Allsup Leads Veterans

Allsup Leads — Veterans

The veterans benefits-gap batch for Allsup. Same customer, same proven process as [allsup-leads-ssdi], different vertical + tiers. The product: handle + own-words quote + ≥1 reachable public profile. Graded on REACHABILITY, not contact data. Allsup owns all outreach — we never contact anyone.

The signal: veterans who DESERVE VA benefits but aren't getting them —

• never filed / didn't know they qualified (a huge, findable population)
• denied or under-rated, fighting for what they're owed
• newly eligible under the PACT Act (burn pit / Agent Orange / Camp Lejeune / Gulf War) and unaware.

Quick start

scripts/run.sh            # pull + tier + build the book (default 30 days)
scripts/run.sh --days 45  # wider window for more volume

Then publish the book and draft the email (steps 3–4).

Workflow

1. Pull + tier + build book — scripts/run.sh. Mines veterans-benefits-gap on Reddit + X + TikTok (fresh, no ledger), tiers by need, writes ~/Desktop/VETERANS-LEADS-<date>.{csv,md}, builds .book/index.html. Report the tier counts.
2. Verify + eyeball the CSV. TikTok can leak the occasional creator/coach (they talk about veterans, not as one) despite the first-person + handle filters — glance at the BEST tier and drop any obvious creator/caregiver rows before sending. If TikTok is too noisy for a given run, re-run pulling only Reddit + X (edit SURFACES in scripts/veterans_pull.py).
3. Publish the book (permanent, forwardable): ~/.codex/skills/here-now/scripts/publish.sh <skill>/.book --client codex
4. Draft the email via the Gmail tool — DRAFT only, never send. Use EMAIL.md with this run's counts + book URL. Tell Brent to drag VETERANS-LEADS-<date>.csv in as the attachment. (Recipient: Allsup's David d.doeren@allsup.com, or Pat pat@contatta.com — ask Brent which for this run.)

The rules (same as SSDI — do not violate)

• Reddit is densest and KEYLESS. Oxylabs only speeds it up. Never exclude Reddit.
• Grade on reachability, not contact data. A Reddit/X/TikTok profile IS a reachable channel.
• NEVER run the heavy resolver (blue-healer maigret / email-finder / SMTP). Yields ~0 on pseudonymous handles, burns API calls, creates PII exposure. The lean pull IS the product.
• No emails/phones — not findable, not the deliverable. The profile is the channel.
• Veterans only. Keep separate from the SSDI skill; don't merge the two.

Tiers (encode need)

• best — denied / appealing / under-rated / 0% / never filed / owed back pay
• better — filing / gathering evidence / C&P exam / secondary condition / intent to file
• good — eligibility / PACT Act / rating questions

Paths & deps

• Vertical: ~/Datamine/datamine_config/verticals/veterans-benefits-gap.yaml (edit phrases/subs here)
• Pull + book: self-contained in this skill's scripts/ (veterans_pull.py, build_book.py)
• Datamine bin: ~/Datamine/.venv/bin/datamine (auto-set as DATAMINE_BIN)
• Publisher: ~/.codex/skills/here-now/scripts/publish.sh (Brent's key already saved)

Codex Runtime

Resolve sibling Codex skills from installed skill roots (including ~/.codex/skills) or the current collection; do not assume a legacy skill root.

Never expose or print secret, credential, or token values.

Mandatory dependencies:

• Datamine environment
• last30days data-source credentials
• here.now publishing credentials

Preflight each dependency using MCP/app capability discovery, CLI availability/version checks, read-only filesystem or Git checks for repositories, and provider auth-status commands without printing secrets, credentials, or tokens. If any mandatory dependency is unavailable, stop and report a concise blocked state naming the missing dependency and the next action needed.