An evidence-first read of Target's public surface, starting with genuine credit (it ships the agent front door its peers don't), then the two things that complete it: the Target Forward commitments made machine-readable, and a per-store proof layer. Every finding is reproducible by anyone repeating the exact command.
Target has done the thing most retailers haven't even started. Before a single finding, the record:
llms.txt: target.com/llms.txt returns HTTP 200 (text/plain), a genuine agent map that documents the category (/c/…/-/N-…), browse, and product (/p/…/-/A-{TCIN}) URL grammar. Among the three institutions reviewed alongside Target, it is the only real agent front door.includeSubDomains, X-Frame-Options, a CSP with frame-ancestors 'self', nosniff, enforced email auth (DMARC p=reject).A retailer that already ships an llms.txt is the easiest first buyer we've found: the work isn't a rebuild, it's the next commit on a file the team already maintains.
No jargon: here is what each fix below is worth to Target in plain terms.
Today an AI can read Target's catalog but can't do or confirm anything against it; closing that gap keeps Target in the basket as assistants start transacting, not just browsing.
Structured, checkable commitments turn a graded-in-prose pledge into trust a rating body or a guest can confirm, instead of a claim they have to take on faith.
A five-line static file gives a security researcher somewhere to report to, cheap protection for a brand this size.
The live llms.txt is a shopping catalog map. It contains zero references to Target Forward, sustainability, or climate, and its own section header calls the owned-brand list "Phase 1", Target itself frames this file as a work in progress. Meanwhile there is no agent card (404), so an agent can read what to buy but has nothing to do or verify. Why it matters: Target's two public commitments, "regenerative" and "inclusive", are exactly the things a proof-verified work layer makes legible and provable: restore work (regenerative) done by team members climbing a step-1-to-mastery ladder (inclusive). The head start is real; the finish is one file and one proof layer away. The reproducible proof:
The llms.txt is genuinely useful for discovery, but the action/verification layer is absent: no /.well-known/agent-card.json (404), no ai.txt (404), no payment or MCP tool schema, no verifiable-outcome endpoint. Why it matters: the next wave of retail agents don't just read, they transact and they cite proof. An agent that can read Target's catalog still can't be handed a task it can complete and settle against Target. The proof:
The Target Forward strategy page (HTTP 200) carries a single ld+json (structured data machines can read) block whose only @type is WebPage, site chrome, not the goals. The 2025 report grades goals in prose (achieved / progressing / evolving), and some 2025 packaging goals are now graded "evolving." Why it matters: honest grading is a credibility asset, and it's exactly the case where a per-place proof receipt rebuilds trust faster than a re-worded goal. "Evolving" reads very differently next to a ledger of verified store restorations. The proof:
Target's hygiene is good: HSTS includeSubDomains, CSP, XFO, nosniff, DMARC p=reject. The one note: no /.well-known/security.txt (404), so there's no published channel for a security researcher, while peer walmart.com serves one (HTTP 200). A five-line static file. Not a vulnerability; a completeness nudge. The proof:
A restore opportunity at a specific store, a native planting strip, a community garden, the swale at the edge of the lot, becomes a bounded work packet: scope, safety, funding path, proof requirements. It's claimed by a team member on a step-1-to-mastery ladder (Build / Grow & Cook / Restore) or by a local, closed with photo + GPS + signature proof, scored by a deterministic proof-checker, and settled to a permanent receipt on Base with its tCO2e retired. Each receipt appends to that store's verified Target Forward record, the per-place proof a graded-in-prose goal is missing, and a legible asset an agent, a rating body, or a guest can read.
This maps onto Target Forward without a rebrand: "regenerative" is the restore work; "inclusive" is the step-1 ladder that lets any team member start today. The restore lanes are live on our board with reproducible counts (curl https://vealth.net/labor/stats): tree_planting 73, urban_canopy_survey 511, garden_bed_planting 64, plus clean_waterway, native_planting, priced for agents at work-packet $0.10 + work-tco2e $0.005 over x402. And the llms.txt extension is a static-file change Target's team already knows how to ship.
| Step | What happens | Who touches what |
|---|---|---|
| 1 | Extend the llms.txt Target already ships (Phase 2 draft in the Agent Kit): add Target Forward + a verified-outcomes section + an agent-card | Target web team, a file they already maintain |
| 2 | One store-community restore packet is authored + funded (community-giving or a single store's budget) | EcoWealth rail; no Target systems touched |
| 3 | The restore is proof-checked (photo+GPS+signature) and settled on Base with tCO2e; receipt appended to the store's Target Forward record | EcoWealth's live contract; Target reads receipts |
| 4 | Target Forward commitments expressed as structured, verifiable per-place data an agent or rating body can read | Target publishes; EcoWealth supplies the schema |
| Claim | Source / re-verify | Observed |
|---|---|---|
target.com/llms.txt = 200 (real agent map); 0 Target Forward/sustainability matches; header says "Phase 1" | curl -s https://www.target.com/llms.txt | grep -icE 'target forward|sustainab|climate' | 2026-07-14 |
| No agent-card (404), no ai.txt (404), no security.txt (404) | curl -s -o /dev/null -w '%{http_code}' https://www.target.com/.well-known/agent-card.json etc. | 2026-07-14 |
Target Forward strategy page ld+json = 1 (@type WebPage) | curl -s .../strategy-target-forward | grep -c application/ld+json | 2026-07-14 |
DMARC p=reject; HSTS includeSubDomains; peer walmart.com serves security.txt (200) | dig +short TXT _dmarc.target.com · curl -sI https://www.target.com/ | grep -i strict-transport | 2026-07-14 |
| Target Forward launched 2021 ("regenerative and inclusive"); 2025 report grades goals achieved/progressing/evolving; 76% renewable electricity 2024; ~400,000 team members | corporate.target.com + WebSearch (cited in evidence.json) | 2026-07-14 |
EWP live: 7,624 postings / $671,012; contract 0x76c17C…A14B; work-packet $0.10, work-tco2e $0.005 | curl -s https://vealth.net/labor/stats · BaseScan · /.well-known/x402 | 2026-07-14 |
work-packet call ($0.10) against our live endpoint and get back a real, bounded, place-based restore packet: a POST to vealth.net/.well-known/vealth/work-packet returns HTTP 402 with the standard schema; a runnable dry-run client is at vealth.net/recipes.html. No contract, no meeting.