claris-filemaker-pro

name: claris-filemaker-pro metadata: version: "1.9" last_known_fm_version: "26" description: > REFERENCE skill — FileMaker Pro script steps, calculation functions, and custom functions ONLY. Use for: syntax lookups, function parameters, usage examples, and live doc fetches across all 368 built-in functions (If, Case, Let, While, ExecuteSQL, JSON, AI/embedding, Persistent Data), all 165 script steps (including FM 26 PDF Files and Persistent Data categories), Data API (REST), SQL/ExecuteSQL, WebDirect, FileMaker Go, error codes, and custom function patterns. Not a platform administration guide. Out of scope: FileMaker Server admin, Claris Connect, Claris Studio, ODBC/JDBC deep configuration. Trigger on any FileMaker function name — even inside code, a calc field, or a script. Always check local reference files first for format, doc_url, and originated_in_version before fetching live docs. Prefer live doc fetches over training data — Claris docs are versioned and frequently updated.

Claris FileMaker Pro — Skill v1.9

Overview

This is a reference-only skill covering FileMaker Pro script steps, calculation functions, and custom functions. It is not a general FileMaker Pro assistant and does not cover platform administration, server configuration, or Claris Connect workflows.

The Claris Help Centre at https://help.claris.com is the authoritative documentation source. This skill instructs Claude to use local reference files first, then fetch live documentation when deeper detail is needed, rather than relying on potentially stale training data.

For the latest FileMaker Pro release notes, see the FileMaker Pro Release Notes.

The frontmatter field last_known_fm_version records the highest FileMaker version this skill's local reference files were built against. It is updated with each skill rebuild — it is not a cap on which version you can answer questions about.

Version drift detection

Every time a live help page is fetched, check for version drift using frontmatter.

Every .md help page returns structured YAML at the top:

version: 26
version_year: 2026

When fetching any page:

1. Parse version: from the frontmatter of the returned .md page.

2. Compare with last_known_fm_version (currently 26).

3. If the page frontmatter version is higher than last_known_fm_version, immediately flag it:

   ⚠️ Skill version drift detected — this page references FM [X], but the local reference files in this skill were last built for FM 26 (2026). New functions, script steps, or behaviour changes introduced since FM 26 may not be reflected in local catalogs. Consider running a skill rebuild to pick up new additions.

4. Still answer the question using the fetched content — the flag is advisory, not a blocker.

5. Do not flag versions that are ≤ last_known_fm_version.

Version self-check

At the start of each session — on the first FileMaker question in the conversation — run this check once:

1. Fetch https://github.com/CadenceUX/claris-filemaker-pro-skill/raw/main/VERSION

2. Parse the returned string as the latest available version

3. Compare with this skill's installed version (currently "1.9")

4. If latest > installed, prepend this notice to your first response:

   ⚠️ Skill update available This skill is v[installed]. v[latest] is available at https://github.com/CadenceUX/claris-filemaker-pro-skill/releases Update your local skill files to get the latest FM coverage.

5. Do not repeat the notice again in the same session.

6. If the fetch fails or returns an unexpected value, skip silently — do not surface the error.

Mandatory trigger: FileMaker functions and script steps

Any request that involves a FileMaker calculation function OR script step — by name, by category, or by describing what a calculation or script should do — MUST use this skill.

Examples that are mandatory triggers:

