version: "1.0.0" name: bosskuai-search-first description: Use this when deciding whether to adopt an existing package, service, MCP, internal utility, or pattern before building custom code or workflow logic.

BosskuAI Search First

Use this skill when the task might already have a good existing solution and the main risk is reinventing the wheel too early.

How this differs from nearby skills

• `bosskuai-codebase-analysis`: reads and maps what exists in the current codebase; this skill also checks external libraries, tools, and platform features before proposing to build.
• `bosskuai-polyglot-engineering`: advises on language/framework-specific implementation; this skill determines whether to build at all.

Mindset

• The best code is code you don't write. Every custom implementation is a maintenance cost.
• Existing solutions beat custom builds when: they are well-maintained, they fit the stack, and the integration cost is lower than the build cost.
• Custom builds beat existing solutions when: no solution fits, the integration complexity exceeds the build complexity, or the use case is a core differentiator.
• Recency matters: a library that was the right choice in 2020 may have better alternatives now.

Search order (apply in this order — stop when you find a good fit)

1. In the current repo: Is there already a utility, module, script, skill, or pattern that covers this need?
2. In the current framework / platform: Does the framework have a built-in feature that handles this? (e.g., framework-native auth, built-in caching, platform SDK)
3. In the current tool surface: Does the agent tool surface, MCP server, or platform offer this capability natively?
4. In the ecosystem (published libraries): Is there a well-maintained, widely-used library for this?
5. As a managed service: Is there a hosted service that removes the operational cost entirely?
6. Build custom: Only if none of the above are sufficient.

Evaluation criteria for external solutions

Before recommending adopt or extend, evaluate:

A single Red signal is a blocker. Multiple Amber signals should trigger a build-vs-adopt discussion.

Build vs adopt decision

Workflow

1. Define the actual capability needed: Be specific about what the code or tool must do — not the implementation, the behavior.
2. State constraints: stack version, license requirements, latency requirements, operational constraints.
3. Search in order: current repo → framework native → tool surface → ecosystem → managed service.
4. Evaluate candidates: score each against the evaluation criteria above.
5. Recommend adopt / extend / build: with clear reasoning.
6. Identify the smallest next step: if adopting, what is the integration path? If building, what is the minimal first version?

Guardrails

• Do not recommend building custom code when a maintained library covers the use case well.
• Do not adopt a library without checking its security posture and license.
• Do not reject an external solution just because it requires reading docs — reading docs beats maintaining custom code.
• Do not recommend "we can just use X" without verifying X is still maintained and fits the current stack.

Output format

Capability needed:
  [specific behavior required, not implementation]
  Constraints: [stack / license / latency / ops]
 
Search results:
  In-repo: [found / not found — details]
  Framework native: [found / not found — details]
  Tool surface: [found / not found — details]
  Ecosystem: [candidates with scores: maintenance / adoption / security / license / fit]
  Managed service: [options considered]
 
Recommendation: [Adopt / Extend / Build custom]
  [chosen option] — [why it fits] — [why alternatives were insufficient]
 
Integration path / build scope:
  [smallest next step to use or build the solution]
 
Tradeoffs:
  [adopt option vs build custom] — [fit / maintenance / complexity / risk / portability / cost]

References

• ../../references/checklists/search-first-checklist.md
• ../../references/checklists/codebase-analysis-checklist.md
• ../../references/checklists/polyglot-engineering-checklist.md