petrowsky/script-lookup

Script Lookup

Locate a FileMaker script by ID or name, resolving to the paired human-readable and Save-As-XML files. Optimized for minimum tool calls.

Performance target: 4 tool calls for ID-based lookups, 5 for name-based.

Interpreting the user's request

Script ID extraction

Treat these as script IDs:

• "ID 123", "script 123", "script id: 123", "#123"

Script name extraction

If no ID is present, treat the remainder as a script name hint, e.g.:

• "review the new invoice for client script" → name hint: "new invoice for client"

Normalize name hints:

• Case-insensitive
• Remove the trailing word "script"
• Collapse repeated whitespace/punctuation

Lookup workflow

Step 1 — Index lookup (PARALLEL)

This is the critical optimization. agent/context/{solution}/scripts.index is a pipe-delimited file (ScriptName|ScriptID|FolderPath) covering every script in the solution. Use it as the primary lookup source — never start with filesystem traversal.

Run these in parallel (single message, multiple tool calls):

Tool call A — Grep the index:

• ID-based: grep "|{id}|" agent/context/*/scripts.index — returns the exact match plus the solution name from the file path.
• Name-based: grep -i "{name_hint}" agent/context/*/scripts.index — returns matching rows. If the user also mentions a solution name, include it in the grep path: agent/context/{solution_hint}*/scripts.index.

Tool call B — List sandbox:

• ls agent/sandbox/ — check for any existing fmxmlsnippet files (in-progress work).

Step 2 — Resolve paths and read excerpt

From the index result, extract:

• Script name (column 1)
• Script ID (column 2)
• Folder path (column 3)
• Solution name (from the index file path: agent/context/{solution}/scripts.index)

Multi-solution handling: If index results come from multiple solution paths, or the user mentions a specific solution, resolve to one. If ambiguous, use AskUserQuestion to ask which solution.

Derive file paths — the sanitized filename follows the pattern {ScriptName} - ID {ScriptID}.txt, nested inside a folder whose name starts with the folder path value from the index. Since the folder's own ID suffix isn't in the index, use a glob:

• Sanitized: agent/xml_parsed/scripts_sanitized/{solution}/{FolderPath}*/{ScriptName} - ID {ScriptID}.txt
• Save-As-XML: agent/xml_parsed/scripts/{solution}/{FolderPath}*/{ScriptName} - ID {ScriptID}.xml

For top-level scripts (empty folder path in index): files are directly in agent/xml_parsed/scripts_sanitized/{solution}/.

Sandbox match: From the sandbox listing (Step 1B), check if any file matches the script name (sandbox files don't include IDs — match by name).

Run in parallel:

Tool call A — Glob for sanitized file: Use Bash: ls agent/xml_parsed/scripts_sanitized/"{solution}"/"{FolderPath}"*/*"ID {ScriptID}"* 2>/dev/null || ls agent/xml_parsed/scripts_sanitized/"{solution}"/*"ID {ScriptID}"* 2>/dev/null

Tool call B — Read first 20 lines of the sanitized script (for the excerpt). If you already know the exact path from the glob, read it directly. Otherwise, combine with tool call A by using a single Bash command that finds and reads:

file=$(find "agent/xml_parsed/scripts_sanitized/{solution}" -name "*ID {ScriptID}*" -type f | head -1) && head -20 "$file"

For name-based lookups where multiple index rows matched: pick the best candidate using the matching rules below, then resolve paths for that candidate.

Step 3 — Script match report + confirmation

Present the report and confirm in one response:

Selected script

• Name: {script name}
• ID: {id}
• Confidence: High/Medium/Low (why)

Paths found

• Sanitized (readable): {path or "not found"}
• Save-As-XML (reference): {path or "not found"}
• fmxmlsnippet (editable base): {path in agent/sandbox, or "not found"}

Alternates (if any)

• Up to 3–5 other candidate scripts from the index results (name + ID)

Quick excerpt

• First few lines from scripts_sanitized to confirm identity

Then use AskUserQuestion: "Is this the correct script? — {Script Name} (ID: {id}) in {solution}"

• Options: yes — "Yes, proceed" / no — "No, that's not it — let me clarify"

Step 4 — Post-confirmation

If confirmed:

• If an fmxmlsnippet already exists in agent/sandbox/, use it as the editable base.
• If none exists, convert the Save-As-XML source:

  python3 agent/scripts/fm_xml_to_snippet.py "agent/xml_parsed/scripts/{solution}/{folder}/{ScriptName} - ID {ScriptID}.xml" "agent/sandbox/{ScriptName}.xml"
  

• Proceed with the next action (handoff to review/refactor, or simply present the script).

