功能描述

EastMoney Roadshow Digest (Transcript + Summary)｜东方财富路演纪要生成. A reliability-first transcript & summary skill for public EastMoney roadshow replays, focused on...

使用说明 (SKILL.md)

EastMoney Roadshow Digest (Transcript + Summary)｜东方财富路演纪要生成

Name: EastMoney Roadshow Digest (Transcript + Summary)｜东方财富路演纪要生成
Author: sunxq1017-hash

Purpose | 用途

Create a clean, reliability-first workflow for public EastMoney roadshow replay pages only.

仅处理公开可访问的东方财富路演回放页，不依赖本地 cookie、私有接口凭据或用户路径，强调执行可控与结果留痕。本发布包不包含本地 vendor、缓存、输出物或调试残留，目标是可在新环境中按统一入口运行。

Input | 输入

url: string
Must match https://roadshow.eastmoney.com/luyan/\x3Cid>

Output | 输出

Write these files under outputs/:

meta.json
transcript.md
clean_transcript.md
summary.md
brief.md
run_report.md

Page parsing succeeds → always write meta.json. Any downstream failure or downgrade → still write run_report.md.

Workflow | 工作流

Validate URL
Parse page and fetch public metadata
Prefer subtitle candidates if available
Discover replay media and run ASR fallback if needed
Clean transcript
Generate summary
Generate executive brief
Always write run report

Role boundaries | 模块边界

providers/eastmoney.py: validate URL, parse page, fetch public metadata, discover subtitle/media candidates only
providers/asr.py: convert audio to text only
main.py: orchestration, file writing, cleaning, summary generation, reporting

Reliability notes | 可靠性说明

If subtitle is absent, attempt media+ASR fallback
If ASR dependencies are missing, degrade gracefully and explain in run_report.md
If page parsing succeeds, still write meta.json even when downstream steps fail
LLM capability is optional environment capability, not a required dependency of the skill package
When LLM capability exists in the runtime, use it automatically for cleaner transcript / summary / brief generation
When environment-provided LLM capability is available, transcript text or intermediate cleaned text may be sent to an external model service for processing
When LLM capability does not exist, fall back automatically to rule-based outputs and explain the path in run_report.md
Do not bind the skill description to any specific LLM vendor or provider
Keep outputs deterministic and conservative; do not invent facts
Keep README.md, SKILL.md, and manifest.json aligned on scope, inputs, outputs, and dependency expectations

Practical guidance | 实操说明

Use replay metadata from the public detail API
Preserve source path in meta.json
Prefer concise summaries grounded in extracted text
If transcript quality is low, say so explicitly in run_report.md
唯一正确执行方式：EASTMONEY_ROADSHOW_ENTRY=python3-main python3 main.py --url \x3Croadshow_url>
禁止使用 uv run、其他包装器、替代 Python 启动路径或任何未在 README / SKILL 明示的入口
调试、验证、重跑必须与正式执行使用同一标准入口；偏离入口的运行结果不得作为 skill 故障判断依据
LLM 增强按环境中可用的 provider key 自动启用；若无 key，则自动降级到规则版输出

Entry enforcement | 入口约束

main.py contains a startup self-check. If the execution entry does not match the documented standard path, it must fail fast and exit with a clear error instead of continuing in a potentially inconsistent environment.

Dependencies | 依赖

Python 3.9+
requests required
faster-whisper required for ASR
av required as a runtime dependency for faster-whisper
ffmpeg required on PATH for audio extraction / ASR fallback
LLM capability is optional if the host runtime provides it; the skill remains usable without it

Model + keys | 模型与密钥

按环境变量自动检测可用 provider，优先顺序如下：

OpenAI: OPENAI_API_KEY
Anthropic: ANTHROPIC_API_KEY
Google: GEMINI_API_KEY（别名：GOOGLE_GENERATIVE_AI_API_KEY, GOOGLE_API_KEY）
xAI: XAI_API_KEY
OpenRouter: OPENROUTER_API_KEY
Moonshot/Kimi: MOONSHOT_API_KEY（别名：KIMI_API_KEY）

若检测到可用 key，clean transcript / summary / brief 可启用外部模型增强；若未检测到，则自动降级为规则版输出。

启用增强时，transcript 或中间清洗文本可能发送至外部模型服务处理。

Description | 描述

EastMoney Roadshow Digest (Transcript + Summary) is a reliability-first skill for turning public 东方财富路演回放页面 into structured transcript, clean transcript, summary, brief, and run report outputs.

