update-context-from-transcript

name

update-context-from-transcript

description

Use this skill when the user provides Bilibili transcript text, raw ASR output, or transcript-derived evidence and wants to update this repository's context.toml for a specific UP 主. This skill is for maintaining author-specific stock context, adding or refining stock-name corrections, distinguishing broad theme terms from single-stock aliases, and conservatively updating portfolio mappings from transcript evidence.

Update Context From Transcript

Use this skill when the task is to update context.toml from transcript evidence for this repository.

Goal

Convert transcript evidence into conservative, maintainable context.toml updates for the summary-context injection flow.

Valid Update Targets

Use the existing schema only:

1. stock_pool
2. authors.portfolio.stocks
3. authors.alias_overrides
4. authors.theme_terms

Pick the narrowest structure that matches the evidence.

Decision Rules

Prefer transcript over summary

If both raw transcript and LLM summary exist, treat the raw transcript as the primary evidence for:

• spoken abbreviations
• ASR mistakes
• blackwords / slang
• broad topic terms

Use the summary only as secondary confirmation.

Be conservative with single-stock mappings

Only add a stock alias or misrecognition when the evidence strongly points to one specific stock.

Good candidates:

• obvious ASR typo with a clear target
• repeated near-homophone for the same stock
• full-name variants
• clearly author-specific shorthand

Do not map broad terms like:

• CPU
• 存储
• 锂矿
• sector names
• basket phrases

These belong in authors.theme_terms, not as stock aliases.

Use authors.theme_terms for broad concepts

Put broad or multi-stock language into authors.theme_terms, for example:

• theme words
• basket phrases
• repeated combo blackwords
• ambiguous category words

theme_terms are topic hints only. They must not be treated as alias -> single stock.

Do not invent unsupported holdings

Only add a stock into authors.portfolio.stocks when at least one is true:

• the user explicitly says it belongs in the UP 主 stock pool / holdings
• screenshot or direct evidence shows it in the portfolio
• repeated transcript evidence strongly supports that it is a standing tracked name and the user intent is to maintain the author's stock pool

A one-off mention in discussion is not enough.

Keep the file maintainable

• Use formal stock names as stock_pool keys and portfolio.stocks values.
• If the stock code is uncertain, leave code = "".
• Do not duplicate one idea across multiple schema fields unless the semantics are different.

Update Workflow

1. Read the transcript carefully.
2. Extract candidate items:
   • formal stock names
   • likely ASR mistakes
   • likely spoken shorthand
   • broad theme words
3. Classify each item:
   • stock-level correction
   • author-specific alias override
   • theme term
   • insufficient evidence
4. Update context.toml conservatively.
5. Preserve existing user-curated entries unless new evidence directly contradicts them.
6. In the final response, state:
   • what was added
   • what was intentionally not added
   • which terms remain ambiguous

Repository-Specific Guidance

This repository already supports:

• stock_pool."<正式名称>"
• common_misrecognitions
• common_aliases
• [[authors]]
• authors.portfolio.stocks
• authors.alias_overrides
• authors.theme_terms

Do not redesign the schema unless explicitly requested.

Editing Rules

• Edit context.toml directly.
• Keep entries readable and grouped.
• Prefer omission over overfitting when evidence is weak.
• Do not convert broad topic words into single-stock aliases.

Close-Out Expectations

When reporting back, summarize:

• confirmed alias/misrecognition additions
• newly added theme_terms
• terms rejected as too broad or too ambiguous