add airline

Data Community

Skill notes

Purpose

Every program page on crazy4points (airline, hotel, etc.) goes through the same end-to-end pipeline: research, draft, hedge, cross-fact-check, author in admin, verify live, submit to search engines, capture sources, wire press-room RSS into Scout, and (for hotels) seed the per-property Decision Engine dataset. This skill enforces that pipeline so it happens consistently. Target automation level — Claude drives end to end. The user does TWO things:

Workflow — surface one step at a time

When triggered, FIRST do this: Read plans/airline-page-runbook.md end-to-end Read plans/sources/_TEMPLATE.md so the source doc structure is fresh Identify the airline from the user's trigger phrase THEN announce: "Starting <airline> — Step 1 first." Don't dump the full runbook. The user wants one step at a time.

Step 1 — Research the program (automated via research-program.mjs)

Primary workflow: scrape the program's own pages directly via the research orchestrator, then WebSearch for gaps the orchestrator flags. User paste-in is only a last resort. Do this: Announce: "Step 1 — running research-program.mjs for <slug>." Confirm the program has scrape_urls seeded: If scrape_urls is {}, write a small migration seeding the canonical URLs (see existing migrations 099, 107 for shape). Standard keys per type:

Step 2 — Draft hedged content + emit Copilot fact-check block (Claude does this)

LEAN WORKFLOW — do NOT dump a full 10-field combined preview to the user. The user does NOT want to read scattered drafts inline. Instead, after drafting internally: Output ONE thing only: the Copilot fact-check block (see Step 3 below for the exact format). The user pastes that into Copilot or ChatGPT, gets back per-claim validations, and shares the response. Internal: keep the full 10-field draft in your head / scratch context, but DO NOT show the user. The Copilot block is the only artifact the user sees in Step 2.

Step 3 — Cross-fact-check via Copilot (THE Step 2 output)

After Step 2 drafts are signed off (combined preview approved), give the user one consolidated block they can paste into Copilot — never make them copy six separate fields one by one. The block has two parts: the prompt, then the content. ALL-CAPS section labels help Copilot anchor each fact-check to the right field. Format to deliver to the user: INTRO [intro draft] LOUNGE ACCESS

Fluxly profile Runtime and permissions are inferred by Fluxly static checks, not author declarations or runtime test results.
Domain
Data
Author-declared agents
No explicit declaration found; this is not inferred or tested compatibility
Static check
88 / 100 · heuristic scan, not runtime safety proof
Runtime evidence
Not run; no isolated runtime test evidence is currently recorded
Author / version / license
@jillyz250 · no license declared
Token usage
Heavy
Setup complexity
Guided setup
External API key
Not required
Operating systems
macOS · Linux · Windows
Runtime requirements
Node.js
Detected file/system behavior
  • Read-only
  • Write / modify
  • Shell exec
Network behavior
External requests
Install commands
None (reference only)

Heads up: Tool permissions are not declared; actual access depends on the runtime.

Output preview add-airline.preview
# Step 3 — Cross-fact-check via Copilot (THE Step 2 output)

**Copilot / ChatGPT fact-check prompt:**

> Fact-check the following content for the [Carrier Name] carrier page. Today's real-world date is [Month DD, YYYY].
>
> **REQUIRED METHOD:** Use web search / browsing tools to verify each claim against current sources. Do NOT rely on training data alone — for events from 2024 onward, training data may be stale or incomplete. If you cannot search, say so up front and stop.
>
> **For each claim:**
> 1. Search the open web with a 2026-dated source filter when relevant.
> 2. Cite the source URL + publication date for any disagreement, AND for confirmation of any recent (post-2024) event.
> 3. If you can't find a 2026-dated source that confirms OR denies a claim, return "❓ UNVERIFIED — needs current source." Do NOT mark a claim incorrect based on training-data memory alone — that's how you fabricate denials of real events.
>
> **Output format per claim:**
> > [claim] — [✅ CORRECT / ⚠️ NEEDS CLARIFICATION / ❌ INCORRECT / ❓ UNVERIFIED] — [URL + publication date if you found one]
>
> Flag only what is factually wrong by current evidence. Do NOT comment on style or voice.

---

**Content block to paste:**

Discussion

Powered by GitHub Discussions. Sign in with GitHub to comment, react, or subscribe.