agentleFS
cook-mdcook-md/cook-mcp/skills/nutrition-goals/SKILL.md

nutrition-goals

Skill1 starsChanged 40 days agoMITView on GitHub

Use when the user wants to change AN EXISTING Cooklang recipe (.cook) or meal plan (.menu) so it hits numeric nutrition targets ("make this 600 kcal per serving", "get my week to 150 g protein a day", "more protein, same calories"). Measures with the Cook nutrition service, edits, and verifies. Not for evaluation-only questions (nutrition-reports), and not for picking or screening recipes that already meet a number (nutrition-reports screening, then meal-planning).

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 5 are distinctive, and 6 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 6

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use when the user wants to change a recipe or meal plan so it hits numeric nutrition targets — "make this 600 kcal per serving", "get my …

    line 10
  2. Distinctive:

    Verify the edited text BEFORE saving.

    line 38
  3. Very specific:

    Tolerance: default ±5 % (within_tol(value, target, 5)).

    line 53
  4. Very specific:

    Read the file with read_recipe (you need its source to edit it), then render it with render_report (input_path = the file).

    line 62
  5. Distinctive:

    Prefer the least invasive tier that can reach the target.

    line 90
  6. Very specific:

    Keep the dish's character: a few decisive changes beat many tiny ones.

    line 100
  7. Very specific:

    Write the full edited file text, then render it with render_report using input (the edited text) and kind ("cook" or "menu") instead of …

    line 107
  8. Very specific:

    Run validate on the edited text, then save with write_recipe (.cook) or write_menu (.menu), always the full file content.

    line 140
  9. Distinctive:

    Every number comes from the service: Never estimate nutrition from your own knowledge — not for baselines, not for candidate verification.

    line 200
  10. Distinctive:

    Keep proposed quantities practical (no "137 g chicken".

    line 212
  11. Very specific:

    "authentication required": call login, show the user the code and link, retry once they've approved.

    line 218

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Skill: Nutrition Goals

Needs the Cook MCP server (tools like render_report and validate). If these tools are missing, the Cook MCP server isn't connected: if you installed the cooklang plugin or extension, check that its cook server is running (e.g. /mcp) or reinstall it; otherwise add it from https://github.com/cook-md/cook-mcp. Never estimate nutrition from your own knowledge instead.

Use when the user wants to change a recipe or meal plan so it hits numeric nutrition targets — "make this 600 kcal per serving", "get my week to 150 g protein per day", "more protein, same calories". You measure with the Cook nutrition service (through render_report), choose edits, verify them before saving, save, and confirm.

Nutrition data needs a cook.md login and Cook Basic or Cook Pro. On "authentication required" call login; on "subscription required" show the user the pricing link from the error. The direct nutrition tools (get_nutrition, aggregate_nutrition, lookup_ingredient, …) return JSON login_required (call login) or plan_required (show the user checkout_url) instead.

This skill changes food to reach numbers. For evaluation-only questions where no change is wanted ("is this healthy?", "how much protein is in this?"), follow the nutrition-reports skill instead.

Not for selection. "Find me three dinners over 35% protein", "which of my recipes are high protein", "plan a week that averages 150 g protein from what I already have" are screening requests: nothing gets edited, and the number is a filter over existing files. Running this skill's per-file measure-and-edit loop over a candidate list is how a three-dinner question turns into a hundred tool calls. Follow nutrition-reports (its "Screening a shortlist" section) and, if a .menu is wanted at the end, meal-planning. Come back here only once a specific file has been chosen and has to change to hit the number.

The loop

  1. Pin down the target (nutrient, value, basis, tolerance).
  2. Measure the baseline and per-ingredient contributions.
  3. Choose edits — tiered, minimally invasive.
  4. Verify the edited text BEFORE saving.
  5. Save the edit and present current → new vs target.
  6. Re-measure the saved file to confirm.
  7. Offer to save the targets as a reusable checks template.

