Proven
docs/shopmine.md
Memory gate
Read before AutoOrder actions; it contains quantity, color, size, evidence, and source-site rules.
AutoOrder memory dashboard
Success and failure memory for real fulfillment decisions.
This screen connects the repository memory, ShopMine operating rules, variant normalization, and failure-prevention dimensions into one visible control surface.
Open service manualRepository source: docs/shopmine.md
Current contract
Evidence first, retry second.
Updated
2026-04-25T00:00:00+09:00
Dimensions
15 analysis axes
Learning records
4
Blocked records
1
Guarded
15D
Failure analysis
Every failed or blocked action must be analyzed across at least ten dimensions before retry.
Proven
3 codes
Variant map
BBB, WWW, and WGB are explicit bundle semantics, not free-form guesses.
Proven
7 codes
Size map
Topten, Eland Mall, and SPAO share the current numeric size normalization rule.
Variant and size proof
No option is guessed.
| Code | Meaning | Evidence rule |
|---|---|---|
| BBB | Black + Black + Black | Treat as a 3-piece black bundle only after order quantity and product option both agree. |
| WWW | White + White + White | Treat as a 3-piece white bundle only after order quantity and product option both agree. |
| WGB | White + Gray + Black | Treat as mixed bundle; verify every source-site swatch before checkout. |
Source size 80
XXS
Aliases: 2XS
Source size 85
XS
No alias required
Source size 90
S
No alias required
Source size 95
M
No alias required
Source size 100
L
No alias required
Source size 105
XL
No alias required
Source size 110
2XL
Aliases: XXL
Failure prevention
15-dimensional root cause grid.
1. Customer order intent
Confirm the buyer ordered one line, multiple lines, or a bundle.
Hook: Block fulfillment if quantity and option text are not both captured.
2. Quantity and bundle count
Separate order quantity from bundle composition such as BBB, WWW, or WGB.
Hook: Record quantity proof before clicking any source-site payment button.
3. Brand and source-site semantics
Topten, Eland Mall, and SPAO can expose the same size code with different option UI.
Hook: Use site-specific evidence notes for every product source.
4. Color swatch mismatch
Do not trust color text alone when swatches or images can disagree.
Hook: Capture or describe clicked swatch evidence for Eland Mall and SPAO unknowns.
5. Size normalization
Map 100 to L, 105 to XL, 110 to 2XL/XXL, 95 to M, 90 to S, 85 to XS, 80 to XXS/2XS.
Hook: Show source size and normalized size side by side in the fulfillment record.
6. UI selection persistence
Dropdowns, swatches, and cart pages can reset after navigation or reload.
Hook: Read back cart option text after every selection and before payment.
7. Stock, price, coupon, and points
A valid variant can still fail because stock, price, coupon, or points changed.
Hook: Record benefit maximization state before checkout, or mark the lane blocked.
8. Account, payment, and shipping session
Login prompts, address verification, or payment challenge can change the flow.
Hook: Treat unknown session state as blocked, never as completed.
9. Automation control limits
Live checkout and ShopMine edits require hardware/UIA routes, not browser devtools automation.
Hook: Use pyautogui, pywinauto, UIA, or approved APIs only for live payment/control actions.
10. Operator or model error
Failure can come from my own misread, stale memory, or overconfident inference.
Hook: Write a failure record with evidence and a concrete guard before retrying.
11. External website change
Source sites can change DOM, labels, images, shipping policy, or checkout order.
Hook: Snapshot new UI state and update the source-site note before continuing.
12. Evidence quality
Screenshots, readback text, and order detail URLs are stronger than memory or list views.
Hook: Never mark verified from a list view alone.
13. Timing, network, and cache
Slow loads or cached cart state can make a previous selection appear current.
Hook: Wait for fresh state and read back the final order detail.
14. Prior memory conflict
Old success rules may not apply when source-site conditions changed.
Hook: Compare current evidence against the latest success and failure memory.
15. Business downside and reversibility
Payment, dispatch, cancellation, and return completion have irreversible or costly downside.
Hook: Resolve hard constraints and irreversible downside before speed or convenience.
Learning records
Keep what worked. Block what failed.
A record is not success unless runtime evidence proves it. A blocked state is not hidden; it is visible until evidence changes.
AOM-001
ProvenRead the AutoOrder skill memory before fulfillment work
Local skills require docs/shopmine.md, success_log, failure_log, and CS memo rules before cycles.
The dashboard and docs make docs/shopmine.md the first operating read.
AOM-002
GuardedDo not store API keys in repository memory
User supplied DeepSeek and Gemini keys, but repository artifacts must not persist secrets.
External critic calls use local DEEPSEEK_API_KEY only, with no key text in docs, prompts, or commits.
AOM-003
Blockedpredok0905/autoorder is not currently accessible
GitHub web and GitHub contents API returned 404 for the requested repository.
Continue from local repository rules until the repo is made public or connector access is granted.
AOM-004
GuardedQuantity and variant are different proof fields
User explicitly warned that one order vs two orders and BBB/WWW/WGB bundle meaning must be checked.
Every order record must store ordered quantity, bundle code, source option, normalized color, and normalized size separately.