Salesforce Data + Tooling API Skill

This skill provides field-level reference for 2130 standard Salesforce objects across two runtime API surfaces: the Enterprise/Data API (standard sObjects you query with SOQL and modify with DML) and the Tooling API (developer/metadata-adjacent records like ApexClass, ApexCodeCoverage, TraceFlag, and EntityDefinition).

Use it to look up authoritative field names, types, and properties before writing SOQL/SOSL, building DML, or reading records at runtime — so queries and writes don't fail with INVALID_FIELD / "No such column" errors.

Standard objects only. These assets cover standard sObjects and
Tooling records. Custom c objects, and custom c fields on standard
objects, are not in this corpus — they don't exist as static docs pages.
For those, describe the live org instead (sf sobject describe --sobject <Name>).

Overview

Each object is documented as a JSON file with:

Field definitions: name, type, and properties (Createable, Filterable, Groupable, Nillable, Sortable, Updateable)
Relationship metadata (relationship name, referenced object, relationship type) for enterprise sObjects
Supported SOAP calls and REST HTTP methods for Tooling records
WSDL schema segment
Usage notes and associated objects (enterprise sObjects)

This skill is for runtime data, not deployment. For authoring *-meta.xml
source files (CustomObject, Flow, Profile, ...) use the Metadata API skill
(platform-metadata-api-context-get). The two are companions, not substitutes.

How to Use This Skill

CRITICAL: Field-Existence Gate (do this BEFORE answering)

NEVER answer a field question — "does field X exist?", "is X filterable/sortable/groupable?", "what's X's API name/type?" — from memory or training data. This includes obvious system fields like Id, Name, CreatedDate, LastModifiedDate, OwnerId, IsDeleted. ALWAYS run the lookup first. You know these fields from training, but this skill exists precisely because your recollection of their properties (Filter/Sort/Group) and of which catalog documents them is unreliable. Confidence is not verification.

Run ONE atomic command that checks BOTH catalogs at once (a field can live in fields, in field_reference, or neither — see the dual-catalog note below):

jq '{fields: .fields["<FieldName>"], field_reference: .field_reference["<FieldName>"]}' assets/enterprise_api/<Object>.json
# e.g. checking CreatedDate on Account:
jq '{fields: .fields["CreatedDate"], field_reference: .field_reference["CreatedDate"]}' assets/enterprise_api/Account.json

Interpret the result: found in fields → use its properties for capability answers. Found only in field_reference → it exists, but Filter/Sort/Group cannot be determined from this skill (that catalog carries no property flags). null in both → not in this corpus (may still be a real live-org/custom field — describe the org).

MANDATORY VERIFICATION GATE. Before you state any field fact, you MUST first print this line verbatim in your reply to the user (like the metadata skill's status line):

field_lookup: object=<Object> field=<FieldName> checked_fields=<yes|no> checked_field_reference=<yes|no> result=<in_fields|in_field_reference|not_found>

If you cannot print this line truthfully with both checks = yes, you have not done the lookup — stop and run the jq command. Do not fabricate a citation like "based on the skill's data" without having run it.

This gate line is a self-verification marker for your conversational response only — do NOT write it into files you generate for the user (.soql, .md, .json, query comment headers, etc.). Deliverables should contain just the answer/query the user asked for; keep the field_lookup: line out of them.

CRITICAL: Section-Specific Consumption

ALWAYS consume only the specific sections you need from JSON files, NOT entire files.

**For assets/enterpriseapi/.json and assets/toolingapi/.json files, always use jq or programmatic JSON parsing to extract only the sections you need.** Do not load these files whole via Read, cat, or readfile — they contain verbose wsdlsegment and field_reference sections that waste 60-80% of tokens on large sObjects like Account.

Each JSON file contains multiple sections. Most use cases only need 1-2:

For query/DML field lists: load only the fields section
For relationship traversal: load fields (the relationshipname / refersto columns are inline)
For Tooling call support: load supportedsoapcalls / supportedrestapihttpmethods
For "can this user query it?" / permission questions: load specialaccessrules (present on both surfaces when the object documents access gates)
For query ceilings / unsupported clauses (record caps, no ORDER BY/queryMore()/OFFSET): load limitations — present only on objects that impose them
Skip by default: wsdlsegment, ispersonaccountfields

Dual-catalog rule (imperative — this is the one people get wrong):

A field can be in fields, in field_reference, in both, or in neither. These are two DIFFERENT catalogs, not a superset relationship.
A miss in fields is NOT a "field doesn't exist" answer. Check field_reference too before concluding anything — the atomic jq command above does both in one shot.
Filter/Sort/Group/Create/Update/Nillable properties come only from fields. A field found only in field_reference has NO retrievable capability flags here — say so; don't guess.

