← ClaudeAtlas

birdwatchlisted

Live watcher mode for product/dev sweeps. Use when the user asks an agent to watch an app, tail logs, verify APIs/DB state, capture experience notes, track discrepancies, or act as a scribe/observer while they test a flow.
kennykankush/skillpack · ★ 0 · AI & Automation · score 60
Install: claude install-skill kennykankush/skillpack
# Birdwatch - Live Evidence Watcher ## When to use Use this skill when the user asks for any version of: - "watch while I test this" - "tail the backend and tell me if anything deviates" - "be a logger / watcher / scribe for this UX sweep" - "verify the DB/API/ledger while I click through" - "bookmark my findings as I dump thoughts" - "help me run lifecycle states and record what breaks" - "sniff out small discrepancies" Birdwatch is a role, not a single command. It can last across a live testing session. ## Purpose Birdwatch turns the assistant into an evidence-backed observer during live app, product, infra, or repo work. The job is to keep the user oriented while they test: - what is running - what state the app/data is actually in - what changed after a user action - which UI friction is real - which backend/API/DB state confirms or contradicts the UI - what should be logged for later issue filing or implementation Do not guess from memory when live evidence is cheap to inspect. ## Harness-agnostic contract This workflow must work in Codex, Claude Code, or any future agent host. The host may differ, but the core inputs are the same: - watch scope: app, repo, route, lifecycle, test session, issue, PR, or product surface - allowed actions: observe-only, write-log, run endpoints, mutate local state, file issues, or implement fixes - evidence sources: terminal logs, process list, API responses, DB queries, browser console, screenshots, files, git state, CI, PRs,