If declined:

• Ask for a corrected script name or ID.
• Re-run from Step 1.

Name-based matching rules

When the index grep returns multiple rows, rank candidates:

1. Exact name match (case-insensitive) — highest confidence
2. Contains match (all tokens from the hint present in the candidate name)
3. Fuzzy match (most tokens match) — pick the best, include alternates in report

Pick the best candidate and continue. The confirmation step is the redirect gate — don't block on a separate disambiguation question unless confidence is truly Low across all candidates.

Fallback: no index available

If agent/context/ has no scripts.index files, fall back to filesystem search:

1. ls agent/xml_parsed/scripts_sanitized/ — determine solution subfolders.
   • One subfolder → use automatically.
   • Multiple → AskUserQuestion to disambiguate.
2. Search within the solution subfolder:
   • ID: find "agent/xml_parsed/scripts_sanitized/{solution}" -name "*ID {id}*" -type f
   • Name: find "agent/xml_parsed/scripts_sanitized/{solution}" -iname "*{hint}*" -type f
3. Continue from Step 2 (resolve paths and read excerpt).

If agent/xml_parsed/ does not exist or is empty, report that explicitly and stop.

Mapping between sanitized and Save-As-XML variants

The sanitized and XML variants are a pair sharing the same name pattern: {ScriptName} - ID {ScriptID} with .txt vs .xml extension, in mirrored folder structures under scripts_sanitized/ vs scripts/.

When ID and name conflict, trust ID.

Handoff: when the user asked to "review" or "refactor"

If the user request is a review/refactor/optimization:

• Use this lookup to identify the correct script and its artifacts.
• Then follow the existing script-review or script-refactor workflow:
  • Prefer an existing fmxmlsnippet version in agent/sandbox/ as the base.
  • If none exists, translate from Save-As-XML using agent/scripts/fm_xml_to_snippet.py.

Examples

Example 1 — ID-based lookup (fast path)

User: "Lets work on script 104"

Step 1 (parallel):

• Grep: grep "|104|" agent/context/*/scripts.index → Quick Find Clients|104|Quick Find in Invoice Solution
• List: ls agent/sandbox/ → check for existing "Quick Find Clients" file

Step 2 (parallel):

• Glob: ls agent/xml_parsed/scripts_sanitized/"Invoice Solution"/"Quick Find"*/*"ID 104"*
• Read: first 20 lines of the matched sanitized file

Step 3: Report + confirm → "Is this the correct script? — Quick Find Clients (ID: 104) in Invoice Solution"

Step 4: On confirmation → convert to sandbox if no fmxmlsnippet exists.

Tool calls: 4 (2 parallel + 1 read/glob + 1 confirm) + 1 convert if needed.

Example 2 — Name-based lookup (fuzzy)

User: "Let's work on the invoices quick find for the invoice solution"

Step 1 (parallel):

• Grep: grep -i "quick find" agent/context/"Invoice Solution"/scripts.index → multiple matches:
  • Quick Find Clients|104|Quick Find
  • Quick Find Invoices|106|Quick Find
  • Quick Find Products|108|Quick Find
  • Quick Find Staff|110|Quick Find
• List: ls agent/sandbox/

Best match: "Quick Find Invoices" (contains "invoices" token from hint).

Step 2: Resolve paths for ID 106, read excerpt.

Step 3: Report with alternates (104, 108, 110) + confirm.

Step 4: Convert on confirmation.

Tool calls: 5 (2 parallel + 1 resolve + 1 confirm + 1 convert).

Example 3 — Multiple solutions

User: "Review the New Invoice script"

Step 1:

• Grep: grep -i "new invoice" agent/context/*/scripts.index → results from two solutions

Step 1.5: AskUserQuestion to disambiguate solution, then continue from Step 2.

Example 4 — Ambiguous name

User: "Show me the invoice script"

Step 1 (parallel):

• Grep: grep -i "invoice" agent/context/*/scripts.index → many matches
• List: ls agent/sandbox/

Pick best candidate, include alternates prominently. Confirmation step acts as redirect gate.