Reading & Triage

Inbox

GitHub Edit page

Inbox

birdclaw inbox is the unified triage queue. It mixes mentions and DMs, applies heuristic and (optionally) OpenAI-driven scoring, and lets you filter low-signal items out so the high-context stuff floats up.

#Default

birdclaw inbox --json

Returns a flat list of actionable items, ranked by:

  • explicit reply state (unreplied first)
  • recency
  • author influence (follower count plus the derived score)
  • low-signal heuristic (templated AI cadence, generic praise, link-only spam → demoted)

#By kind

birdclaw inbox --kind mentions --limit 10 --json
birdclaw inbox --kind dms --limit 10 --json
birdclaw inbox --kind mixed --limit 20 --json

mixed is the default. The mixed queue interleaves mentions and DMs while preserving the same ranking heuristic.

#OpenAI scoring

Pass --score to refresh stored OpenAI scores before listing. This requires OPENAI_API_KEY in the environment (or in ~/.profile):

birdclaw inbox --score --hide-low-signal --limit 8 --json
birdclaw inbox --score --kind mentions --min-score 60 --limit 12 --json

What --score does:

  • batches items that do not yet have a fresh score
  • sends them to OpenAI with a structured prompt that asks for a 0–100 actionability score and a one-line reason
  • stores (score, reason, model, scored_at) against the canonical record
  • never mutates the canonical row itself

The raw mention or DM is the source of truth. Scores are overlays.

Without OPENAI_API_KEY:

  • --score is a no-op and prints a warning to stderr
  • the heuristic ranker still works
  • --hide-low-signal and --min-score still filter on cached scores when present

#Filters

  • --limit <n> — max items to return
  • --kind mentions|dms|mixed
  • --replied / --unreplied
  • --score — refresh OpenAI scores before listing
  • --min-score <n> — only include items at or above this score
  • --hide-low-signal — drop items flagged by the low-signal heuristic OR scored under the low-signal threshold
  • --account <id>

#Web UI

The Inbox lane wraps the same query. It exposes the score reason in the row hover and lets you re-score on demand without dropping back to the CLI.

#How to think about scoring

OpenAI scoring is an overlay, not a verdict. Use it to:

  • bucket noisy days — "show me anything OpenAI thinks is at least 50/100"
  • demote obvious slop — --hide-low-signal plus --min-score 30 clears the worst cases
  • batch decisions — pre-rank a 200-item queue, then walk top-to-bottom

Do not use it to:

  • decide whom to block — read the actual content first
  • replace your own judgment on borderline cases — fall back to profiles replies for templated-AI checks

#See also

  • Mentions — mentions export is the same queue without ranking
  • DMs — list/sync/search reference
  • Configuration — OPENAI_API_KEY setup