krystophny/helpy

helpy

Local Go binary for office-workflow automation: web search, TUGonline (lecture / room / exam / candidate / Lehrerhebung), spreadsheets (.xlsx via excelize), Nextcloud shares + versions, ICS feed management, and SAP budget reports. The same surface the helpy MCP daemon serves on 127.0.0.1:8090, driven from bash.

Use this skill whenever the user asks for:

• web search or fetch (web_search, web_fetch)
• TU Graz TUGonline operations (tugonline_*)
• ICS feed subscription and feed-event queries (ics_*)
• xlsx reading, writing, formulas, recalc, charts, pivot tables, styles, validation, etc. (sheet_*)
• Nextcloud share or version management (nextcloud_*)
• TU Graz SAP budget / OData reports (sap_*)
• read-only ops scans / drift reports / dashboard previews (ops_*)

Discover

helpy tools list                       # name<TAB>short-description
helpy tools list --format json         # full schema; pipe through jq
helpy tools list --format json | jq '.[] | select(.name | test("sheet_"))'

Call a tool

# whole arguments object
helpy tools call web_search --args '{"query":"qwen3 long context","limit":5}'

# arguments from a file
helpy tools call sheet_set_range --args-file /tmp/sheet-write.json

# key=value (values parse as JSON if they parse, else string)
helpy tools call ics_events --arg days=14 --arg calendar="teaching"

Stdout is structuredContent (preferred) or joined text content. Exit codes: 0 ok, 1 tool error, 2 JSON-RPC / invocation error.

Common recipes

# web search
helpy tools call web_search --args '{"query":"f2py compatibility python 3.14","limit":10}'

# fetch a single URL as readable text
helpy tools call web_fetch --args '{"url":"https://example.invalid/article","max_chars":6000}'

# TUGonline lecture-participant list
helpy tools call tugonline_lecture_search --args '{"query":"plasma physics 2026"}'

# read a workbook range
helpy tools call sheet_read_range --args '{"file":"/tmp/budget.xlsx","sheet":"Sheet1","range":"A1:F40"}'

# create a workbook with a named sheet
helpy tools call sheet_workbook_edit --args '{"file":"/tmp/report.xlsx","op":"create","sheets":["Data"]}'

# recalc engine: auto (cross-sheet + 3D refs handled natively)
helpy tools call sheet_recalc --args '{"file":"/tmp/budget.xlsx","engine":"auto"}'

# SAP budget summary for an order
helpy tools call sap_budget_report --args '{"order":"100xxxxxx","detail":"summary","report":"budget"}'

xlsx rules (carryover from CLAUDE.md)

• xlsx access only through these helpy tools. Forbidden: soffice/LibreOffice, openpyxl, xlsxwriter, pandas.read_excel|ExcelWriter|ExcelFile, exceljs, xlsx-cli, xlsx2csv, hand-rolled zip pipelines.
• xlsx file paths may be under /tmp and must be absolute. Use the JSON key file, not path.
• For recalc use helpy tools call sheet_recalc with "engine":"auto" (or "none"); 3D and cross-sheet refs work natively.
• Missing xlsx feature → add a helpy tool or extend github.com/sloppy-org/excelize (fork at ~/code/sloppy/excelize, branch 2303-recalc). Do not work around it in Python or shell.

Notes

• Web search via helpy tools call web_search is preferred for live info. Keep search limit around 3-5 unless the user asks for breadth; fetch one chosen result with web_fetch.
• web_search returns { "result": { "query": "...", "results": [...] } }; count results from .result.results | length.
• Do not use helpy for generic personal/work calendar lookups. Use sloptools tools call calendar_events for Google/EWS calendars. Use helpy only for explicit ICS feed operations such as ics_list, ics_events, ics_add, and ics_remove.
• web_fetch defaults to a compact excerpt; pass max_chars only when the task needs a longer page.
• Literature lookups: check ~/Zotero/zotero.sqlite first, only then fall back to web_search.
• All tools are localhost-only; no public bind.