<details> <summary><b>Why the two catalogs differ (reference — optional reading)</b></summary>

fields comes from Salesforce's SOAP "Fields" table (query/DML properties: Create/Filter/Sort/Group/Update/Nillable). fieldreference comes from a different UI-facing "Field List" table (label/length/precision/scale) that some objects document separately from the SOAP table — and it can include many fields fields never lists at all (e.g. objects with very large boolean-flag catalogs, or assorted admin-facing fields the SOAP table omits). Across the corpus, thousands of fields exist ONLY in fieldreference, and a smaller number exist only in fields. This is why a field-existence answer requires checking both, and why capability questions can only be answered from fields. </details>

Which folder?

Standard sObjects you query and modify at runtime (Account, Contact, Opportunity, Lead, Case, ...) → assets/enterprise_api/
Developer / diagnostics records (ApexClass, ApexCodeCoverageAggregate, TraceFlag, EntityDefinition, MetadataComponentDependency, SymbolTable) → assets/tooling_api/
Custom c objects and custom c fields → NOT in these assets. Describe the live org: sf sobject describe --sobject <Name> --target-org <alias>. (A custom field like Region__c on standard Account won't be in assets/enterprise_api/Account.json either.)

Example Queries (Section-Specific)

Do:

"Show me only the 'fields' section from assets/enterprise_api/Account.json"
"What are the filterable fields on Opportunity?"
"Which fields on Contact are relationships, and what do they refer to?"
"What SOAP calls does ApexClass support in the Tooling API?"
"Field X isn't in fields — check field_reference before saying it doesn't exist"

Don't:

"Load Account.json" (pulls the huge wsdlsegment; load fields and fieldreference separately instead of the whole file)

JSON File Structure

Enterprise (data) sObjects live in assets/enterpriseapi/; Tooling records live in assets/toolingapi/. Both share a common core, with a few surface-specific sections.

{
  "sections": ["title", "description", "fields", "wsdl_segment", ...],
  "title": "Account | Enterprise API",
  "description": "Plain-text description of the object.",
  "fields_columns": ["type", "properties", "description", "relationship_name", "refers_to", "relationship_type"],
  "fields": {
    "fieldName": {
      "type": "string | reference | picklist | ...",
      "properties": "Create, Filter, Group, Nillable, Sort, Update",
      "description": "Field description",
      "relationship_name": "(reference fields only)",
      "refers_to": "(reference fields only) e.g. Account"
    }
  },
  "wsdl_segment": "<xsd:complexType>...</xsd:complexType>"
}

fields is a dict keyed by field API name (e.g. fields["AnnualRevenue"]), not a list — iterate with .items()/.keys(), don't index it positionally.

How sections are derived (read this before assuming a section list)

Each JSON file is generated from whatever ## Heading sections actually appear on that object's Salesforce docs page — the JSON schema is not a fixed, uniform template applied identically to every object. Two consequences follow directly from that:

No section is guaranteed present on every object, fields included. A section

exists in a file only if the source docs page had a matching heading. Some Tooling docs pages have no fields table at all (thin/beta pages), so fields can be missing; a small number of SOAP-header-style pages label their field table singular (field) instead of plural. Always check the file's own sections array (or "fields" in sections) before assuming a section exists — don't hard code an expectation from having looked at one or two example objects.

A page can carry extra sections beyond the common ones, one per additional

## heading Salesforce's docs used on that specific page (e.g. a nested complex type referenced by a field, like a picklist's value-metadata description, or a record-type-info block). These are normalized into snakecase keys and stored either as their own top-level section, or — when the converter recognizes the heading as a distinct sub-type rather than a plain content section — nested under a subtypes dict keyed by the sub-type's name. Treat the "Available Sections" list below as the common/frequent case, not an exhaustive enum — the authoritative list for any one object is that object's own sections array.

