SuperAPI AI Directory

context-analyzer

Coding & development

View repository

Updated 10-15 17:31

About this skill

Context Analyzer

Iron rule: before establishing the minimum sufficient context, do not jump to conclusions, and even more so do not directly modify code.

Workflow

  • Step 1: Scan the structure first ⚠️ REQUIRED
    • 1.1 Check the root directory, key subdirectories, and configuration files.
    • 1.2 Ignore build artifacts and dependency directories, and focus on the real source code and documentation.
  • Step 2: Automatically discover constraint documents ⚠️ REQUIRED
    • 2.1 Prioritize reading root directory documentation, AGENTS.md, README, and contribution and security documents.
    • 2.2 If docs/, .github/, or specific specification files exist, read them additionally by topic.
  • Step 3: Lock onto task-relevant code
    • 3.1 Narrow the scope using file names, symbol names, configuration items, and call chains.
    • 3.2 Identify entry points, core modules, dependency boundaries, and risk areas.
  • Step 4: Output a context summary
    • 4.1 Provide key files, known constraints, unknowns, and next-step recommendations.
    • 4.2 Distinguish between "confirmed facts" and "assumptions to be verified".

Output goals

  • List of task-relevant files.
  • Sources of specifications to follow.
  • Summary of key data flows, dependencies, and call chains.
  • Information still missing and the minimum next step.

Common project entry points

  • package.json: confirm scripts, dependencies, and the actual toolchain.
  • server/api, pages, components, composables: determine the frontend-backend boundaries and call chains.
  • lib/auth.ts, lib/auth-client.ts, middleware/auth.global.ts, server/api/auth/*: check these first when troubleshooting authentication and permission flows.
  • README, AGENTS.md, CONTRIBUTING.md, SECURITY.md: prioritize extracting project rules and collaboration constraints.

Anti-patterns

  • Reading every file in the repository, causing the analysis to lose focus.
  • Assuming implementation details are consistent just because file names look similar.
  • Failing to distinguish facts from guesses and directly giving implementation suggestions.

Pre-delivery checklist

  • Task-relevant files have been located, rather than speaking in generalities.
  • Actually existing specification documents have been scanned first.
  • Conclusions, evidence, and unknowns have been stated separately.
  • The output is sufficient to support the next step of planning or implementation.