Tool Scout

Use this Skill when the user or AI has a clear software-tool need but does not know which tool type can solve it.

Trigger Conditions

Trigger this Skill when any of the following are true:

• The user asks to find a tool, software, open-source project, library, plugin, extension, MCP server, Skill, Agent, SaaS tool, API, automation, CLI, bridge, integration, or alternative.
• The user says or implies: "is there an existing tool?", "do not reinvent the wheel", "find a ready-made solution", "what can solve this?", "which tool should I use?", "有没有现成工具", "找一个工具", "不要重复造轮子".
• The AI is about to build a feature or project and can reasonably infer that an existing external tool may solve a meaningful part of it.
• The task mentions tool types but the correct type is unclear, for example Skill vs MCP vs Agent vs GitHub repo vs extension.
• The task involves integration, bridge, workflow automation, local agent control, file conversion, content capture, browser automation, dev tooling, or AI workflow tooling.

Do not trigger this Skill for:

• A known exact package lookup where the user already named the tool and only wants docs.
• A normal coding task where no external tool choice is relevant.
• A pure knowledge question that is not about software tools.
• A high-stakes legal, medical, or financial recommendation.

Goal

Return a ranked shortlist of existing tools that can solve the user's stated job, not a generic directory dump.

The user does not need to know whether the answer is a Skill, MCP server, Agent, GitHub repo, npm package, VS Code extension, Open VSX extension, SaaS product, or paid tool. Tool Scout should search across types.

Process

1. Restate the job to be done in one sentence.
2. Extract hard constraints:
   • actor or controller
   • target system
   • direction of control or data flow
   • required platform
   • local vs hosted preference
   • free/open-source preference
3. If the request is an add-on, helper, plugin, overlay, automation, workflow, or integration for a named product, run a native feature audit before external competitor search:
   • Check official docs, help center, release notes, and changelog.
   • Inspect or ask the user to inspect the current product UI: selected text actions, right-click/context menu, hover menu, toolbar, side panels, command palette, slash commands, and keyboard shortcuts.
   • Check whether extension/plugin/API/custom-command support already covers the job.
   • Treat native capability as a first-class "existing tool"; if it solves 70%+ of the job, say so before recommending external tools.
   • Do not assume a product lacks a feature just because external search results are stronger.
4. Generate multiple query families:
   • user wording
   • developer wording
   • capability wording
   • product type wording
   • English and Chinese synonyms where relevant
   • source-specific wording, such as in:readme, MCP server, VS Code extension, npm package
5. Run searchers in parallel.
6. Normalize and deduplicate candidates.
7. Apply V0/V1 gates:
   • V0: source exists and is not obviously dead.
   • V1: description, README-level content, metadata, or tool schema shows evidence that it can solve the task.
8. Rank only candidates that pass V0/V1.
9. Explain the ranking in plain language, including whether native product functionality already covers the need.

Command

From this Skill directory:

python3 scripts/tool_scout.py "USER_TOOL_NEED"

Recommended full run:

python3 scripts/tool_scout.py "USER_TOOL_NEED" --limit 10

Machine-readable output:

python3 scripts/tool_scout.py "USER_TOOL_NEED" --json

Limit to explicit sources:

python3 scripts/tool_scout.py "USER_TOOL_NEED" --sources github,npm,vscode,openvsx,mcp,glama,agentskill,web

Searchers

First-version searchers:

• GitHub repository search
• npm registry search
• Official MCP Registry
• Glama MCP directory
• OpenAgentSkill API
• agentskill.sh API
• VS Code Marketplace
• Open VSX Registry
• Brave Search if BRAVE_API_KEY is configured
• Jina search as a best-effort fallback

Optional keys:

• GITHUB_TOKEN
• SMITHERY_API_KEY
• PULSEMCP_API_KEY
• BRAVE_API_KEY

Ranking

Only V0/V1-passing candidates are ranked.

• Goal match: 45%
• Evidence strength: 20%
• Project quality: 20%
• Landing friction: 10%
• Multi-source corroboration: 5%

When presenting results, include:

• tool name
• type
• source URL
• why it matches
• V0/V1 status
• ranking explanation
• known caveats

Output Style

Lead with the best candidates. Keep the first screen short.

Use this shape:

I found N credible candidates. Top choices:

1. Tool name - type - why it is ranked first
2. Tool name - type - why it is ranked second
3. Tool name - type - why it is ranked third

Rejected or lower-confidence results:
- Name - why it did not pass V1 or why it is less relevant