Available Sections (common case — not exhaustive; see above)

title, description, fields_columns: present on essentially all objects
fields: present on the large majority of objects, but see point 1 above — verify

via sections rather than assuming

Enterprise-only: usage, associatedobjects, ispersonaccountfields, fieldreference, fieldreference_columns
Tooling-only: supportedsoapcalls, supportedrestapihttpmethods
specialaccessrules: **who can query/access the object and any permission or

license gate** (e.g. "Customer Portal users can't access this object", "requires Omnistudio licenses"). Present on both surfaces. Read this before concluding a query will run for a given user — it's a query-eligibility fact no field flag captures.

limitations: query ceilings and unsupported SOQL clauses the object imposes

(e.g. MetadataComponentDependency: max 2000 records via Tooling API / 100,000 via Bulk API 2.0; ORDER BY, OFFSET, queryMore(), and *Name-field filters are not supported). Read this before writing a query against an object that has it — these constraints are not derivable from the fields properties.

wsdl_segment: schema definition (verbose — skip unless you need the raw type)
subtypes and any other snakecase-named section not listed above: object-specific

nested schema description, present only when that object's docs page had a matching heading — inspect the file's sections array to discover them per-object

The fieldscolumns array tells you which columns each object's fields entries carry — check it per-object rather than assuming a fixed shape. Enterprise sObjects typically add relationshipname, refersto, and relationshiptype; Tooling records typically carry only type, properties, description — but a meaningful minority of Tooling objects also carry relationshiptype, so don't treat it as enterprise-exclusive. relationshiptype values also have spelling variants in the source data (e.g. "Lookup" vs. "Look up") — compare loosely (e.g. .replace(" ", "").lower()) rather than exact string equality.

Type strings are case-inconsistent across objects — do not assume a single casing. The same concept appears under multiple castings depending on the source doc page: string/String, boolean/Boolean, dateTime/DateTime/datetime. Compare type values case-insensitively (e.g. .lower()) rather than with == against one casing. Beyond the common string | reference | picklist types, real data also includes int, boolean, double, currency, date, dateTime, textarea, address, url, phone, email, time, anyType, base64, and others — treat the type list as open-ended.

A field can lack structured type/properties even when the object otherwise has them — check description as a fallback before concluding the data is missing. Because each object's JSON is generated from that object's own docs page, a page that presents a field's type/properties as prose ("Type: ... Properties: ... Description: ...") inside a single description paragraph — rather than as separate table columns — produces a fields entry with type/properties empty and that prose stuffed into description. This shows up for a range of reasons (generic template pages reused across many standard objects, e.g. history/share/feed-tracking companion objects) and can affect anywhere from one field to every field on an object — in the worst case fieldscolumns itself may say only ["description"]. If type/properties is empty or fieldscolumns looks suspiciously thin, check whether description starts with "Type: " and parse it as a fallback before concluding the field/object has no type information.

Picklist allowed values sometimes appear in description prose — check before assuming they're absent. No fields entry has a structured picklistValues list. Whether a picklist's allowed values are enumerated in description (as "Possible values are: ..." followed by the values) depends entirely on whether the source docs page enumerated them for that specific field — many do, many don't (e.g. org-configurable picklists backed by a separate value-set object typically only give illustrative examples like "such as New, Closed, or Escalated" instead of a full list). When present, the format itself is inconsistent — sometimes bare space/comma-separated tokens, sometimes Value—explanation em-dash pairs per line — so there's no single reliable regex; extract the substring after "Possible values are:" and parse case-by-case, and don't assume every picklist has a discoverable value set here.

A field's description may reference a relationship (e.g. "this is a dependent picklist", "this field controls X") without naming or resolving the other side of that relationship — the enterprise sObject's own fields section is not always where that answer lives. The Tooling API side of this dataset documents Salesforce's field-metadata objects (schema-about-schema, e.g. an object exposing per-field describe-style attributes) which can carry structured attributes — like a controlling field's identity for a dependent picklist — that the enterprise sObject's docs page never named explicitly. If an enterprise field's description points at a relationship it doesn't resolve, check whether a Tooling-side field-metadata object answers it before concluding the data doesn't exist anywhere in this skill.

