agentsclimarketplace

Opportunity tool selection

Skill ArthurZakirov/OpportunityOS/skills/opportunity-tool-selection

Opportunity workflows for applications, registrations, browser handoffs, and supporting documents.

Install
npx -y skills add ArthurZakirov/OpportunityOS --skill opportunity-tool-selection

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Decide which tool mode to use for opportunity workflows: normal web search, browser automation, MCP/API access, local files, or chat-only reasoning.

SKILL.md

3.5 KB, 703 tokens by cl100k_base, as published. Nobody here has run it

Opportunity Tool Selection

Purpose

Use this skill when an opportunity workflow needs a clear decision about whether to use web search, browser automation, MCP/API tools, local files, or ordinary chat reasoning.

This skill is generic. It applies to jobs, apartments, programs, vendor research, marketplaces, registrations, and similar search-then-act workflows.

Tool Modes

Chat-only reasoning

Use chat-only reasoning when:

  • The user asks for a plan, rubric, schema, checklist, or critique.
  • The answer can be produced from already loaded context.
  • The task is conceptual and does not require live facts, authenticated data, or page interaction.

Do not use chat-only reasoning for current listings, prices, availability, deadlines, company facts, or database state.

Normal web search

Use normal web search when:

  • Finding public pages through search engines.
  • Reading static company, listing, funding, event, or documentation pages.
  • Collecting citations for public facts.
  • Comparing sources before deciding where browser automation is needed.

Prefer original source pages over aggregators once a promising result is found.

Browser or agent mode

Use browser automation when:

  • A site is JavaScript-rendered and static fetch/search does not show the real content.
  • The workflow requires clicking filters, expanding cards, paging, scrolling, or screenshots.
  • A form, portal, dashboard, or listing UI must be inspected visually.
  • The task needs a handoff point before CAPTCHA, login, payment, legal acceptance, identity verification, or final submission.

Use browser automation for interaction and extraction. Do not use it to edit a structured system of record when an MCP/API tool is available.

MCP or API tools

Use MCP/API tools when:

  • Reading, creating, or updating records in Notion, Google Sheets, Linear, Slack, Gmail, GitHub, or another connected system.
  • Checking whether a record already exists.
  • Updating statuses, schema, properties, comments, logs, or relations.
  • Querying authenticated data that should not be scraped through the browser UI.

Prefer MCP/API access over browser UI for systems of record. Browser UI is more fragile and easier to confuse with source extraction.

Local files

Use local files when:

  • The workflow depends on private profile data, document manifests, answer policies, logs, resumes, PDFs, or exported trackers.
  • The user wants a diffable artifact.
  • The data should remain private and not be copied into a public repository.

Never publish private local data into reusable skills or public examples.

Decision Order

  1. Identify the system of record.
  2. If a connected MCP/API exists for that system, use it for reads and writes.
  3. Use web search for public discovery.
  4. Use browser automation only for dynamic extraction or form/page interaction.
  5. Use local files for private policies, documents, and durable logs.
  6. Use chat-only reasoning for planning, synthesis, and schema design.

Boundaries

Do not use automation to bypass:

  • CAPTCHA or bot checks
  • MFA, OTP, or login walls
  • Payment
  • Legal acceptance
  • Identity verification
  • Terms or access restrictions

When a task crosses one of these boundaries, stop and hand control back to the user with the exact blocker and next action.

Output

Return:

  • selected tool mode
  • reason
  • fallback mode
  • system of record
  • read/write boundary
  • blocker risks

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.