ifixkart-admin/.agents/skills/context_optimizer/SKILL.md
Adithiyan bbd3a3e959 feat: initial commit — full platform snapshot
- Backend: FastAPI app with all models, services, schemas, routes
- Admin Panel: Next.js dashboard with all modules
  (CRM, POS, repairs, warranties, billing, notifications, etc.)
- Ecommerce: storefront (submodule) + product catalog
- AI Agents: .agents/ skill library for Antigravity IDE
- Docs: architecture, billing, notification gateway markdown
- Submodules: Backend/OpenWA, Ecommerce/File 2/storefront
2026-10-13 12:59:47 +05:30

1.8 KiB

name description
Context Optimizer Enforces token-saving context retrieval, task classification, and targeted documentation/mcp tool loading rules.

Skill: Context Optimizer

1. Purpose

Minimize token usage and prevent context window pollution by loading only target files, documentation, and tools relevant to the active task.

2. When to Use

Mandatory first step of any workspace task, before reading large directories or executing files.

3. Preconditions

  • docs/12-ai/context-index.md and docs/12-ai/current-state.md must exist.

4. Operational Rules

DO NOT:

  • Read the entire repository or list directories recursively by default.
  • Read all documentation files or scan unrelated folders.
  • Load every MCP tool definition or run tool discovery repeatedly.
  • Re-read unchanged files or large log dumps.
  • Repeat previously established architectural findings in subsequent turns.

DO:

  • Consult docs/12-ai/context-index.md first to locate documentation paths.
  • Read docs/12-ai/current-state.md to identify the active feature branch and current work state.
  • Classify the task (UI, Backend, Database, Quality, etc.) and restrict file readings to that scope.
  • Perform targeted query searches (grep_search) rather than reading whole files.
  • Summarize large terminal or log outputs immediately.
  • Persist important intermediate architectural findings to local markdown summaries.

5. Step-by-Step Checklist

  1. Retrieve docs/12-ai/context-index.md and docs/12-ai/current-state.md.
  2. Determine the Task Type (UI, Backend, Database, Mixed).
  3. Search for the specific file names or lines of code using targeted queries.
  4. Load only the specific section of files matching the query.
  5. Defer loading additional MCP tool adapters (like PostHog, Figma, or Perplexity) unless actively required by the task type.