Image Understanding
Skill748 starsChanged 3 months agoMITView on GitHub
Analyze images sent by the user, such as portfolio screenshots or candlestick charts, and combine the result with tools for a fuller assessment
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 3 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
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
- Very specific: line 13
When the user sends an image attachment, first consume the current-turn 【图片文字提取】 block produced by the attachment service.
- Distinctive: line 20
If extraction is partial, answer from the fields that are actually present and ask one minimal confirmation only for the specific number or …
- Distinctive: line 37
Keep the filename boundary when several screenshots overlap
- Very specific: line 38
Only offer portfolio(action="add") after summarizing the extracted values and receiving confirmation.
- Distinctive: line 56
If the message contains multiple image attachments, analyze them one by one
- Very specific: line 58
After identifying a portfolio screenshot, summarize the extracted result and let the user confirm before writing it into portfolio
- Very specific: line 59
Use other current-turn evidence such as portfolio(action="view"), data_fetch, or web_search when it helps answer the question.
≈ “Common” is an estimate: at least 1 in 5 of the 308,232 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
Image Understanding Skill
When the user sends an image attachment, first consume the current-turn
【图片文字提取】 block produced by the attachment service. It is trusted
current-attachment evidence and is grouped by filename. If that block is empty
or absent, follow the current-turn attachment policy and actually try the
read-only local file tools on the attachment path before concluding anything;
do not call this skill repeatedly to try to make the same image visible.
If extraction is partial, answer from the fields that are actually present and ask one minimal confirmation only for the specific number or label that matters to the user's decision. Never replace the whole answer with a generic tool, OCR, or research failure.
Nothing about this pipeline is user-facing. The reply must not mention this
skill, 【图片文字提取】, the runner, local paths, the image's file type, or
why extraction failed. When an image genuinely cannot be read, say only that
this picture did not come through clearly and name the one value you need
("看不清持仓那一栏，成本价是多少？") — never ask the user to re-send it in a
different format.
Supported Scenarios
1. Identify Portfolio Screenshots
- Extract the ticker, company name, share count, cost basis, and similar details from the image
- Keep the filename boundary when several screenshots overlap
- Only offer
portfolio(action="add")after summarizing the extracted values and receiving confirmation; image analysis itself is read-only
2. Analyze Market Charts
- Recognize candlestick trends and technical indicators such as moving averages, MACD, and RSI
- Combine the chart with real-time news from
web_searchfor a broader judgment
Tool Guide
| Tool call | Purpose |
portfolio(action="add", ...) | Record the extracted holdings |
web_search(query="...") | Search for news around the chart's time frame |
data_fetch(data_type="quote", symbol="...") | Get live market data for comparison |
Notes
- If the message contains multiple image attachments, analyze them one by one
- For unclear numbers, always ask the user to confirm instead of guessing
- After identifying a portfolio screenshot, summarize the extracted result and let the user confirm before writing it into
portfolio - Use other current-turn evidence such as
portfolio(action="view"),data_fetch, orweb_searchwhen it helps answer the question; a missing image field is a narrow evidence gap, not authority to refuse the task
In this repository37 other files it gives its agents.
- AGENTS.mdAGENTS.md
- memory/agents.mdAGENTS.md
- alipay-aipaySkill
- event-engine-baseline-testingSkill
- event-engine-push-reviewSkill
- Formal ReleaseSkill
- Analyst CoverageSkill
- Chart VisualizationSkill
- Company Latest DevelopmentsSkill
- Company PortraitSkill
- company-thesis-ratingsSkill
- Deep Stock ResearchSkill
- Earnings ReadoutSkill
- earnings-researchSkill
- ETF AnalysisSkill
- First PrinciplesSkill
- FundamentalsSkill
- Gold AnalysisSkill
- hari-investSkill
- Hone AdministratorSkill
- Image GenerationSkill
- Industry MapSkill
- Influencer ViewsSkill
- Market AnalysisSkill
- Missed Digest ItemsSkill
- Moat AnalysisSkill
- Notification PreferencesSkill
- Options AnalysisSkill
- PDF UnderstandingSkill
- Portfolio ManagementSkill
- Position AdviceSkill
- Scarcity DifferentiationSkill
- Scheduled Task ManagementSkill
- Sector To StockSkill
- Skill ManagementSkill
- Stock ResearchSkill
- Valuation AuditSkill
Does 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.