demand via MCP |
| Local cache | N/A | .agentkanban/MEMORY.md |
.agentkanban/TECHNICAL.md |
| Visibility | Every agent session (via file) | Every resume + local file | Local file + MCP on demand |
| Best for | Project-wide coding rules and conventions | Cross-task decisions and patterns | Detailed reference documentation |
| Size | Up to 10,000 chars | Up to 50,000 chars | Up to 100,000 chars |
| Updated by | Humans via board settings | Agents and humans | Agents and humans |
Use AGENTS.md for instructions that should be present in every session, regardless of task. Use Memory for decisions and context that accumulate over time. Use Technical for deep documentation that agents pull on demand.
Permissions
Only board members with Edit permission can modify board settings. Members with read-only access can view the settings but cannot change them.
Appearance
The Appearance tab controls visual preferences for the board.
Board background colour
Set a background colour for your board view. This is a personal preference — it only affects your own view, not other board members.
Choose from the five preset pastel colours (Rose, Peach, Lemon, Mint, Sky), or use the system colour picker at the end of the palette to select any custom colour. Click the "None" button to remove the background and return to the default.
Tag colours
Tags are board-scoped labels you can assign to tasks. Each tag is automatically assigned a colour from the preset palette when created. You can change tag colours from the Appearance tab — these changes apply to all board members.
For each tag, the settings dialog shows:
- the tag name as a coloured pill
- five preset colour swatches to pick from
- a system colour picker for custom colours
- a delete button to remove the tag from the board
Deleting a tag removes it from the board registry and from every task that uses it. A confirmation dialog is shown before deletion.
Tags are created from the task detail page by typing a name in the tag input field. Tag names must be lowercase, alphanumeric with hyphens only, and a maximum of 20 characters. See Editing Tasks for details on adding tags to tasks.