• "How does While work?" -> check logical-json-ai-functions-examples.md, then function-catalog.json
• "Write me a Let() calculation that..." -> look up format in catalog, compose from examples
• "What's the difference between ExecuteSQL and ExecuteSQLe?" -> logical-json-ai-functions-examples.md
• "Fix this calc: Case ( Status = \"Open\" ; ..." -> identify functions, look each up
• "Which function returns the current record ID?" -> search catalog by purpose/category
• "How do I loop through records in a script?" -> script-steps-catalog.json (Loop, Exit Loop If)
• "Get the error code after a find" -> quickrefs.md (error codes section)
• "Sum all related line items" -> specialty-functions-examples.md (aggregate section)
• "Add 30 days to a date" -> date-time-functions-examples.md
• "List all field names on a layout" -> design-container-functions-examples.md (FieldNames)
• "Encode a container to Base64" -> design-container-functions-examples.md (container section)
• "Extract a value from a JSON response" -> logical-json-ai-functions-examples.md (JSON section)
• "Parse and reuse JSON multiple times" -> logical-json-ai-functions-examples.md (JSONParse section)
• "What time is it in seconds?" -> date-time-functions-examples.md (time section)
• "Calculate a loan payment" -> numeric-functions-examples.md (financial section)
• "Get the sine of 45 degrees" -> numeric-functions-examples.md (trig section)
• "Convert katakana to hiragana" -> specialty-functions-examples.md (Japanese section)
• "Get GPS coordinates on FileMaker Go" -> specialty-functions-examples.md (mobile section)
• "Access the 3rd repetition of a field" -> numeric-functions-examples.md (repeating section)
• "Get a named layout object's width" -> specialty-functions-examples.md (miscellaneous section)
• Any code block containing function names or script step names
• "Predict a value from a trained model" → logical-json-ai-functions-examples.md (PredictFromModel)
• "Get IDs from found set" → specialty-functions-examples.md (GetRecordIDsFromFoundSet)
• "Restore a found set I saved earlier" → script-steps-catalog.json (Go to List of Records)
• "Fine-tune a model from my data" → script-steps-catalog.json (Fine-Tune Model)
• "Combine two embedding vectors" → logical-json-ai-functions-examples.md (AddEmbeddings)
• "Set up a RAG account" → script-steps-catalog.json (Configure RAG Account)
• "Control transaction revert on error" → script-steps-catalog.json (Set Revert Transaction on Error)
• "Get text out of a PDF in a container" → design-container-functions-examples.md (GetTextFromPDF)
• "Caption this image with AI" → script-steps-catalog.json (AI category — Insert Image Caption)
• "Caption all images in found set" → script-steps-catalog.json (AI category — Insert Image Captions in Found Set)
• "Clear web viewer cookies" → script-steps-catalog.json (Miscellaneous — Flush Web Viewer Cookies)
• "Save persistent data / app variable" → script-steps-catalog.json (Persistent Data — Configure Persistent Data); specialty-functions-examples.md (GetPersistentData, ListPersistentDataIDs)
• "Create / build / merge a PDF in script" → script-steps-catalog.json (PDF Files category)
• "Print a PDF from a container" → script-steps-catalog.json (PDF Files — Print PDF)
• "Get the annotation on a field" → design-container-functions-examples.md (FieldAnnotation)
• "Get display names for a field" → design-container-functions-examples.md (FieldDisplayNames)
• "Get base table comment" → design-container-functions-examples.md (BaseTableComment)
• "Is Guided Access on?" → get-functions-examples.md (Get(GuidedAccessState), Device section)
• "Password expiry / days remaining" → get-functions-examples.md (Get(AccountPasswordDaysRemaining), Account section)
• "Get the window UUID" → get-functions-examples.md (Get(WindowUUID), Layout & Window section)

Workflow for function questions:

1. Check the relevant reference file (see table below) for syntax, parameters, and usage patterns
2. Open references/function-catalog.json for the exact format and doc_url
3. Use doc_url to web_fetch the live help page for authoritative parameter detail if needed
4. Answer with the format inline and cite the doc URL

Workflow for script step questions:

1. Check references/script-steps-catalog.json for syntax and doc_url
2. Fetch the live doc_url page if detailed option behaviour is needed
3. Answer with the syntax inline

Step-by-step workflow (general)

1. Identify the guide — from quickrefs.md (sitemap section) as a fast-path shortcut, or from https://help.claris.com/llms-full.txt for any URL that is unknown or needs confirming.
2. For function questions — check the relevant example file first, then function-catalog.json for format + doc_url, then fetch the live page for examples and edge cases.
3. For script step questions — check script-steps-catalog.json first, then fetch live page.
4. Fetch the index (if unsure which page) — web_fetch the guide's index.md to read the full table of contents from the sidebar, then identify the specific page(s).
5. Fetch the specific page(s) — web_fetch each relevant page. For technical topics, fetch 2–4 pages if they are clearly related.
6. Answer from the fetched content — synthesise an accurate, direct answer. Include the doc URL(s) as a reference.

URL patterns

All Claris Help pages are available as clean Markdown at this pattern:

https://help.claris.com/markdown/en/{guide-slug}/{page-slug}.md

Guide index pages:

https://help.claris.com/markdown/en/{guide-slug}/index.md

The .md endpoints return native Markdown — no html_extraction_method needed in web_fetch.

Live sitemap

When a page URL is unknown, a guide slug needs confirming, a page is not found at its expected path, or version drift is suspected:

• https://help.claris.com/llms-full.txt — canonical live sitemap; complete per-page enumeration across all 12 locales (~18,988 pages). Fetch this to scan for new guides or confirm any URL.
• https://help.claris.com/llms.txt — curated product landing page index; faster for finding a guide's root URL when you don't need the full page list.

