fund-screenshot-digitization
Skill4 starsChanged 3 days agoMITView on GitHub
Digitize fund transaction screenshots into accurate, page-level Excel records and a validated single-sheet statistics input. Use when a fund website cannot export transactions, screenshots are the source of truth, or an existing batch needs careful reprocessing, merging, or QA.
In this repository10 other files it gives its agents.
The file
Show its rules and text
We split this skill into the rules it gives an agent
13 rules. About 0 are common, 9 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈9
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Use this skill for high-fidelity transcription of Chinese fund transaction screenshots.
- Distinctive: line 15
Never assume every page has 20 rows.
- Distinctive: line 18
Use one agent per page (or a small disjoint page range).
- Distinctive: line 19
Read every visible row from the original image.
- Very specific: line 20
Keep --, an explicitly blank field, and numeric zero distinct.
- Distinctive: line 21
Treat fund codes as six-character text, including leading zeroes.
- Very specific: line 25
Prefer the supplied page-13 template when the user requests matching formatting.
- Distinctive: line 27
Run a formula/error scan and render or otherwise inspect the populated sheet before accepting the page.
- Very specific: line 30
Run scripts/merge_workbooks.py after page artifacts are complete.
- Very specific: line 32
Run scripts/validate_workbooks.py on both the page directory and merged output.
- Distinctive: line 36
Use 4–8 agents for throughput.
- Distinctive: line 40
Do not delete source screenshots or accepted page artifacts during processing.
- Distinctive: line 48
Run scripts with the workspace's maintained Python environment when one is specified.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Fund Screenshot Digitization
Use this skill for high-fidelity transcription of Chinese fund transaction screenshots. Treat the image—not OCR guesses, fund knowledge, or neighboring pages—as authoritative.
Workflow
- Inventory and order sources
- List image files with
rg --filesor the equivalent and sort by filename using the host's locale-independent ordering. - Assign page numbers from that sorted list. Record the source filename in every page result.
- Never assume every page has 20 rows; count visible rows and preserve short pages.
- List image files with
- Transcribe one page at a time
- Use one agent per page (or a small disjoint page range). Agents may read images and write only their own page artifact; they must not write a shared workbook.
- Read every visible row from the original image. Preserve product names, transaction types, status text, punctuation, decimals, and units exactly.
- Keep
--, an explicitly blank field, and numeric zero distinct. Do not infer hidden time, fund code, account, or amounts. - Treat fund codes as six-character text, including leading zeroes. Keep request and confirmation values numeric with separate unit columns.
- Mark an unclear field as null and record the reason in the agent's review note; do not silently repair it from another page.
- Write the page artifact
- Prefer the supplied page-13 template when the user requests matching formatting. Keep the two standard sheets:
交易记录for human review and数字字段for machine use. - Populate all visible rows, preserve the source filename, and leave unused template rows empty on short pages.
- Run a formula/error scan and render or otherwise inspect the populated sheet before accepting the page. A rendering API failure is not a data pass: report it and perform a structural check.
- Prefer the supplied page-13 template when the user requests matching formatting. Keep the two standard sheets:
- Merge only after page QA
- Run
scripts/merge_workbooks.pyafter page artifacts are complete. It selects the best artifact for each page, reads the numeric schema, and writes one sheet named交易明细. - The merged columns are:
页码,页内序号,交易日期,交易时刻,产品名称,基金代码,业务类型,申请数值,申请单位,确认数值,确认单位,关联账户,状态,来源文件. - Run
scripts/validate_workbooks.pyon both the page directory and merged output. Sort by page then row, and reject missing pages, duplicate(页码, 页内序号), unexpected row counts, malformed codes, or formulas/errors.
- Run
Parallelism and handoff
- Use 4–8 agents for throughput; one page per agent is safest.
- Give every agent a disjoint output filename such as
交易记录_第014页_数字化_精细复核版.xlsx. - The parent agent owns inventory, exception review, merging, final validation, and delivery.
- For low-quality pages or high-value numeric fields, start a second independent transcription only for the flagged page and compare row-by-row.
- Do not delete source screenshots or accepted page artifacts during processing. Clean only named temporary crops, renders, OCR dumps, and scripts after delivery.
Bundled scripts
scripts/merge_workbooks.py: deterministic page selection, schema normalization, one-sheet merge, and numeric formatting.scripts/validate_workbooks.py: page coverage, row uniqueness, schema, fund-code, formula, and merged-file checks.references/data-contract.md: field meanings and rules for null,--, units, and short pages.
Run scripts with the workspace's maintained Python environment when one is specified. Example:
python scripts\merge_workbooks.py --input-dir D:\Download\天天基金交易记录 --output D:\Download\天天基金交易记录\天天基金交易记录_合并统计输入.xlsx --expected-pages 190
python scripts\validate_workbooks.py --input-dir D:\Download\天天基金交易记录 --expected-pages 190
python scripts\validate_workbooks.py --merged D:\Download\天天基金交易记录\天天基金交易记录_合并统计输入.xlsx --expected-pages 190Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

