mcpField
S
smart-loader

smart-loader

github.com/egorfedorovAgent workflows125208 installsUpdated 1d ago
Repository

About smart-loader

smart-loader is a Claude skill that uses task descriptions and past tracking patterns to suggest relevant files or templates. It keeps context loading under user control and limits suggestions to a few files.

Key features

  1. 01Suggests files from past sessions
  2. 02Checks task templates
  3. 03Offers context loading for approval
  4. 04Flags files that may waste tokens
  5. 05Limits suggestions to one per task

Use cases

  1. 01Starting a bug fix or feature task
  2. 02Finding files for a new task
  3. 03Reusing a matching task template

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/egorfedorov/claude-context-optimizer --skill smart-loader

After installing, type/smart-loaderin Claude Code to confirm it loaded.

Details

From the project README

Smart Context Loader

When a user starts a new task or asks to work on something, check if there are historical patterns that suggest which files they'll need.

When to Activate

  • User describes a new task (bug fix, feature, refactor, review)
  • User says "let's work on..." or "I need to fix..."
  • Session just started and user is describing what they want to do
  • User asks "what files do I need for this?"

How to Use

  1. Check for existing templates matching the task type: Glob ~/.claude-context-optimizer/templates/*.json.

  2. Check historical patterns for the current directory:

node ${CLAUDE_PLUGIN_ROOT}/src/tracker.js suggest "$(pwd)"
  1. Based on the task description and historical data:
    • If a matching template exists, mention it: "I see you have a '{name}' template. Want me to load that context?"
    • If historical patterns suggest files, mention: "Based on previous sessions, you usually need these files for this type of task: ..."
    • If patterns show files to avoid, mention: "Tip: [file] has been consistently unused in past sessions — I'll skip it."
    • If nothing matches, proceed normally

Auto-Optimization Tips

When you detect patterns that waste tokens, proactively suggest improvements:

  • If a large file (300+ lines) is being read fully: "I'll use offset/limit to read only the relevant section"
  • If a file was wasted in 2+ past sessions: skip it silently or mention why
  • If the user's task matches a known template: offer to apply it

Important

  • Do NOT automatically read files without user confirmation
  • Only SUGGEST, don't force context loading
  • Keep suggestions brief (top 3-5 files max)
  • If no tracking data exists yet, skip silently — don't mention the optimizer
  • Be helpful, not annoying — max 1 suggestion per task start

Comments

Markdown supported