multi-script-scaffold

name

multi-script-scaffold

description

Implements the Untitled Placeholder Technique for multi-script systems. Guides the developer through creating N placeholder scripts in FM, captures their IDs via Push Context, then generates all scripts with correct Perform Script wiring in one pass. Use when the user wants to scaffold a set of interdependent scripts. Triggers on phrases like "multi-script", "scaffold scripts", "placeholder technique", "untitled placeholder", or "build a script system".

Multi-Script Scaffold

Implements the Untitled Placeholder Technique: create N placeholders → capture IDs → generate all scripts with correct wiring → deploy → rename.

---

Step 1: Understand the script system

If the developer has not already described the scripts to build, ask:

• How many scripts are needed?
• What does each script do (name + purpose)?
• Which scripts call which (the dependency graph)?

Build a simple dependency table, e.g.:

#	Script Name	Calls
1	Process Invoice	Invoice - Validate, Invoice - Save
2	Invoice - Validate	—
3	Invoice - Save	—

Confirm this with the developer before proceeding.

---

Step 2: Read CONTEXT.json

Read agent/CONTEXT.json. Extract:

• solution — for resolving automation config
• scripts — check if any of the target scripts already exist (by name); if so, note their IDs — those scripts do NOT need placeholders
• current_layout — for context during generation

Identify how many new placeholders are needed (exclude any scripts that already exist in CONTEXT.json).

---

Step 3: Instruct placeholder creation (Tier 1/2) or auto-create (Tier 3)

Read agent/config/automation.json to determine the active deployment tier.

Tier 1 or Tier 2

Tell the developer:

In FileMaker Script Workspace, click the + button N times. FileMaker will name each one New Script. Before running Push Context, rename each placeholder to its final name:

New Script #	Rename to
1st	Script Name A
2nd	Script Name B
…	…

Once all N scripts are renamed, run Push Context (Scripts menu → agentic-fm → Push Context) with the task description: "Scaffold: [brief description]"

Why rename first? FileMaker names every new script New Script. Push Context keys scripts by name — if multiple scripts share the same name, only the last one's ID is captured.

Wait for the developer to confirm Push Context has run.

Tier 3

Confirm with the developer before proceeding:

I'll use AppleScript to create N placeholder scripts in the Script Workspace with their final names, then run Push Context automatically. Ready to proceed?

If confirmed, trigger placeholder creation via POST {companion_url}/trigger with:

{ "fm_app_name": "...", "script": "AGFMScriptBridge", "parameter": "{\"script\": \"...\", \"parameter\": \"...\"}" }

(Consult SKILL_INTERFACES.md for the full Tier 3 AppleScript path. At Tier 3, AppleScript creates scripts with their final names directly — no rename step needed.)

---

Step 4: Re-read CONTEXT.json (capture placeholder IDs)

After the developer confirms Push Context has run, re-read agent/CONTEXT.json.

Locate the newly created Untitled, Untitled 2, … Untitled N scripts in the scripts object and extract their IDs.

Build the assignment map — which placeholder ID maps to which target script:

Placeholder	ID	Will become
Untitled	301	Process Invoice
Untitled 2	302	Invoice - Validate
Untitled 3	303	Invoice - Save

Confirm the mapping with the developer if there is any ambiguity (e.g. more Untitled scripts than expected).

---

Step 5: Generate all scripts

With all IDs resolved, generate every script as fmxmlsnippet XML written to agent/sandbox/.

Naming convention: {Script Name}.xml (spaces replaced with underscores or hyphens, developer preference).

Rules:

1. Use the real numeric IDs from the placeholder map for all <Script id="N" name="..."/> references in Perform Script steps.
2. Follow all conventions in agent/docs/CODING_CONVENTIONS.md.
3. Grep the step catalog for every step type used.
4. Validate each file with python3 agent/scripts/validate_snippet.py agent/sandbox/<file>.xml before proceeding to deployment.

Fix any validation errors before continuing.

---

Step 6: Webviewer output (if available)

Check webviewer availability:

curl -s --max-time 2 -o /dev/null -w "%{http_code}" http://localhost:8080

If reachable (HTTP 200), push each script as a preview payload:

curl -s -X POST http://local.hub:8765/webviewer/push \
  -H "Content-Type: application/json" \
  -d '{"type": "preview", "content": "<HR script text>"}'

Push scripts sequentially so the developer can review each in Monaco before deployment begins.

If not reachable, output each script in HR format to the terminal.

---

Step 7: Deploy

Deploy each script using agent/scripts/deploy.py, targeting its corresponding placeholder by script name.

Tier 1: For each script:

python3 agent/scripts/deploy.py agent/sandbox/<Script Name>.xml "<Placeholder Name>"

Present instructions in the standard format:

The script is on your clipboard. To install it:

1. Open Untitled [N] in Script Workspace
2. ⌘A — select all existing steps
3. ⌘V — paste

Repeat for each script. Present all paste instructions up front so the developer can work through them in sequence without waiting.

Tier 2: deploy.py auto-pastes into each placeholder. Confirm success for each before moving to the next.

Tier 3: deploy.py handles everything. Report result per script.

---

Step 8: Final verification (optional)

Suggest running a context refresh after renaming to confirm all script IDs are correctly wired:

Once you've renamed all scripts, run Push Context again and I can verify the IDs match the wiring in the generated scripts.

---

Notes

• Always confirm the placeholder-to-script mapping before generating code — a wrong assignment means all Perform Script calls in that script will target the wrong script.
• If the developer has already created scripts in a prior session and knows the IDs, skip Steps 3–4 and use those IDs directly.
• Scripts with no inter-script dependencies can be generated without placeholders — use existing IDs from CONTEXT.json directly.
• The webviewer push is per-script, not a batch — send one preview per script so the developer can review them in sequence.