← 返回 Skills 市场
Ride Receipts
作者
Maxim Tuleyko
· GitHub ↗
· v0.1.12
· MIT-0
475
总下载
0
收藏
0
当前安装
13
版本数
在 OpenClaw 中安装
/install ride-insights-test
功能描述
Build a local SQLite ride-history database from Gmail ride receipt emails using gog for fetch and OpenClaw Gateway /v1/responses for extraction. Use when you...
安全使用建议
This skill appears coherent and implements the described pipeline, but it handles sensitive data: your Gmail receipt HTML/JSON will be written to disk and (by default) sent to a local/private OpenClaw Gateway for extraction. Before installing or running: 1) verify you have and trust a local Gateway and have the gateway token available (OPENCLAW_GATEWAY_TOKEN or ~/.openclaw/openclaw.json); 2) confirm you are comfortable using the gog CLI (it will use your gog auth to read Gmail) and that you select the correct account when prompted; 3) do not set OPENCLAW_ALLOW_NONLOCAL_GATEWAY=1 unless you explicitly accept sending raw receipts to a remote Gateway; 4) review the included Python scripts if you want to audit behavior (they are self-contained and show the data flows); and 5) run the pipeline in an isolated environment if you are concerned about sensitive financial/location data being stored or processed.
功能分析
Type: OpenClaw Skill
Name: ride-receipts-llm
Version: 0.1.12
The skill bundle is a legitimate tool for building a local ride-history database from Gmail receipts. It uses the 'gog' CLI for email retrieval and an LLM gateway for structured data extraction. Security-conscious features are present, such as a host-validation check in 'extract_rides_gateway.py' to prevent sending sensitive data to non-local endpoints by default, and explicit instructions in 'SKILL.md' for the agent to seek user confirmation before processing financial/location data.
能力评估
Purpose & Capability
Name/description match the included scripts. Requiring 'gog' and 'python3' is appropriate for fetching Gmail and running the Python pipeline. The declared gateway config paths align with the code that contacts a Gateway /v1/responses endpoint.
Instruction Scope
SKILL.md and the scripts instruct fetching raw email content (including full HTML) into emails.json, then sending that raw JSON to a Gateway model for extraction. The instructions explicitly require user consent before sending sensitive emails to a model and the extraction script enforces local/private Gateway hosts by default. One inconsistency: the registry lists no required env vars, but the SKILL.md and scripts reference environment variables such as OPENCLAW_GATEWAY_TOKEN, OPENCLAW_GATEWAY_URL, OPENCLAW_ALLOW_NONLOCAL_GATEWAY and also read ~/.openclaw/openclaw.json for the token.
Install Mechanism
No install spec; this is an instruction-only skill with bundled Python scripts. Nothing is downloaded from third-party URLs or installed automatically. Risk from installation is low, but the provided scripts will be executed locally by the agent using python3.
Credentials
The skill does not request unrelated credentials. It legitimately needs a Gateway auth token (sensitive) and a gog-authenticated Gmail account; the token can come from OPENCLAW_GATEWAY_TOKEN or ~/.openclaw/openclaw.json. Registry metadata did not declare any required env vars, so users should be aware the skill will look for the gateway token and optional env flags at runtime. This is proportionate to the task but involves sensitive tokens and Gmail access via the user's gog credentials.
Persistence & Privilege
The skill is not always-enabled and does not request elevated system-wide privileges. It writes its own output files (emails.json, rides.json, rides.sqlite) in the working data directory and does not modify other skills or global agent settings.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install ride-insights-test - 安装完成后，直接呼叫该 Skill 的名称或使用
/ride-insights-test触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v0.1.12
- Added anonymized CSV export: new script `export_anonymized_rides_csv.py` enables export of ride records from SQLite as a shareable, privacy-preserving report.
- The CSV contains only non-personal fields (e.g., provider, month, rounded times, city/country, amounts, distance, duration).
- Update documentation to describe when and how to use export, precise output columns, and export rules.
- No changes to core extraction or database loading; the new feature is a safe reporting option for sharing aggregated ride data.
v0.1.11
- Added summary_rides_sqlite.py script for schema-aware summarization of the rides SQLite database.
- Updated instructions to clarify Gmail account selection flow and prevent assumptions about default account names.
- Enhanced documentation: now requires dynamically inspecting the rides table schema before generating SQL queries or summaries.
- Added new guidance for producing behavioral ride insights—clients should use the extracted JSON plus human-level factual summaries rather than SQL dumps.
- Improved pipeline and output documentation for clearer usage and reliable field handling.
v0.1.10
- Added a metadata section specifying required binaries and Gateway configuration.
- No functional or pipeline changes; documentation and requirements are now more explicit.
- Version alignment and internal documentation update only.
v0.1.9
- Adds explicit warnings about the sensitivity of receipt email data and the contents of emails.json (including full HTML).
- Requires user confirmation before sending raw email data to the Gateway, especially if not using a local/private instance.
