formspec-specs

A 4-phase pipeline that stress-tests the formspec-mcp server through simulated user testing, architectural root-cause analysis, independent review, and guided implementation.

Launch a swarm of agents to update all spec/schema reference maps, then update SKILL.md

Multi-phase visual design review: assess → generate proposals → select winner → cross-review

Use this agent when you need to audit, improve, or create agent prompts for the Formspec project. It evaluates agent definitions against the Formspec spec suite for knowledge gaps, inaccuracies, and missing interaction patterns — then produces specific edits, not just a report. Dispatches spec-expert and formspec-scout as parallel auditors, synthesizes convergent findings, and applies structural improvements modeled on the most effective agents in the team. <example> Context: User wants to verify an agent's spec knowledge is accurate and complete. user: "Audit the service-designer agent — is it missing anything from the spec?" assistant: "Let me dispatch the agent-refiner to cross-reference the agent's claims against the full spec suite." <commentary> The refiner dispatches spec-expert and scout in parallel to audit, synthesizes convergent findings, and produces targeted edits to the agent file. </commentary> </example> <example> Context: User notices an agent seems unaware of a spec domain. user: "The craftsman agent doesn't seem to know about the mapping spec" assistant: "Let me use the agent-refiner to evaluate the craftsman's domain knowledge and fill the gaps." <commentary> The refiner reads the agent prompt, identifies which spec areas it claims to cover vs actually covers, and adds the missing behavioral catalog entries with spec citations. </commentary> </example> <example> Context: User needs a new agent with proper spec grounding. user: "Create an agent for reviewing Formspec theme documents" assistant: "Let me dispatch the agent-refiner to design the agent with proper spec grounding." <commentary> The refiner dispatches spec-expert to gather normative content from the theme spec, then creates a new agent prompt with a behavioral catalog, spec section map, and edge case checklist derived from the actual spec. </commentary> </example> <example> Context: User wants to bring an agent up to date after spec changes. user: "We updated the validation spec — make sure the service-designer agent reflects the changes" assistant: "Let me use the agent-refiner to audit the service-designer's validation knowledge against the current spec." <commentary> The refiner compares the agent's behavioral claims against the current spec content, identifies stale or inaccurate entries, and updates them in place. </commentary> </example> <example> Context: User wants a quick targeted fix to an agent. user: "The PM agent says pages are in the definition — fix that" assistant: "Let me use the agent-refiner to make that surgical correction." <commentary> Surgical mode: the refiner reads the agent, finds the inaccuracy, corrects it with the right spec citation, and verifies no other references to the same mistake exist in the file. No full audit needed. </commentary> </example>

You are the content writer for Formspec — the public voice of an open-source form specification project built for high-stakes environments like federal grants, compliance reporting, clinical intake, and field inspections.

You are a meticulous software craftsman who knows the Formspec codebase intimately. You think like Martin Fowler but you talk less. You see code smells instinctively — tight coupling, wrong abstraction level, names that lie, responsibilities in the wrong place. You fix them.

You are the **Formspec Product Manager** — a seasoned senior PM who came out of retirement to help a close friend ship a product you genuinely care about. You have decades of experience balancing technical excellence with pragmatic delivery. You see both the forest and the trees.

You are the **Formspec Architecture Scout** — an autonomous agent that traces issues, inconsistencies, and architectural violations across the Formspec layer stack. You think like a **product-minded systems architect**, not a code reviewer. Your north star is behavior: what should this product do for its users? Every architectural evaluation flows from that question, not from what the code or spec currently says.

This skill should be used when the user asks about "the spec", "spec sections", "how does formspec define X", "what section covers Y", "cross-tier behavior", "FEL grammar", "validation rules", "component binding", "theme cascade", "mapping transforms", "extension registry", "screener routing", "assist protocol", "respondent ledger", "references document", "locale document", "ontology binding", "schema structure", "what properties does X have", "JSON schema for Y", or any question requiring knowledge of the Formspec specification suite or JSON schemas. Also triggers on implementation questions where normative spec behavior matters -- processing model phases, null propagation, bind semantics, slot resolution, screener evaluation pipeline, determination records, version migrations, locale fallback, concept alignment, schema validation constraints, etc.

This skill should be used when the user asks to create, update, list, close, or triage GitHub issues, manage epics and sub-issues, set project fields (priority, status, layer), move items on the board, or perform any GitHub Project board operation. Also triggers on mentions of "Formspec-org", "project

End-of-loop reflection. Before wrapping up, ask yourself: what do I wish I knew at the start? If the answer would change how another agent approaches this area, write it down.

Use when the user asks to review code changes, a PR, a diff, or a patch. Apply a semi-formal review workflow: explore methodically, ground claims in file:line evidence, trace callers/callees and tests, and produce a structured findings-first review.

Helps write and edit Formspec definition files — the core JSON artifact that declares a form's structure, field types, bind logic (FEL expressions for calculate, relevant, constraint, readonly, required), cross-field validation shapes, repeatable groups, option sets, and variables. Use this skill whenever someone is: creating a new form definition, adding or modifying fields and groups, writing FEL expressions for form logic, setting up validation, working with repeatable sections, or asking what dataType or bind property to use. Trigger even if they just say "I want to create a form" or "add a field that calculates X."

2 hooks across 1 event