Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Persistent dietary safety layer for DoorDash CLI (dd-cli) ordering. Stores a personal/household dietary profile (allergens with severity tiers, diets, dislikes) that every cart is vetted against before checkout, with a deterministic tripwire — the vetting step saves the full cart contents to a vetted-cart dump, and a PreToolUse hook re-greps that dump against the anaphylaxis-severity allergen list before any checkout URL is allowed. Use when the user mentions allergies or dietary restrictions, w
.claude/skills/davila7-doordash-allergy-shield/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 239% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 41% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 72% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 123% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 24% | 0% |
Claude has no cross-session memory: tell it "I'm allergic to peanuts" today and tomorrow it happily adds pad thai. This skill persists a dietary profile and makes cart vetting mandatory — with a deterministic hook double-checking the highest-severity allergens before any checkout URL is emitted.
> Unofficial community skill built on DoorDash's doordash-oss/doordash-cli.
cart show output plus a synonyms table. It cannot see full ingredient lists and knows nothing about cross-contamination in the kitchen.
DoorDash payment page — where the human reviews the real order — is the documented final check.
cheap; the reverse is not. Hook matching is substring-based (short allergen names like "oat" can match inside "goat cheese") — annoying, never dangerous.
references/allergen-synonyms.md table is intentionally broader thanthe compact synonym subset hardcoded in the checkout-gate hooks; adding a synonym to the reference doc alone does not change what the hooks enforce.
| Piece | Role | |---|---| | this skill | vetting protocol + profile management | | references/allergen-synonyms.md | hidden-allergen lookup table | | hook doordash/doordash-allergy-checkout-gate | deterministic re-check at checkout | | command /doordash-profile | manage the dietary profile interactively |
~/.claude/doordash-profile/dietary.json — the profile. Structure:json{ "people": { "me": { "allergens": [ { "name": "peanut", "severity": "anaphylaxis" }, { "name": "shellfish", "severity": "avoid" } ], "diets": ["vegetarian"], "dislikes": ["cilantro"] }, "sam": { "allergens": [{ "name": "egg", "severity": "avoid" }], "diets": [], "dislikes": [] } } }
Severity tiers: anaphylaxis (hook-enforced, never overridable in-session), avoid (requires explicit human acknowledgment to keep the item), preference (mention it, don't gate on it).
~/.claude/doordash-profile/vetted/<cart-uuid>.json — one dump per vetted cart:the full raw cart show output plus your verdict and a timestamp. The checkout hook greps this artifact — it trusts the dump, not your claim.
Read dietary.json. Missing → offer to create it interactively (or via /doordash-profile). If the user is ordering for others ("lunch for me and Sam"), make sure each eater exists in the profile or ask for their restrictions.
dd-cli cart show --cart-uuid <X> and capture the full output.AND hidden sources from references/allergen-synonyms.md (satay→peanut, aioli→egg, ponzu→soy+fish, …).
anaphylaxis → remove the item (dd-cli cart remove-item) and tellthe user why. Do not offer to keep it.
avoid → present the conflict; keep only with explicit acknowledgment.preference → mention it in passing.UNVERIFIED and require the user's explicit sign-off for eaters with anaphylaxis entries.
cart show output + verdict + ISO timestampto ~/.claude/doordash-profile/vetted/<cart-uuid>.json. Without this file the checkout hook blocks — vetting is not optional.
Just run dd-cli order checkout-url --cart-uuid <X> (or the dd-guard wrapper when doordash-spend-guard is installed — the two gates compose). The hook independently verifies: dump exists, is newer than the last cart mutation it saw, and contains no anaphylaxis-tier keyword. If it blocks, do NOT try to work around it — fix the cart, re-vet, and explain to the user what tripped.
dietary.json from the shell; profile changes go through/doordash-profile (interactive, confirmed).
cart show fresh — thedump IS the audit artifact.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 15,186 | 5,733 | -62% | 1 | 1 | 0% | 1,640 | 1,533 | -7% | 0 | 0 | — |
case-02 | fail→fail | 8,279 | 6,424 | -22% | 1 | 1 | 0% | 629 | 1,609 | +156% | 0 | 0 | — |
case-03 | fail→fail | 10,956 | 6,041 | -45% | 1 | 1 | 0% | 1,111 | 1,605 | +44% | 0 | 0 | — |
case-04 | fail→pass | 4,538 | 10,954 | +141% | 1 | 1 | 0% | 766 | 2,596 | +239% | 0 | 0 | — |
case-05 | fail→pass | 9,707 | 7,568 | -22% | 1 | 1 | 0% | 1,709 | 2,413 | +41% | 0 | 0 | — |
case-06 | fail→fail | 5,195 | 6,907 | +33% | 1 | 1 | 0% | 28 | 1,566 | +5493% | 0 | 0 | — |
case-07 | fail→pass | 6,600 | 3,500 | -47% | 1 | 1 | 0% | 1,021 | 1,758 | +72% | 0 | 0 | — |
case-08 | fail→pass | 5,668 | 5,758 | +2% | 1 | 1 | 0% | 1,004 | 2,239 | +123% | 0 | 0 | — |
case-09 | fail→pass | 6,298 | 3,143 | -50% | 1 | 1 | 0% | 1,360 | 1,689 | +24% | 0 | 0 | — |
case-10 | fail→pass | 6,883 | 4,058 | -41% | 1 | 1 | 0% | 1,387 | 1,923 | +39% | 0 | 0 | — |
case-11 | pass→pass | 6,414 | 3,884 | -39% | 1 | 1 | 0% | 1,072 | 1,834 | +71% | 0 | 0 | — |
case-12 | pass→pass | 9,297 | 3,194 | -66% | 1 | 1 | 0% | 1,572 | 1,791 | +14% | 0 | 0 | — |
case-13 | pass→pass | 4,815 | 5,646 | +17% | 1 | 1 | 0% | 791 | 2,203 | +179% | 0 | 0 | — |
case-14 | pass→fail | 8,171 | 3,922 | -52% | 1 | 1 | 0% | 1,175 | 1,767 | +50% | 0 | 0 | — |
case-23 | fail→pass | 7,143 | 2,790 | -61% | 1 | 1 | 0% | 1,370 | 1,586 | +16% | 0 | 0 | — |
case-15 | fail→pass | 9,317 | 9,054 | -3% | 1 | 1 | 0% | 735 | 1,998 | +172% | 0 | 0 | — |
case-16 | fail→pass | 9,921 | 3,400 | -66% | 1 | 1 | 0% | 1,698 | 1,807 | +6% | 0 | 0 | — |
case-17 | fail→pass | 9,442 | 5,178 | -45% | 1 | 1 | 0% | 1,532 | 2,006 | +31% | 0 | 0 | — |
case-18 | fail→pass | 8,287 | 1,743 | -79% | 1 | 1 | 0% | 1,171 | 1,391 | +19% | 0 | 0 | — |
case-19 | fail→pass | 10,171 | 2,667 | -74% | 1 | 1 | 0% | 1,591 | 1,547 | -3% | 0 | 0 | — |
case-20 | fail→pass | 7,790 | 4,083 | -48% | 1 | 1 | 0% | 1,130 | 1,888 | +67% | 0 | 0 | — |
case-21 | fail→pass | 5,609 | 4,543 | -19% | 1 | 1 | 0% | 929 | 1,983 | +113% | 0 | 0 | — |
case-22 | fail→pass | 8,033 | 3,267 | -59% | 1 | 1 | 0% | 1,254 | 1,683 | +34% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 19 counted toward the lift figure. The other 4 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +61 percentage points is the difference between those two pass rates over the 19 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.