AGENT SKILL

skilltta-task-guidance

Task-specific operational guidance synthesized at test time from retrieved training examples.

License 许可
Apache-2.0
最近更新
2026-11-14
Stars
★ 0
Forks
1

01 前言 Front Matter

name
skilltta-task-guidance
description
Task-specific operational guidance synthesized at test time from retrieved training examples.

03 SKILL.md 正文

SKILL.md

When to use

Use this skill for multi-step Excel workbook tasks that require:

  • Downloading an external data source (e.g., an IMF/commodity workbook) and extracting a specific series into a named sheet.
  • Populating designated cells across multiple sheets with Excel formulas (not precomputed Python values), because the instruction restricts computation to Excel.
  • Performing financial-style calculations: log returns, rolling volatilities (annualized), cross-sheet lookups (INDEX/MATCH or XLOOKUP), and a final risk metric such as Reserves-at-Risk (RaR).
  • Saving the final workbook to an exact output path.

Before acting, the agent should gather:

  • The exact structure of the target workbook: sheet names, row/column layout of each "Answer" block, and any HINTs (e.g., "multiply log returns by 100").
  • The exact URL/source and which series/units to extract (e.g., "US$ per troy ounce").
  • Which rows correspond to which sub-step, and the expected formula semantics (latest value vs. average vs. annualized).
  • Whether openpyxl (or similar) is used to write formulas as strings while preserving existing sheet content and formatting.

Possible Failure Modes

  • Hardcoding numeric results into cells instead of writing Excel formulas — violates "only use Excel for formula, computation, etc."
  • Downloading the wrong sheet/series from the source workbook (commodity files often have many series; pick gold in USD/troy oz, monthly frequency).
  • Misaligning the monthly time series (e.g., off-by-one when computing log return LN(P_t/P_{t-1}), or forgetting to multiply by 100 per HINT).
  • Volatility mistakes:
    • Using population vs. sample stdev inconsistently.
    • Forgetting annualization factor (*SQRT(12) for monthly data).
    • Using the wrong rolling window (3 vs. 12) or not using the latest window for the Answer sheet.
  • STEP 2 errors:
    • Missing countries that appear only in the Volume sheet with 2025 data — must be added and converted to Value using Jan–Sep 2025 average gold price.
    • Using annual or wrong-period average instead of Jan–Sep average as the 2025 price proxy.
  • STEP 3 errors:
    • Using VLOOKUP or hardcoded joins instead of INDEX+MATCH / XLOOKUP as instructed.
    • Failing to delete countries lacking 2025 Total Reserves data before computing RaR.
    • RaR formula misapplied (typical form: Gold Value × Volatility × Total Reserves ratio, or Gold Value / Total Reserves × Volatility — follow the sheet's row semantics).
  • Overwriting or destroying existing sheets/formatting when writing with openpyxl (use load_workbook, don't recreate).
  • Saving to the wrong path or wrong extension; the target expects /root/output/rar_result.xlsx exactly.
  • Copying concrete numbers, sheet layouts, or country lists from retrieved BigCodeBench examples — those are unrelated pandas exercises, not templates for this task.

Possible procedures

  1. Fetch source data

    • Download the IMF global commodity Excel from the given page. Inspect sheet names; locate the monthly gold price series in USD per troy ounce.
    • Copy the (Date, Price) columns into the workbook's "Gold price" sheet, starting where the template expects.
  2. Compute return/volatility formulas (in Excel)

    • Column C (monthly log return): =LN(B3/B2)*100 from the second data row onward.
    • Column D (3-month volatility): rolling stdev of last 3 log returns, annualized: =STDEV(C_{t-2}:C_t)*SQRT(12).
    • Column E (12-month volatility): =STDEV(C_{t-11}:C_t)*SQRT(12).
    • Decide STDEV vs STDEV.S consistently; document choice.
  3. Fill Answer STEP 1 (rows 3–6)

    • Use references to the latest available row in "Gold price" for 3-mo and 12-mo volatility.
    • For the "3-month annualized" cell, derive from the 3-month volatility cell (already annualized if using SQRT(12); otherwise apply the annualization here).
  4. STEP 2 (rows 11–13)

    • From "Value" sheet: list every country with a non-empty 2025 value; write country names and values via cell references.
    • Cross-check "Volume" sheet: any country with 2025 volume data but absent from "Value" must be appended; compute Value = Volume × (Jan–Sep 2025 average gold price, referenced via AVERAGE over the relevant Jan–Sep cells in "Gold price").
    • Row 13 "Gold price exposure": sum or per-country exposure — read the row label to confirm; typically total gold reserves value.
  5. STEP 3 (rows 20–24)

    • Copy country list, values, and volatility from STEP 2 by cell reference.
    • Row 23 Total Reserves 2025: use =INDEX(TotalReserves!range, MATCH(country, TotalReserves!keys, 0)) or =XLOOKUP(country, keys, values2025, "").
    • Remove (delete row / blank out) any country whose lookup returns no 2025 data before computing RaR.
    • Row 24 RaR: apply the formula implied by the sheet header (commonly Gold Value × Volatility / Total Reserves — verify against the template's row labels).
  6. Persist

    • Open the template with openpyxl.load_workbook, write formulas as strings (starting with =), save to /root/output/rar_result.xlsx. Ensure /root/output exists.

Decision points / recovery checks:

  • If the IMF workbook layout changed, search sheets for the string "Gold" and unit "troy" to locate the correct column.
  • If volatility cells show #DIV/0! or #NUM!, verify enough prior rows exist for the rolling window.
  • If XLOOKUP is unavailable in the Excel engine used, fall back to INDEX+MATCH.
  • If unsure whether a country belongs in STEP 2's "Volume-only" additions, verify it truly has 2025 data and is truly absent from Value (case/whitespace-normalize names).

Verification Checklist

  • [ ] /root/output/rar_result.xlsx exists and opens cleanly.
  • [ ] "Gold price" sheet contains monthly gold USD/troy oz data with columns C (log return ×100), D (3-mo annualized vol), E (12-mo annualized vol), all as Excel formulas, not literal numbers.
  • [ ] STEP 1 answer cells (rows 3–6) reference the latest row of "Gold price" and include the correctly annualized 3-month figure.
  • [ ] STEP 2 (rows 11–12) includes every 2025 country from "Value" plus any 2025-only countries from "Volume", with the latter's values computed as Volume × AVERAGE(Jan–Sep 2025 gold price) via formula.
  • [ ] Row 13 gold price exposure is a formula, consistent with the row's intent.
  • [ ] STEP 3 (rows 20–22) mirrors STEP 2 via cell references; row 23 uses INDEX+MATCH or XLOOKUP against a Total Reserves sheet; countries without 2025 total reserve data are removed.
  • [ ] Row 24 RaR formula is present and evaluates to a finite number for every remaining country.
  • [ ] No cells contain hardcoded results where a formula is required; no computation was done outside Excel and pasted as values.
  • [ ] Existing sheets, headers, and formatting from the template are preserved; only intended cells were modified.
  • [ ] Nothing was copied from unrelated retrieved examples (they are generic pandas tasks, not templates).
  • [ ] Recovery: if any formula shows an error (#N/A, #DIV/0!, #REF!), trace the referenced range and fix before finalizing.

04 同仓库更多 Skills