Token导航 LogoToken导航TokenDH.com
研究检索需要联网clawhub未标认证来源可访问clear审计提醒

ocas-look奥卡斯外观

Agent Skill

ocas-look 用于查找、检索和筛选相关信息,适合在 OpenClaw 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

2,752

周安装

117

GitHub Stars

公开资料未说明

下载量

964
OpenClaw

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

MIT-0

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:ocas-look(奥卡斯外观)
来源仓库:https://github.com/indigokarasu/ocas-look
安装命令:
openclaw skills install ocas-look
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 OpenClaw 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

ClawHubOpenClaw
openclaw skills install ocas-look

简介

将用户图像转换为日历事件、膳食宏等可执行文档草稿。

  • 适用于收据、产品或公民报告的自动化录入场景。
  • 安装后上传图片即可获得结构化操作建议。ocas-look 属于研究检索类 Skill,可作为该场景下的辅助能力补充。
  • 需确保图像清晰且包含有效信息,否则识别率下降。
  • 输出为草稿,需经用户确认后方可执行相关动作。

SKILL.md

name
ocas-look
source
https://github.com/indigokarasu/look
install
openclaw skill install https://github.com/indigokarasu/look
description
Use when converting a user-provided image into a validated, decision-ready action draft: events from flyers, macros from meal photos, places to save, products to price, receipts to log, documents to file, or civic issues to report. Trigger phrases: 'look at this image', 'what is this', 'scan this receipt', 'what event is this', 'how many calories', 'save this place', 'update look'. Do not use for generic OCR, computer vision research, or surveillance.
metadata
{"openclaw":{"emoji":"👁️"}}

Look

Look bridges the physical world and the digital agent — it takes a user-provided image, infers what the user probably wants done with it, and produces a validated, decision-ready action draft across domains including calendar events, meal macros, places, product comparisons, receipts, documents, and civic reports. It resolves ambiguity through research and option reduction before asking any clarifying questions, and nothing executes without explicit per-draft confirmation.

When to use

  • A flyer or poster → calendar event or ticket purchase draft
  • A meal photo → macro estimation
  • A storefront or sign → save to try-list
  • A product photo → pricing comparison or order draft
  • A civic issue photo → 311 report draft
  • A receipt → expense entry
  • A document → searchable PDF and filing draft

What this skill does not do

  • Generic OCR utility
  • Universal computer vision toolkit
  • Background surveillance or tracking
  • Generic automation framework

Responsibility boundary

Look owns image-to-action conversion: ingest, context inference, domain routing, draft generation, and execution with confirmation.

Look does not own: web research (Sift), knowledge graph writes (Elephas), preference persistence (Taste), communications (Dispatch).

Supported domains

Events, food macros, places, products, civic issues, receipts, documents.

Read references/domain_playbooks.md for detailed per-domain behavior.

Core workflow

  1. Ingest image(s) with EXIF if available
  2. Merge extraction evidence (OCR, entities, context)
  3. Infer context and likely intent
  4. Route to appropriate domain
  5. Research and validate externally
  6. Filter by constraints (dietary, preferences, permissions)
  7. Reduce options before asking questions
  8. Generate 1-3 decision-ready ActionDrafts
  9. Emit Signal files for extracted entities (places, events, products) to ~/openclaw/db/ocas-elephas/intake/{signal_id}.signal.json. Use Signal schema from spec-ocas-shared-schemas.md.
  10. Write journal via look.journal

Clarification happens only after option reduction, not before.

Commands

  • look.ingest.image — ingest image(s) with optional EXIF and device pre-parse
  • look.propose.actions — generate ActionDrafts with DecisionRecords
  • look.execute.action — execute a confirmed draft (requires explicit approval)
  • look.rollback.action — attempt rollback for reversible actions
  • look.status — last ingest, pending drafts, items awaiting confirmation
  • look.config.set — update configuration
  • look.journal — write journal for the current run; called at end of every run
  • look.update — pull latest from GitHub source; preserves journals and data

Confirmation and rollback rules

  • Draft-first always. No execution without explicit confirmation.
  • High-risk actions (purchases, 311 submission, health writes): require per-draft confirmation token.
  • Reversible actions (calendar, maps): expose rollback information.

Permission discipline

Default deny. Request minimally. Drafting continues even without execution permissions. Blocked execution reported, not silently skipped.