A given object's JSON can carry sections beyond the common list above — inspect that object's own sections array rather than assuming a closed set. Any additional ## heading on the source docs page becomes its own snakecase-named section (or, when the converter recognizes it as a distinct nested type, a subtypes entry). These appear only on the specific objects whose docs happened to include that heading — a per-object schema-description payload, not junk to ignore. There's no fixed enumeration of every possible extra section name; discover them per-object via sections.

See the [Index Table](references/dataandtoolingindextable.md) for the full object listing and per-object extra sections — it's ~125 KB, so grep -i "<ObjectName>" it for a single lookup rather than reading the whole file.

More detail: worked query/DML examples, relationship-traversal patterns, and a full section glossary live in [references/usageguide.md](references/usageguide.md). Load it with the Read tool only when needed.

File Location

Object JSON files are split by API surface:

assets/
├── enterprise_api/      # standard sObjects (SOQL/DML)
│   ├── Account.json
│   ├── Contact.json
│   └── ...
└── tooling_api/         # developer / diagnostics records
    ├── ApexClass.json
    ├── TraceFlag.json
    └── ...

Files are referenced relative to the skill root, e.g. assets/enterpriseapi/Account.json or assets/toolingapi/ApexClass.json.

Working Examples

Runnable section-loading examples (each demonstrates: fields extraction, filterable-field lookup via properties, relationship traversal, the fields vs fieldreference dual catalog, and reading Tooling supportedrestapihttp_methods):

Python: [examples/pythonsectionloading.py](examples/pythonsectionloading.py) — json.load() with section extraction
JavaScript/Node.js: [examples/javascriptsectionloading.js](examples/javascriptsectionloading.js) — JSON.parse() with section extraction
Bash + jq: [examples/bashsectionloading.sh](examples/bashsectionloading.sh) — jq command-line JSON processing

See [examples/README.md](examples/README.md) for usage and the property→SOQL/DML capability table.

Query & DML Generation Requirements

When generating SOQL/SOSL or DML against these objects, follow these rules so the statement actually runs.

Verify the field exists and its API name

**Look up the field first (both catalogs, atomic jq) and print the

field_lookup: gate line — even for system fields you're sure about.** See the Field-Existence Gate at the top of "How to Use This Skill". Field API names are case-insensitive to Salesforce but must resolve to a real field. Custom fields end in c; custom relationships traverse with r.

Do not invent fields. A guessed column produces

INVALID_FIELD: No such column 'X' on entity 'Y'. If unsure, load the fields section and confirm.

Don't case-fold fields keys when matching. At least one object

(LoginEventLog) has both UserName and Username as distinct real fields in the same fields dict — a lowercase/case-insensitive lookup would silently collide them. Match the exact key casing as given. These are two different columns holding potentially different data — they are NOT the same field that "resolves to whichever the platform picks." Select and filter each by its exact casing; do not describe them as interchangeable or say a query "resolves ambiguously," because the platform treats them as distinct fields.

Respect field properties

The properties string on each field controls what you can do with it:

Filter → usable in a WHERE clause. A field without Filter cannot be filtered.
Sort → usable in ORDER BY.
Group → usable in GROUP BY.
Createable / Updateable → settable via insert / update DML. System and

formula fields are read-only (no Create/Update) — writing them fails.

Nillable → may be null; a non-nillable field is required on insert.

This skill's properties string has no External ID flag — upsert requires a field marked External ID in the org, and that flag isn't captured by this data source (it only appears in Metadata API describe output, not the HTML docs this skill is built from). To find or set a field's External ID flag, use the Metadata API skill (platform-metadata-api-context-get) and check CustomField's externalId attribute — don't guess an upsert field from this skill's properties alone.

Relationships (enterprise sObjects)

For reference-type fields, the relationshipname column gives the parent relationship for SOQL traversal and refersto names the target object(s):

-- OwnerId (reference, relationship_name = Owner, refers_to = User)
SELECT Id, Owner.Name FROM Account
-- Custom lookup Foo__c traverses as Foo__r
SELECT Id, Foo__r.Name FROM My_Object__c

Tooling API vs Data API

Tooling records (assets/toolingapi/) are queried through the Tooling API endpoint (/services/data/vXX.0/tooling/query), not the regular Data API. Check supportedsoapcalls / supportedrestapihttp_methods for what each record supports. Many Tooling objects are read-only.

