agentsclimarketplace
Sign in

Tool discovery

Skill wde123sadw/agent-preflight/skills/tool-discovery

Find a small shortlist of task-fit skills, plugins, MCP servers, CLIs, libraries, APIs, or applications for a proven capability gap. Use only after the required capability and current environment are understood. Search trustworthy and current sources, prefer official evidence, compare viable fallbacks, and return candidates for review without installing or enabling them.From its SKILL.md

Install
npx -y skills add wde123sadw/agent-preflight --skill tool-discovery

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

2 things to look at

  • 26 days oldThe repository was created 26 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • 1 stars1 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.

SKILL.md

3.0 KB, 511 tokens by cl100k_base, as published. Nobody here has run it

Tool Discovery

Preconditions

Require all of the following:

  • the task outcome is sufficiently clear;
  • $capability-gap-analysis or equivalent inspection identified a named gap;
  • the gap is required or the user asked for alternatives;
  • read-only discovery is in scope.

If these are missing, return to clarification or capability analysis.

Write the discovery query

Describe:

  • required capability and artifact;
  • platform and runtime;
  • inputs, outputs, and quality bar;
  • local, cloud, privacy, cost, and license constraints;
  • required integration surface: Skill, plugin, MCP, CLI, library, API, or application;
  • verification requirement.

Do not begin with a preferred brand unless the user required it.

Search the right sources

Read source-ladder.md. Search current sources because tools, versions, maintenance, and security change over time.

Prefer this order:

  1. capabilities already exposed by the active agent platform;
  2. official vendor tools, documentation, marketplaces, and package registries;
  3. protocol or ecosystem registries with verifiable publisher identity;
  4. established capability hubs such as CLI-Hub for agent-native professional software;
  5. official GitHub repositories and releases;
  6. community lists, discussions, and search results as leads only.

Use community popularity as a discovery signal, never as primary proof of fitness or safety.

Verify each candidate

Use primary sources to record:

  • canonical name, owner, URL, license, and current version or release;
  • exact capability match and documented limits;
  • supported platform, runtime, formats, and integration;
  • installation and removal method;
  • authentication, permissions, data flow, network use, and secret handling;
  • pricing, quota, local/cloud execution, and vendor dependency;
  • maintenance indicators and known security notices;
  • documented verification, dry-run, health check, or preview path.

Clearly mark unknown fields. Do not fill them from memory.

Limit the shortlist

Return at most three candidates by default:

  • best fit;
  • best low-risk or local alternative;
  • best no-install fallback, when one exists.

Do not produce a marketplace dump. Explain why excluded tools are unnecessary only when the exclusion affects the decision.

Emit candidate records

Use candidate-record.md. Include source links beside supported claims. State whether discovery found no trustworthy fit.

End with a handoff to $tool-review-gate. Do not install, enable, authenticate, request credentials, or mutate configuration.

What ships with it: 3 files

2.9 KB alongside SKILL.md

agents/

references/

Keep looking

Skills are one crate of 326,782. 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.