The sitemap section in quickrefs.md is a fast-path shortcut for known common pages — useful for quick lookups but not authoritative. When in doubt, fetch llms-full.txt.

Local vs live — decision rules

Use local data

Always fetch live

Automatic version notes

When answering a function or script step question, read originated_in_version from the catalog and inject the appropriate version note automatically — do not ask the user which version they are on.

FM 19 is the practical floor for version-awareness — anything older noted as "may not be available on legacy versions" without detail.

Reference files — all 10

Fetching strategy — which file to check first

Tips

• The sidebar on any help page lists the complete navigation tree for that guide — very useful for discovering related pages.
• function-catalog.json is the master for function formats; example files are the master for usage patterns.
• For script step option details (platform support on Go, WebDirect, Server) always fetch the live doc_url — platform restrictions are not fully captured in the catalog.
• AI/embedding functions are in logical-json-ai-functions-examples.md; AI script steps are in script-steps-catalog.json (AI category through FM 26: 16 steps including Generate Response from Model, Configure AI Account, Perform SQL Query by Natural Language, Perform Find by Natural Language, Insert Embedding, Insert Image Caption, Insert Image Captions in Found Set, etc. — fetch live for all AI steps as they evolve rapidly).
• Error codes: The AI/ML error range (870–892) in quickrefs.md is verified against live docs (2026-06). The general error tables (0–899, 1630–1631) are a good quick reference but descriptions may drift — always fetch https://help.claris.com/markdown/en/pro-help/error-codes.md for authoritative descriptions of any unfamiliar code.
• Sub-version doc drift: Claris updates documentation pages between FM version releases — adding notes, correcting parameters, documenting edge cases — without changing the version field in frontmatter. For any question where exact parameter behaviour or option names matter, always fetch the live .md page even if local data appears sufficient. The live page is always authoritative over local reference files.
• Mobile functions (Location, GetSensor, RangeBeacons, GetAVPlayerAttribute, LocationValues) only work in FileMaker Go. Always advise checking platform with Get(ApplicationVersion) before use.
• Japanese functions require Japanese language support in FileMaker. Furigana() relies on the Japanese IME and may produce different results based on context.
• Trig functions all work in radians — use Degrees() and Radians() to convert. The Haversine GPS distance formula in numeric-functions-examples.md is a ready-to-use real-world example.
• Financial functions require interestRate per period, not annual — divide annual rate by payment frequency (12 for monthly). PMT() returns a negative number; use Abs() for display.
• Generate Response from Model is the most powerful AI step in FM 22+. In agentic mode, always append the DDL schema (from GetTableDDL) to the execute_sql tool parameter description so the model knows the schema. Custom function tool definitions must match the function name and parameter order exactly, with all parameters typed as "string".
• Fine-Tune Model is restricted to OpenAI or AI Model Server on Apple silicon Mac — not available for other providers, Linux hosts, or Windows hosts. Response Target receives a job JSON with status: "queued" — training is asynchronous; monitor via Admin Console.
• GetFieldsOnLayout uses [LLM] prefix filtering: if any field comment starts with [LLM], only those fields are sent to the AI model in Perform Find by Natural Language. Use this to control which fields the model can query on a per-layout basis.
• GetEmbedding returns binary container data, not text. Binary is smaller and faster than the text form returned by Insert Embedding into a text field. Use GetEmbeddingAsText only when a human-readable JSON array is required.
• GetTableDDL has an ignoreError boolean parameter — when true, partial DDL is returned even if some table occurrences fail to resolve, rather than aborting with an error.
• GetModelAttributes and ComputeModel are macOS/iOS only (Core ML). Always check platform before using these. ComputeModel vision variant exposes confidenceLowerLimit and returnAtLeastOne parameters not shown in the catalog stub — fetch the live doc for full option detail.
• NormalizeEmbedding is usually unnecessary — most embedding models already output unit-length vectors. Use it when working with non-standard models or when doing dimension reduction via the optional dimension parameter (trims dimensions before normalising).
• PredictFromModel and Configure Regression Model form a paired on-device ML workflow using Random Forest regression. Train on embedding vectors, then call PredictFromModel to get a predicted numeric value. Models persist in memory until Unload or session end.