1. Pin down the target

  • Nutrient keys: macro keys are kcal, protein_g, fat_g, sat_fat_g, carb_g, sugar_g, fiber_g; micros/vitamins by key via nutrient_total (e.g. "sodium_mg", "iron_mg"). nutrient_total returns 0 for an unknown key — if a micro baseline reads 0 for an ingredient list that plainly contains it, suspect the key, not the food.
  • Basis: per serving, whole recipe, or per day (for .menu plans). If the recipe declares servings and the user just says "600 kcal", per-serving vs total is genuinely ambiguous — ask. Otherwise don't ask; state your reading.
  • Tolerance: default ±5 % (within_tol(value, target, 5)); use the user's figure when given. For "keep X the same", hold X within ±5 % of its current value.
  • Already given: targets, restrictions or servings the user stated earlier in the conversation (including in a prompt they ran) count as provided. Do not ask for them again — restate your reading and proceed.

2. Measure the baseline

Read the file with read_recipe (you need its source to edit it), then render it with render_report (input_path = the file). Emit machine-readable lines — the ITEM lines are your lever map (which ingredients dominate which nutrient):

{% set t = macros(ingredients) %}
{% set servings = metadata.servings | default(1) | int %}
TOTAL kcal={{ t.kcal | round(1) }} protein={{ t.protein_g | round(1) }} fat={{ t.fat_g | round(1) }} carb={{ t.carb_g | round(1) }} fiber={{ t.fiber_g | round(1) }}
PER_SERVING ({{ servings }}) kcal={{ (t.kcal / servings) | round(1) }} protein={{ (t.protein_g / servings) | round(1) }}
{% set agg = aggregate_nutrition(ingredients) %}
{% for it in agg.items -%}
ITEM {{ it.ingredient }} | {{ it.amount.value }} {{ it.amount.unit }} | kcal={{ it.macros.kcal | round(1) }} protein={{ it.macros.protein_g | round(1) }} | {{ it.confidence }}
{% endfor -%}
{% for f in agg.failures -%}
FAILED {{ f.ingredient }} — {{ f.error.message }}
{% endfor -%}

FAILED ingredients contribute nothing to the totals — tell the user the numbers exclude them rather than silently trusting the total. Flag estimated confidence items when they dominate a target nutrient.

For .menu plans the template reads plan.days (each day's scaled .ingredients) instead of ingredients; see the "Meal plans" section of nutrition-reports and https://nutrition.cook.md/docs/guides/plan.

3. Choose edits — tiered

Prefer the least invasive tier that can reach the target; move down only when the current tier cannot:

  1. Quantities & servings — scale ingredient amounts, or plan servings.
  2. Swaps that preserve the dish — yogurt → Greek yogurt, half the oil plus broth, white rice → lentils. lookup_ingredient finds catalog names, check_category keeps a swap in its category (e.g. oily fish), and get_nutrition compares per-amount profiles before you commit.
  3. Add/remove ingredients — or recipes in a plan — last.

Keep the dish's character: a few decisive changes beat many tiny ones. Round to practical cooking amounts and keep the recipe's existing unit style. The cooklang://syntax resource covers the edit mechanics for .cook and .menu files.

4. Verify BEFORE saving

Write the full edited file text, then render it with render_report using input (the edited text) and kind ("cook" or "menu") instead of input_path — the same measuring template, so numbers are comparable. Nothing is saved yet. Refine at most ~3 rounds; if the targets stay unreachable within the allowed tiers, stop and present the closest achievable numbers with a plain explanation of the gap (e.g. "without swapping the rice, 40 g protein forces ~750 kcal").

To compare a few alternatives quickly without writing each one out, sum nutrition_for_amount over a candidate list (item shape — macros live under .macros); it raises a template error when an ingredient can't be resolved, so fix the name/unit and re-render:

{% set candidate = [
  {"name": "chicken breast", "amount": 300, "unit": "g"},
  {"name": "olive oil", "amount": 1, "unit": "tbsp"},
  {"name": "basmati rice", "amount": 150, "unit": "g"}
] %}
{% set servings = 2 %}
{% set ns = namespace(kcal=0.0, protein=0.0) %}
{% for c in candidate %}
{% set n = nutrition_for_amount(c.name, c.amount, c.unit) %}
{% set ns.kcal = ns.kcal + n.macros.kcal %}
{% set ns.protein = ns.protein + n.macros.protein_g %}
{% endfor %}
CANDIDATE total kcal={{ ns.kcal | round(1) }} protein={{ ns.protein | round(1) }}
CANDIDATE per-serving kcal={{ (ns.kcal / servings) | round(1) }}
KCAL_OK={{ within_tol(ns.kcal / servings, 600, 5) }}

