Agents

Banco supports agentic workflows. Upon initialization, three files are created for agent context:

• .banco/BANCO.md — banco-managed file with detailed instructions on how agents should work with this project; regenerated by banco on init
• AGENTS.md — reads .banco/BANCO.md and is meant for user-defined instructions; never overwritten by banco
• CLAUDE.md — reads AGENTS.md; never overwritten by banco

.banco/BANCO.md instructs agents to:

• Run banco context first to get a JSON snapshot of the project state (providers, modules, items, and the labels required to build commands)
• Always use banco new to create items — never write files directly
• Derive label names and values from the context rather than guessing
• Use status and tags metadata on synced provider items to make informed decisions — e.g. prioritize open issues, filter by label, skip closed tasks when working on active work

Skills

Banco can install agent skills into a project with banco skills install <agent>. For the claude agent, skills are written to .claude/skills/<skill-name>/SKILL.md — the location Claude Code discovers project skills from — so they become invokable immediately. See the skills command for details.

Example prompts

• How many tasks do I have in the backlog?
• Move task “Write full specs” to done
• Create a note about today’s meeting in the meetings folder
• Show me all bookmarks tagged under tools/rust
• What tasks are currently in progress?