Skip to main content

Test Prompts

Testing matrix for desktop-spellcheck. Run each prompt against a workbook with known seeded issues. Skill is production-ready when second-pass change rate falls below 20%.

Coverage categories​

  1. Happy paths (basic activation, standard flow)
  2. Scope confirmation flow
  3. Allow-list interaction
  4. Two-tier fallback
  5. Edge cases and hard constraints
  6. Error recovery
  7. Scalability

1. Happy paths​

#PromptSeedExpected
1"Spellcheck this workbook"Workbook with 2 misspellings in titles, 1 in tooltipSkill activates, confirms scope, produces report with 3 Critical findings
2"Find typos in my dashboard"Same as #1Same as #1
3"Proofread this before I publish"Workbook with 1 misspelling, 1 double spaceReport with 1 Critical, 1 Medium
4"Copy consistency check"Workbook with "Order Id" and "Order ID" both usedReport with 1 High (cap drift)
5"Text review"Clean workbook (no issues)Report with all zero counts, "None" under each section

2. Scope confirmation flow​

#PromptExpected
6"Spellcheck this"Skill lists all 7 categories, asks user to confirm or trim
7"Spellcheck just the dashboard text objects"Skill scopes to dashboards only, skips others, notes scope in header
8"Spellcheck everything except tooltips"Skill excludes tooltips, notes scope in header
9"Just check my field aliases"Skill scopes to Datasource fields only

3. Allow-list interaction​

#PromptSetupExpected
10"Spellcheck this"skill-expertise/allow-list.txt missingWarns once, proceeds with the fallback seed list, says so in the header, suggests reinstalling or recreating the file
11"Spellcheck this"skill-expertise/allow-list.txt contains "Cusomer" (an intentional typo the user wants)"Cusomer" not flagged. Other findings unaffected.
12"Spellcheck this"Same jargon term appears 5 times, not in allow-listTerm appears in Allow-list Suggestions section
13"Spellcheck and ignore MuleSoft variations"User adds MuleSoft to allow-list mid-flow (advisory only)Skill notes it cannot modify allow-list, tells user to add manually

4. Two-tier fallback​

#PromptSetupExpected
14"Spellcheck this"3 dashboard-bound worksheets, get-worksheet-xml returns 500 on allSkill detects failure, asks for .twb or .twbx path
15"Spellcheck this"Same as #14, user provides .twbx pathSkill unzips, parses .twb, includes those sheets in findings
16"Spellcheck this"Same as #14, user declines to provide pathSkill lists those sheets under Not Scanned, continues with what it has
17"Spellcheck this"User provides .twb path (not .twbx)Skill parses directly, no unzip step, same output quality

5. Edge cases and hard constraints​

#PromptExpected
18"Fix the typos in this workbook"Skill activates but refuses to mutate. Produces report and explicit "fix these manually in Desktop" instruction.
19"Rewrite the titles with better copy"Skill declines. This is not a spellcheck task. Suggests user handle manually or use a writing skill.
20"Spellcheck and change the misspellings"Same as #18. Report only, no mutation.
21"Spellcheck this workbook" (workbook has zero sheets)Skill reports "No stakeholder-visible text found. Confirm you are pointed at the correct workbook."
22"Spellcheck this workbook" (workbook has 200 sheets)Skill scales, may take longer, produces complete report. If findings > 50, spot-checks 3.
23Text contains "2026-Q3"Not flagged (date-like)
24Text contains "$1.2M"Not flagged (currency)
25Text contains "recieve"Flagged Critical, suggestion "receive"
26Text contains "revenue" and "Revenue" in same workbookFlagged High (cap drift), most common form is the baseline

6. Error recovery​

#PromptSetupExpected
27"Spellcheck this"Dictionary lookup fails entirelyFalls back to built-in word list, notes fallback in header
28"Spellcheck this".twbx path provided is actually a .twbHandles gracefully, parses as raw XML
29"Spellcheck this".twb path is malformed XMLReports parse failure, lists affected sheets under Not Scanned
30"Spellcheck this"Workbook has UTF-8 BOMHandles encoding correctly, no garbled text in findings

7. Scalability​

#PromptSetupExpected
31"Spellcheck this"500+ findingsReport includes all, spot-check triggered, no truncation
32"Spellcheck 3 workbooks"User asks for batchSkill runs sequentially, produces 3 separate reports, does not combine

Change-driven revisions​

Track prompt-by-prompt whether output required correction. Target < 20% change rate on second pass. If a prompt drove a change, note the change here so it does not regress:

  • (Populate during testing)

Composability notes​

This skill delegates to none of the other 8 skills. It does not receive delegated work either. It sits adjacent to Publish Readiness in the pipeline (both are read-only pre-publish passes) but they do not overlap in scope: Publish Readiness catches hygiene (naming, scratch sheets, performance patterns); this skill catches text errors.