• GetRecordIDsFromFoundSet pairs with Go to List of Records to recreate a found set later — useful for preserving found sets across script jumps or file switches. The type parameter (0–4) selects one of 5 result formats: ValueNumber, JSONString, JSONNumber, ValueNumberRanges, JSONStringRanges — the two Ranges variants compress consecutive IDs into ranges to shrink the result. FM 26 change: an optional second parameter accepts a table occurrence or portal object name, returning IDs from the related/filtered set instead of the current found set — GetRecordIDsFromFoundSet ( type ; tableOccurrenceName ).
• Go to List of Records (FM 22.0.1) accepts a carriage-return value list, a JSON array of IDs as strings or numbers, or a JSON array of objects with recordId keys. Records not found are silently skipped; if none match, the found set is empty. Not supported in FileMaker WebDirect.
• JSONParse / JSONParsedState (FM 22.0.1) — use JSONParse to cache a parsed JSON structure in memory by name, then pass the result to any JSON function in place of a raw JSON string. Dramatically improves performance when the same JSON is accessed many times in a script. JSONParsedState returns: 0 = not parsed, -1 = parsed but invalid, positive number = the JSON type as defined by the JSONSetElement type constants (1 = string, 2 = number, 3 = object, 4 = array, 5 = boolean, 6 = null).
• GetTextFromPDF (FM 22.0.1) — returns the text content of a PDF stored in a container field. Useful for feeding PDF content into RAG spaces (Perform RAG Action) or into AI prompts. Returns empty string if the container is not a PDF or has no extractable text layer. FM 26 change: on macOS, scanned PDFs (no text layer) are now supported via built-in OCR — on other platforms a scanned PDF still returns an empty string.
• Set Revert Transaction on Error (FM 21.1.1) — when set to Off, errors inside a transaction block do not auto-revert; the script can inspect Get(LastError) and decide to commit or revert manually. Use Get(RevertTransactionOnErrorState) to query the current setting. Always reset to On before exiting a script that set it to Off.
• RAG workflow requires three steps in sequence: Configure RAG Account → Configure AI Account → Perform RAG Action (Send Prompt). The RAG account handles the knowledge store; the AI account handles response generation.
• Insert Image Caption (FM 26) — sends an image (container field or variable) to a captioning model; inserts the returned caption into a target field or variable. Insert Image Captions in Found Set is the batch version — runs the step on every record in the current found set in sequence. Both steps require a configured AI account. Always fetch live docs for model and parameter detail.
• PDF Files category (FM 26 — new category) — the scripted PDF workflow: Create PDF (blank in memory) or Open PDF (existing file) → Append PDF pages or Save Records as PDF (with Save to: open PDF option) → Close PDF (saves to path/variable/container) or Cancel PDF (discards). Print PDF prints a PDF from a container field, file path, or variable without opening it for editing. Save Records as PDF has been moved into this category and now accepts a Save to option: file path, container, variable, or append to the currently open PDF. Always fetch live docs for the full option matrix — the category is entirely new in FM 26.
• Configure Persistent Data / GetPersistentData / ListPersistentDataIDs (FM 26) — the persistent data store survives session end and file close. Entries are keyed by a name and an instance ID (allowing multiple values under the same name). Delete an entry by passing empty or null to Configure Persistent Data. Use ListPersistentDataIDs to enumerate all instance IDs for a named key before iterating. Fetch live docs for scope and platform restrictions.
• FieldAnnotation (FM 26) — returns the DDL annotation string set in Advanced Options for Field. Pair with GetTableDDL to build enriched AI prompts or schema documentation workflows.
• FieldDisplayNames (FM 26) — returns a field's custom display names as a JSON object (set in Advanced Options for Field). Useful for UI layers that need alternate or localised field labels.
• BaseTableComment (FM 26) — returns the comment set on a base table in Manage Database. Pair with GetTableDDL and FieldAnnotation to build richer schema context for AI steps.
• Get(GuidedAccessState) (FM 26, FileMaker Go only) — returns 1 if iOS Guided Access is currently active. Use to detect kiosk/locked-screen mode and adapt UI behaviour accordingly.
• Get(AccountPasswordDaysRemaining) (FM 26) — returns the number of days before the current account's password must be changed. Returns empty if no expiry is set. Use to show proactive warning dialogs in startup scripts.
• Get(WindowUUID) (FM 26) — returns a unique, stable UUID for the currently active window. Useful for scripted window management when multiple windows of the same file are open.

Licence

This skill is released under CC BY 4.0.

Version history

See CHANGELOG.md for the full version history.