- By default, extraction now refuses non-local Gateway URLs for sensitive data unless explicitly overridden by the user.
- Enhances documentation to clarify when and how sensitive data is stored or transmitted.
v0.1.8
Major pipeline simplification: switched to a portable Gateway-based flow with unified JSON artifacts.
- Pipeline now fetches all ride receipts into a single `emails.json` file instead of per-email files.
- Extraction uses the local OpenClaw Gateway `/v1/responses` endpoint for all parsing in `extract_rides_gateway.py`.
- Output rides are written as a JSON array (`rides.json`), then inserted directly into SQLite.
- Removed multiple scripts and references related to manual extraction, repair, and enrichment steps.
- Updated skill and files to reflect the new, streamlined pipeline and naming (`ride-receipts-gateway-llm`).
v0.1.7
- Clarified that the Gateway-backed LLM batch extraction script performs only extraction and repair, not validation or SQLite import; users must run the validator script separately.
- Expanded notes in the batch extraction step to highlight that running validation and import is now a separate, explicit step.
- No file changes detected; documentation update only.
v0.1.6
ride-receipts-llm 0.1.6
- Switched LLM repair/enrichment from embedded subagents to Gateway-backed LLM calls with new prompt templates and scripts.
- Now requires explicit confirmation/environment variable to allow non-local Gateway LLM endpoints for processing sensitive data.
- Replaced and renamed batch extraction/repair scripts for compatibility with Gateway-backed LLM pipeline.
- Updated documentation to match new LLM usage model (sequential, one-at-a-time Gateway requests).
- Removed older subagent-specific scripts and prompt templates.
v0.1.5
- Add support for Gateway-backed batch extraction and repair: requires local OpenClaw Gateway HTTP endpoint and token.
- Update extraction pipeline to include Gateway HTTP LLM interface via `run_embedded_extraction_batch.mjs`.
- Add environment variable documentation for `OPENCLAW_GATEWAY_URL` and `OPENCLAW_GATEWAY_TOKEN`, including default behaviors.
- Clarify that Gateway must be reachable and enabled for batch extraction steps.
- No behavioral changes for deterministic extraction; Gateway batch is optional.
v0.1.4
No changes detected in this version.
- No file or documentation changes were made in version 0.1.4.
v0.1.3
ride-receipts-llm 0.1.3 introduces major improvements to the extraction, repair, and enrichment pipeline for ride receipt processing.
- Switches storage to structured per-email (`data/emails/*.json`) and per-ride (`data/rides/*.json`) files, with new scripts for fetching, extraction, and DB import.
- Adds validation and automated repair workflow for problematic extractions, including LLM-assisted repair and city/country enrichment steps.
- Promotes conservative repair: never overwrite good data with nulls, process city/country enrichment one ride at a time, and prevent field hallucination.
- Expands ride provider support (adds Free Now).
- Introduces workflows for reporting, insight generation, and anonymized CSV export.
- Updates all user-facing descriptions, emphasizing safety, transparency, and reproducibility.
v0.1.2
ride-receipts-llm v0.1.2
- Added Openclaw-compatible metadata for binary/config requirements and homepage link.
- Clarified prerequisite to prefer configured Gmail account, prompting user only if missing.
- Updated instructions to explicitly confirm user consent before sending email HTML to LLM.
- Improved description of sensitive data handling and clarified CLI/tool dependencies.
v0.1.1
- Clarified the skill workflow: build, refresh, export, and query a local rides database using LLM extraction from full Gmail receipt HTML.
- Added explicit prerequisite for `gog` Gmail CLI access and emphasized sensitivity of processed data.
- Improved extraction stage description to highlight one-shot plus targeted repair, non-hallucination, and additive merging rules.
- Streamlined step overviews and updated command examples for better usability.
- Reworded description and section headings for conciseness and clarity.
v0.1.0
Initial release: Extract, process, and store ride receipt data from Gmail with LLM-powered parsing.
- Builds a local SQLite ride history using LLM-based extraction from Uber, Bolt, Yandex Go, and Lyft email receipts.
- Ensures high-quality, provider-agnostic data extraction without template-based parsing.
- Supports loading, exporting, querying, and refreshing ride databases.
- Includes repair logic for missing fields using targeted LLM calls.
- Provides reproducible CLI pipeline for setup, email fetching, extraction, and data insertion.
元数据
常见问题
Ride Receipts 是什么？
Build a local SQLite ride-history database from Gmail ride receipt emails using gog for fetch and OpenClaw Gateway /v1/responses for extraction. Use when you... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 475 次。
如何安装 Ride Receipts？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install ride-insights-test」即可一键安装，无需额外配置。
Ride Receipts 是免费的吗？
是的，Ride Receipts 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Ride Receipts 支持哪些平台？
Ride Receipts 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 Ride Receipts？
由 Maxim Tuleyko（@tuleyko）开发并维护，当前版本 v0.1.12。
推荐 Skills