filemaker-ai-core/AGENTS.md

FileMaker AI Core

This repository provides reusable, client-neutral guidance for FileMaker AI work. It is normally consumed as a pinned Git submodule by a client workspace.

Scope and precedence

• Keep this repository free of client names, DDR exports, tickets, production data, credentials, and client-specific business rules.
• Read the consuming project's PROJECT.md before applying this guidance. Its confirmed client decisions, documented conventions, and verified current FileMaker evidence take priority over this core.
• Preserve established legacy patterns unless a confirmed requirement deliberately changes them.

Evidence priority

When sources conflict, use this order unless the project's PROJECT.md documents a stricter project-specific rule:

1. Confirmed current client decision in the ticket or project record
2. Verified current FileMaker behavior or current parsed DDR evidence
3. Other confirmed project documentation
4. An organization-specific shared standard explicitly selected by the project
5. This FileMaker Core guidance

Report meaningful conflicts instead of silently resolving them.

Standard repository map

Use these locations for a workspace that follows the standard project layout. A project may document only deviations in PROJECT.md.

• ../DDR/: local raw FileMaker DDR input. Never copy it into the repository or commit it; move it only as part of an explicitly approved filesystem migration.
• TICKETS/: one folder per ticket or workstream. Keep its source requests, handoffs, analysis, questions, reports, assets, and ticket-specific scripts together. Use a scripts/ subfolder when it helps navigation.
• FMSCRIPTS/: tracked complete .fmscript files that are not clearly ticket-specific or apply across several workstreams; keep one .fmscript file per FileMaker script.
• NOTES/: confirmed knowledge and business rules that apply across more than one ticket.
• DDR-Parsed/: tracked generated FileMaker reference material. Do not edit it manually.
• ddr-parse.sh: run zsh ddr-parse.sh from the project root to parse the latest DDR into DDR-Parsed/ when an up-to-date reference is needed.

Select the relevant skill

Read only the skill that matches the work; do not load every shared reference by default.

• Ticket, request, email, or handoff: skills/ticket-analysis/SKILL.md
• Current implementation, schema, script, calculation, layout, relationship, security, reporting, or integration analysis: skills/ddr-analysis/SKILL.md
• Any .fmscript generation, edit, review, or syntax question: skills/filemaker-fmscript/SKILL.md and the examples it selects
• New or changed FileMaker object, field, variable, parameter, or result names: skills/filemaker-naming/SKILL.md
• Project-manager reply or short billable work report: skills/ticket-delivery/SKILL.md

Standard workflow

1. Put a new ticket, request, or email in its own TICKETS/<ticket or workstream>/ folder.
2. Analyze it with the relevant shared skill before proposing implementation.
3. Review affected current definitions and dependencies in DDR-Parsed/.
4. Turn unclear business rules into concise confirmation questions rather than assumptions.
5. After implementation and DDR regeneration, re-check affected definitions, calculations, layouts, scripts, and report dependencies.
6. Prepare client-facing delivery only after the work is reviewed or when explicitly requested.

Core rules

• Do not invent business rules, schema details, permissions, or implementation status. Separate confirmed behavior, verified current behavior, proposals, and open questions.
• Use parsed DDR as generated reference data. Do not edit it manually; defer to direct FileMaker verification when the two conflict.
• Keep work practical and minimal. Review affected data, permissions, workflows, layouts, reporting, integrations, and regression risk where relevant.
• Do not make production changes, send client messages, or modify project files unless explicitly asked.
• Do not expose credentials or other sensitive values found in source material.

Response format

• Do not hard-wrap Markdown prose. Use line breaks only where Markdown structure or meaning requires them.
• Present every FileMaker script snippet, including partial sections, in a fenced text block and use valid .fmscript format.