本技能定位为：面向公开东方财富路演回放页的、以稳定性优先的纪要生成工具，可输出 transcript、clean transcript、summary、brief 与 run report。

Validation standard | 验证标准

调试、验证、重跑、正式执行全部使用同一入口：EASTMONEY_ROADSHOW_ENTRY=python3-main python3 main.py --url \x3Croadshow_url>
对发布包的任何验证结论，必须基于该标准入口；偏离入口的结果无效

Non-goals | 非目标

Do not handle private/live-only rooms
Do not log in
Do not require browser cookies
Do not publish automatically

安全使用建议

This package appears to do what it says: parse public EastMoney roadshow pages, prefer subtitles, fall back to media+ASR, clean text, and optionally call an external LLM if an API key is present. Before installing or running it: - Be aware that any transcript/cleaned text will be sent to an external model provider if you have OPENAI_API_KEY, OPENROUTER_API_KEY, or MOONSHOT_API_KEY/KIMI set in the environment. If you do not want transcript data leaving the host, do not set those keys. - SKILL.md / README claim detection for more LLM providers (Anthropic, Google, xAI) but the code only checks OpenAI, OpenRouter, and Moonshot/Kimi—expect this mismatch. If you rely on another provider, verify and/or modify get_llm_config before use. - The skill requires ffmpeg on PATH and faster-whisper + av to run ASR locally; running ASR may be CPU- or disk-intensive. - The package inserts a local .vendor path into sys.path; inspect that directory (if present) before running to ensure no unexpected vendored modules. - Follow the author's enforced entry: set EASTMONEY_ROADSHOW_ENTRY=python3-main when running main.py so the launcher does not exit. - For higher assurance, run the skill in an isolated environment (container/VM) and review the outputs/run_report.md after a run to confirm no unexpected network activity. If you want full parity with SKILL.md's claimed provider list, ask the author to align code and docs.

功能分析

Type: OpenClaw Skill Name: eastmoney-roadshow-digest Version: v0.1.2-rc1 The skill bundle is a legitimate tool for transcribing and summarizing public EastMoney roadshow replays. It uses `faster-whisper` for local ASR and optionally integrates with major LLM providers (OpenAI, Moonshot, etc.) for text refinement, with clear disclosures regarding data being sent to external APIs. The code implements robust URL validation in `providers/eastmoney.py`, structured error handling, and a mandatory environment variable check in `main.py` to ensure correct execution. No evidence of malicious intent, data exfiltration, or harmful prompt injection was found.

能力评估

✓ Purpose & Capability

Name/description match the code: providers/eastmoney.py validates and fetches public page/API data; providers/asr.py extracts and transcribes audio; main.py orchestrates, cleans text, writes outputs, and (optionally) calls LLMs. No unexpected cloud credentials or unrelated system services are requested.

ℹ Instruction Scope

SKILL.md explicitly limits scope to public roadshow URLs and forbids use of cookies/private credentials. The runtime honors that: requests target the public EastMoney detail API and media URLs. The only notable behavior is intentional and declared: when an LLM API key is present in the environment, cleaned transcript text (or intermediate text) will be POSTed to the configured LLM endpoint for enhancement. If no key is present the code falls back to rule-based processing and records downgrades in run_report.md.

✓ Install Mechanism

No installer/remote download steps are present in the manifest. Python dependencies are standard (requests, faster-whisper, av) and ffmpeg is required at runtime; dependencies are declared in requirements.txt. No downloads from arbitrary URLs or extract/install steps are present.