5. Save the edit

  • Run validate on the edited text, then save with write_recipe (.cook) or write_menu (.menu), always the full file content. This overwrites the user's file: if they asked to see options first, or the change is large (tier 3), confirm before saving.
  • Tell the user what you changed in one or two lines per ingredient (olive oil 3 tbsp → 1 tbsp), not a full diff, and where you saved it.
  • Present current → new vs target (per serving/day and total, from steps 2 and 4) as a small table.

6. Confirm on the saved file

Re-render the saved file with input_path and the same measuring template. If the numbers differ from step 4, say so and fix — the saved file is what the user will cook from.

7. Offer to persist the goals

Offer (don't push) to save the targets as a reusable checks template so future edits can be re-validated (the nutrition-reports flow), at reports/goals.<recipe-slug>.md.jinja (or under config/reports/ if that is where the collection keeps its templates — Cook Editor does), with write_config:

{% import "ck" as ck %}
{% set t = macros(ingredients) %}
{% set servings = metadata.servings | default(1) | int %}
# Goals — {{ metadata.title | default("Recipe") }}

- Energy: {{ (t.kcal / servings) | round }} kcal per serving
- Protein: {{ (t.protein_g / servings) | round(1) }} g per serving

{{ ck.between(t.kcal / servings, 570, 630, "kcal per serving") }}
{{ ck.min(t.protein_g / servings, {"min": 40}, "Protein per serving (g)") }}

{% set checks = all_checks() %}
**{{ checks | selectattr("ok") | list | length }} of {{ checks | length }} checks passed.**

Render it later with template_path: "reports/goals.<recipe-slug>.md.jinja".

(ck.min/ck.max/ck.range take a target object like {"min": 40}; ck.between takes scalar bounds.)

Function cheatsheet (condensed)

Full reference at https://nutrition.cook.md/docs/guides/functions.

FunctionReturns
macros(ingredients)flattened totals {kcal, protein_g, fat_g, sat_fat_g, carb_g, sugar_g, fiber_g}
aggregate_nutrition(ingredients){items, failures, totals, confidence_breakdown}; each item has .ingredient, .amount.value/.unit/.mass_g, .macros.*, .confidence
nutrition_for_amount(name, amount, unit, prep?, standard?)one full item (same shape as items[]) for an arbitrary hypothetical amount
nutrient_total(ingredients, key)number; micros then vitamins ("sodium_mg", "vit_d_iu")
within_tol(actual, target, tol_pct)bool — within tol_pct % of target
compare(actual, target, op)bool — op is gte/lte/eq
convert(amount, from, to, ingredient?)unit conversion (ingredient needed for volume/count)
record_check(label, ok) / all_checks()record and read passed/failed checks (used by ck macros)

Guardrails

  • Every number comes from the service. Never estimate nutrition from your own knowledge — not for baselines, not for candidate verification.
  • Surface FAILED and low-confidence ingredients; never present totals that silently exclude them. Measure with aggregate_nutrition / macros (both skip unresolved ingredients; aggregate_nutrition lists them under failures), never with nutrition_for(ingredient) in a loop — that aborts on the first miss.
  • Non-English recipes: the nutrition database resolves English ingredient names only. When the FAILED list is mostly non-English names, tell the user once — "Nutrition lookups currently understand English ingredient names — <language> isn't supported yet; email support@cook.md if you'd like it added sooner" — and offer to proceed with English equivalents via nutrition_for_amount, clearly marked as translated by you.
  • Keep proposed quantities practical (no "137 g chicken"; prefer 140 g).

When the render errors

render_report returns an error result on failure:

  • "authentication required": call login, show the user the code and link, retry once they've approved.
  • "subscription required": nutrition math needs Cook Basic or Cook Pro — show the pricing link from the error. You may still make a qualitative edit if the user wants, clearly labelled as unverified. Do not keep retrying.
  • Template error (minijinja reports the line): fix the template, render again.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.