Agent-Browser PRD Playbooks

Use this skill to draft ready-to-run agent-browser command blocks for every Daedalus PRD user story in prd.json (US-001..US-007). It does not run the browser; it outputs scripts to execute manually.

Prerequisites

•agent-browser --version succeeds; if missing: npm install -g agent-browser
•Chromium downloaded: agent-browser install (Linux: agent-browser install --with-deps)
•App base URL known (e.g., http://localhost:3000 or deployed URL)
•
Auth: choose one
- •Header-based: --headers '{"Authorization":"Bearer <token>"}'
- •Saved state: log in once then agent-browser state save auth.json; reuse with state load auth.json
•Keep refs-first flow: always run snapshot -i before interacting; re-snapshot after navigation.

Inputs

•prdPath: path to PRD JSON (default prd.json at repo root)
•baseUrl: app URL
•authHeaders or authStatePath: for authenticated areas
•Optional: feature flags or tenant identifiers per story

Steps

•Read prdPath and list userStories ids/titles.
•For each story, set a dedicated session --session <id> to isolate cookies.
•Login (if needed) once per session; then state save for reuse.
•Build a refs-first command block: open → snapshot -i → interactions → wait → verify (url/title/text) → screenshot --full.
•Prefer semantics (find role/text/label) when refs are unstable; avoid destructive actions (billing changes) unless explicitly approved.
•Capture evidence: final URL, title, key text via get text @eN, and screenshot --full <id>.png.

Command Playbooks per PRD Story

Replace <BASE_URL> with your target, provide real selectors/refs from the latest snapshot -i, and fill any placeholders like <CSV_PATH>.

US-001 – Enrich a CSV of leads (enrich batch)

bash

agent-browser --session US-001 open <BASE_URL>/enrich/batch
agent-browser snapshot -i
# TODO: click upload control
agent-browser find role button click --name "Upload" || agent-browser click @e?
agent-browser snapshot -i
agent-browser fill @e? "<CSV_PATH>"  # if file picker not allowed, attach via UI flow manually
agent-browser snapshot -i
agent-browser click @e?  # Submit/process
agent-browser wait --text "Enrichment complete" || agent-browser wait --text "Processing"
agent-browser get url
agent-browser get title
agent-browser get text @e?  # Verify row count or success message
agent-browser screenshot --full US-001.png

US-002 – Enrich a single lead (enrich single)

bash

agent-browser --session US-002 open <BASE_URL>/enrich
agent-browser snapshot -i
agent-browser find label "Email" fill "demo@example.com" || agent-browser fill @e? "demo@example.com"
agent-browser find label "Name" fill "Demo User" || true
agent-browser find role button click --name "Enrich" || agent-browser click @e?
agent-browser wait --text "Enrichment complete" || agent-browser wait --text "Profile"
agent-browser get url
agent-browser get title
agent-browser get text @e?  # Verify role/company output
agent-browser screenshot --full US-002.png

US-003 – Extract brand identity from a URL (extract)

bash

agent-browser --session US-003 open <BASE_URL>/extract
agent-browser snapshot -i
agent-browser find label "Website" fill "https://example.com" || agent-browser fill @e? "https://example.com"
agent-browser find role button click --name "Extract" || agent-browser click @e?
agent-browser wait --text "Colors" || agent-browser wait --text "Fonts"
agent-browser get url
agent-browser get title
agent-browser get text @e?  # Verify brand attributes present
agent-browser screenshot --full US-003.png

US-004 – Monitor critical pages for changes (observe)

bash

agent-browser --session US-004 open <BASE_URL>/observe
agent-browser snapshot -i
agent-browser find label "URL" fill "https://example.com/pricing" || agent-browser fill @e? "https://example.com/pricing"
agent-browser find label "Frequency" fill "daily" || agent-browser select @e? "Daily"
agent-browser find role button click --name "Create monitor" || agent-browser click @e?
agent-browser wait --text "Monitor created" || agent-browser wait --text "Next run"
agent-browser get url
agent-browser get title
agent-browser get text @e?  # Verify monitor row appears
agent-browser screenshot --full US-004.png

US-005 – Run scouts to discover new signals (scout)

bash

agent-browser --session US-005 open <BASE_URL>/scout
agent-browser snapshot -i
agent-browser find label "Query" fill "site:news.ycombinator.com fintech" || agent-browser fill @e? "site:news.ycombinator.com fintech"
agent-browser find role button click --name "Run scout" || agent-browser click @e?
agent-browser wait --text "Results" || agent-browser wait --text "New findings"
agent-browser get url
agent-browser get title
agent-browser get text @e?  # Verify deduped results listed
agent-browser screenshot --full US-005.png

US-006 – Run research sessions with agents (agent)

bash

agent-browser --session US-006 open <BASE_URL>/agent
agent-browser snapshot -i
agent-browser find label "Prompt" fill "Summarize competitor positioning for Acme" || agent-browser fill @e? "Summarize competitor positioning for Acme"
agent-browser find role button click --name "Run" || agent-browser click @e?
agent-browser wait --text "Thinking" || agent-browser wait --text "Sources"
agent-browser get url
agent-browser get title
agent-browser get text @e?  # Verify answer section populated
agent-browser screenshot --full US-006.png

US-007 – Manage plans, usage, and billing (platform/billing)

bash

agent-browser --session US-007 open <BASE_URL>/billing
agent-browser snapshot -i
# Read-only unless explicitly approved; avoid purchasing/upgrading without consent
agent-browser get text @e?  # Current plan/limits
agent-browser screenshot --full US-007.png

Session & Resilience

•Use --session <id> per story to keep cookies separate.
•After login in a session: agent-browser state save auth-<id>.json; preload with state load auth-<id>.json.
•On failure/timeouts: rerun snapshot -i, use wait --text or wait --url, then retry the action.
•If MFA/CAPTCHA blocks progress, stop and request human input.

Outputs

•Per-story command block (above) tailored with real refs/selectors.
•Evidence per story: final URL, title, key text assertion, and screenshot path <id>.png.

Verification

•Validate command syntax (no missing args; refs exist from latest snapshot).
•Confirm non-destructive intent (especially billing); abort upgrades/changes unless explicitly approved.
•Ensure auth method is scoped (headers per origin or session state file).
•Re-snapshot after each navigation before using @e? refs.