Not for deployment. To author or edit *-meta.xml source (CustomObject,
Flow, Profile, ...) use the Metadata API skill — the objects here are the
runtime/queryable representation, not the deployable metadata form.

Duplicate and Ambiguous Object Names

Several names exist in BOTH the Metadata API and here (ApexClass, ApexTrigger, CustomField, CustomObject, EmailTemplate, Layout, Profile, PermissionSet, RecordType, ValidationRule, Flow, ...). They mean different things:

This skill = the runtime/queryable record: fields you SELECT, filter, and

(sometimes) write via the Data or Tooling API.

Metadata API skill = the *-meta.xml source form you author and deploy.

Resolve ambiguity with these signals:

"query", "SOQL", "SOSL", "DML", "insert/update/upsert", "what fields/columns",

"filterable", "record", "runtime", "REST/SOAP" → this skill.

"authoring", "deploy", "retrieve", package.xml, force-app/, sfdx,

.meta.xml, "blueprint/template" → Metadata API skill.

Tooling-specific: "Tooling API", ApexCodeCoverage, EntityDefinition,

TraceFlag, SymbolTable, "code coverage", "compile errors", "debug log" → Tooling half of this skill (assets/tooling_api/).

If invoked directly by name with no other signal, default to the runtime data interpretation and disclose the assumption.

Troubleshooting

File Not Found

File names are case-sensitive PascalCase matching the object API name

(Account.json, ApexClass.json), no separators.

Check the correct folder: enterprise sObjects in assets/enterprise_api/,

Tooling records in assets/tooling_api/. A name can exist in only one, or in both with different fields.

Before declaring "not found", search the index table (it lists every object name

in the corpus) with a targeted grep rather than reading the whole ~125 KB file: grep -i "<ObjectName>" references/dataandtoolingindextable.md. Match case-insensitively and allow for near-misses (spacing, plural, minor spelling) before concluding the object is absent. Read the full file only for a broad survey.

INVALID_FIELD / No such column

The field is not on that object, or you used the wrong API name. Load the

fields section and confirm the exact name (custom fields end in __c).

The field exists but lacks the needed property: filtering a non-Filter

field, sorting a non-Sort field, or writing a read-only (no Create/Update) field all fail. Check the properties string.

Relationship query fails

Use relationship_name (not the id field) to traverse: Owner.Name, not

OwnerId.Name. Custom lookups traverse with __r.

Confirm refers_to — polymorphic fields (e.g. WhoId, WhatId) refer to

multiple objects and need TYPEOF or the correct relationship.

Tooling query returns nothing / errors

Tooling objects must be queried against the Tooling API endpoint

(/tooling/query), not the standard Data API. Verify with supportedrestapihttpmethods / supportedsoapcalls.

Field is in wsdl_segment but not fields

Complex nested types have their sub-fields in wsdl_segment. Pull just the

matching complexType with jq -r '.wsdl_segment' file.json | grep -A 30 'complexType name="Foo"' instead of loading the whole segment.

Common Objects

Enterprise / Data API (SOQL + DML)

Account: Represents an individual account, which is an organization or person involved with your business (such as customers, competitors, and partners).
Contact: Represents a contact, which is a person associated with an account.
Opportunity: Represents an opportunity, which is a sale or pending deal.
Lead: Represents a prospect or lead.
Case: Represents a case, which is a customer issue or problem.
User: Represents a user in your organization.
Task: Represents a business activity such as making a phone call or other to-do items.
Event: Represents an event in the calendar.

Tooling API (developer records)

ApexClass: Represents the saved copy of an Apex class.
ApexTrigger: Represents the saved copy of an Apex trigger.
ApexCodeCoverageAggregate: Represents aggregate code coverage test results for an Apex class or trigger.
TraceFlag: Represents a trace flag that triggers an Apex debug log at the specified logging level.
EntityDefinition: Provides row-based access to metadata about standard and custom objects.

platform-data-and-tooling-api-context-get

Installation

Summary

Similar popular skills

Also in this package

Agent compatibility

Repository health

Skill metadata

Package contents

History

SKILL.md