Boundaries

  • Never invent OCR text, barcodes, prices, or license plates
  • EXIF capture location is not the event venue
  • iOS relay pre-parse is optional evidence, not truth
  • The skill must work without relay upload

Storage layout

~/openclaw/data/ocas-look/
  config.json
  state.json
  events.jsonl
  decisions.jsonl
  reports/
  artifacts/

~/openclaw/journals/ocas-look/
  YYYY-MM-DD/
    {run_id}.json

Default config.json:

{
  "skill_id": "ocas-look",
  "skill_version": "2.3.0",
  "config_version": "1",
  "created_at": "",
  "updated_at": "",
  "domains": {
    "events": true,
    "food": true,
    "places": true,
    "products": true,
    "civic": true,
    "receipts": true,
    "documents": true
  },
  "user_profile": {
    "diet": "vegetarian"
  },
  "commerce": {
    "auto_purchase": false
  },
  "retention": {
    "days": 30,
    "max_records": 10000
  }
}

OKRs

Universal OKRs from spec-ocas-journal.md apply to all runs.

skill_okrs:
  - name: draft_accuracy
    metric: fraction of ActionDrafts accepted without modification
    direction: maximize
    target: 0.75
    evaluation_window: 30_runs
  - name: domain_routing_accuracy
    metric: fraction of images routed to correct domain on first attempt
    direction: maximize
    target: 0.90
    evaluation_window: 30_runs
  - name: confirmation_compliance
    metric: fraction of high-risk actions requiring confirmation before execution
    direction: maximize
    target: 1.0
    evaluation_window: 30_runs

Optional skill cooperation

  • Sift — web research for validation during draft generation
  • Elephas — emit Signal files for extracted entities after draft generation

Journal outputs

Observation Journal — all image ingestion and draft generation runs.

Initialization

On first invocation of any Look command, run look.init:

  1. Create ~/openclaw/data/ocas-look/ and subdirectories (reports/, artifacts/)
  2. Write default config.json and state.json if absent
  3. Create empty JSONL files: events.jsonl, decisions.jsonl
  4. Create ~/openclaw/journals/ocas-look/
  5. Ensure ~/openclaw/db/ocas-elephas/intake/ exists (create if missing)
  6. Register cron job look:update if not already present (check openclaw cron list first)
  7. Log initialization as a DecisionRecord in decisions.jsonl

Background tasks

Job nameMechanismScheduleCommand
look:updatecron0 0 * * * (midnight daily)look.update
openclaw cron add --name look:update --schedule "0 0 * * *" --command "look.update" --sessionTarget isolated --lightContext true --timezone America/Los_Angeles

Self-update

look.update pulls the latest package from the source: URL in this file's frontmatter. Runs silently — no output unless the version changed or an error occurred.

  1. Read source: from frontmatter → extract {owner}/{repo} from URL
  2. Read local version from skill.json
  3. Fetch remote version: gh api "repos/{owner}/{repo}/contents/skill.json" --jq '.content' | base64 -d | python3 -c "import sys,json;print(json.load(sys.stdin)['version'])"
  4. If remote version equals local version → stop silently
  5. Download and install:
   TMPDIR=$(mktemp -d)
   gh api "repos/{owner}/{repo}/tarball/main" > "$TMPDIR/archive.tar.gz"
   mkdir "$TMPDIR/extracted"
   tar xzf "$TMPDIR/archive.tar.gz" -C "$TMPDIR/extracted" --strip-components=1
   cp -R "$TMPDIR/extracted/"* ./
   rm -rf "$TMPDIR"
  1. On failure → retry once. If second attempt fails, report the error and stop.
  2. Output exactly: I updated Look from version {old} to {new}

Visibility

public

Support file map

FileWhen to read
references/schemas.mdBefore creating evidence, drafts, or receipts
references/domain_playbooks.mdBefore domain routing or draft generation
references/decision_policy.mdBefore risk assessment or confirmation decisions
references/command_reference.mdBefore any command execution
references/storage_and_config.mdBefore config changes or storage operations
references/journal.mdBefore look.journal; at end of every run

适合场景

01

OpenClaw 用户查找和安装 Skill 时

02

用户想查找某类 Agent Skill 时

03

需要根据任务场景推荐可安装能力包时

04

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

补充不同宿主或平台的使用分布数据

能力 5

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

OpenClaw

70.85%
按下载量换算683

安全审计

VirusTotal

可疑

ClawScan

可疑

Static analysis

通过

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。来源安全扫描存在 warning/failed 结果,不能写成本站确认安全。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills