AI Integration

Bowerbird is designed to work seamlessly with AI coding assistants.

JSON Mode

Machine-readable support is command-specific: use --output json and --json only where that command’s help lists them. The canonical discovery workflow and the main note mutation workflows are scriptable:

# AI reads vault state
bwrb list task --output json

# AI creates notes
bwrb new task --json '{"name": "Generated task", "status": "backlog"}'

Agent Skill

For AI assistants that support skills (like OpenCode), Bowerbird provides a skill document at docs/skill/SKILL.md with:

• Command patterns for common operations
• JSON mode examples
• Best practices for note management

Use Cases

Inbox Processing

AI can triage notes:

# Read unprocessed items
bwrb list idea --where "status = 'raw'" --output json

# Update after review
bwrb edit "My Idea" --json '{"status": "developing"}'

Daily-Note Sweep (Coverage)

The goal: ramble into daily notes every day and know nothing got swept under the rug. The convention is a single boolean field on your daily-note type that records whether a note has been reviewed/swept for extractable items (tasks, ideas, people to link). bwrb itself ships no daily-note type — types are defined per-vault — so you declare the field in your schema:

{
  "types": {
    "daily-note": {
      "output_dir": "Daily Notes",
      "fields": {
        "reviewed": {
          "prompt": "boolean",
          "description": "Whether this note has been swept for tasks, ideas, and people to extract. Absent or false means not yet swept."
        }
      }
    }
  }
}

The workflow:

1. New daily notes start without reviewed (or with reviewed: false).
2. An agent (or you) sweeps a note, extracts what matters, then stamps reviewed: true.
3. At any time, list the notes that still need a sweep.

The recipe — find unswept notes:

bwrb list --type daily-note --where "reviewed != true"

Use != true, not == false. A note that has never been touched has no reviewed field at all, and reviewed == false only matches notes where the field is explicitly false — it silently skips the never-reviewed notes, which are exactly the ones most likely to have been swept under the rug. reviewed != true matches both the explicit false and the missing-field cases.

Save it as a dashboard so the un-swept queue is one command away:

# Save the query as a reusable dashboard...
bwrb list --type daily-note --where "reviewed != true" --save-as unswept-daily-notes

# ...then run it any time
bwrb dashboard unswept-daily-notes

Stamping a note as swept is a normal edit:

bwrb edit "Daily Notes/2026-06-20" --json '{"reviewed": true}'

The risk in this loop is the agent forgetting to stamp reviewed: true. That is prompting discipline, not a bwrb feature — but the reviewed != true query is the deterministic backstop that makes a forgotten stamp visible instead of lost.

Content Generation

AI can create structured notes:

bwrb new task --json '{
  "name": "Research topic X",
  "status": "backlog",
  "priority": "medium"
}'

Batch Analysis

AI can analyze vault contents:

# Export for analysis
bwrb list --output json > vault-export.json

Future: AI Commands

Post-v1.0, Bowerbird will include dedicated AI commands for:

• Ingest (process external content)
• Suggest (recommend connections)
• Summarize (generate overviews)

See Also

• JSON Mode
• Product Roadmap