ℹ Credentials

The package does not require environment secrets to operate. It reads optional LLM keys from environment to enable model enhancement (get_llm_config checks OPENAI_API_KEY, OPENROUTER_API_KEY, and MOONSHOT_API_KEY/KIMI). SKILL.md/README list a broader detection order (Anthropic, Google, xAI, etc.) but the code only implements a subset—this is an inconsistency (informational/sloppy) but not an access escalation. No unrelated secrets (AWS, GitHub tokens, etc.) are requested.

✓ Persistence & Privilege

The skill does not request always:true and does not persistently modify other skills or system-wide settings. It writes its own outputs under outputs/ and inserts a local .vendor directory onto sys.path (typical for vendored deps); the entry guard requires a specific env marker but that is a contained enforcement, not privilege escalation.

版本历史

vv0.1.2-rc1

v0.1.2-rc1 LLM Integration Fixed & Enabled Fixed critical issue where LLM enhancement was detected but not actually executed Implemented real LLM call via OpenAI-compatible API Added multi-provider support (OpenAI / OpenRouter / Moonshot-compatible endpoints) Removed hard dependency on any single provider Introduced unified call_llm() interface Behavior Improvements LLM enhancement is now only marked as enabled after successful execution Added robust fallback when LLM is unavailable or fails Improved run report to clearly indicate LLM usage status Architecture Alignment Converted from implicit external calls to explicit, documented LLM enhancement Ensured consistency between code behavior and documentation Maintained compatibility with environments without LLM configuration v0.1.2-rc1 LLM 能力修复并正式接通修复“检测到 LLM 但实际未执行”的核心问题实现基于 OpenAI-compatible API 的真实 LLM 调用支持多模型提供方（OpenAI / OpenRouter / Moonshot 等兼容接口）移除对单一供应商的绑定引入统一的 call_llm() 调用接口行为改进仅在 LLM 实际成功执行后才标记为启用增强失败时自动降级，不影响整体流程 run_report 中明确记录 LLM 是否启用及执行结果架构与规范对齐从“隐式外部调用”改为“显式 LLM 增强能力” 保证代码行为与文档描述完全一致在未配置模型的情况下仍可降级运行

v0.1.2-release-candidate

This release improves reliability, execution consistency, and summary quality: Unified the standard execution entry and prevented invalid invocation paths Stabilized ASR runtime handling and clarified dependency expectations Upgraded clean and summary from rule-based extraction to LLM-driven processing Improved summary readability from raw sentence extraction to structured insights Added explicit documentation for optional external LLM usage to align behavior and disclosure Overall, the skill moves from a demo-level tool to a usable transcript-to-summary workflow. 本次版本优化了稳定性、执行一致性与纪要质量：统一标准执行入口，避免非规范调用导致的异常结果稳定 ASR 运行链路，并明确依赖要求将 clean 与 summary 从规则拼接升级为基于大模型的语义处理将纪要从“原句摘取”提升为“结构化观点提炼”，显著提升可读性补充外部 LLM 可选调用说明，确保行为与文档一致整体上，该技能已从 demo 级工具提升为可实际使用的纪要生成流程。

v0.1.0-release-candidate

EN Structural upgrade to the transcript-to-summary workflow: standardized execution entry, stronger ASR runtime handling, clearer dependency expectations, and better summary quality with optional LLM enhancement when available. This release is published as 0.1.0-release-candidate for final validation before a stable 0.2.0. 中文本次版本对“转写→纪要”链路进行了结构升级：统一执行入口、强化 ASR 运行稳定性、明确依赖要求，并在具备 LLM 能力时提供更高质量的纪要增强。当前先以 0.1.0-release-candidate 发布，用于最终验证，稳定后再升至 0.2.0。

v0.1.0

Initial public release packaging with bounded ASR execution, transcript cleanup, summary outputs, and clean-environment validation.

元数据

Slug eastmoney-roadshow-digest

版本 v0.1.2-rc1

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 4

常见问题