Agent-Skills.md

Agent Skills: Cursor Context Management

Cursor context management with /summarize command, @ mentions for file inclusion, and @codebase discovery. Use when operating in Cursor IDE or when context exceeds 60%. Activate for context optimization in Cursor sessions.

UncategorizedID: tianjianjiang/smith/smith-ctx-cursor

Repository

tianjianjiang/smith
tianjianjiang

Install this agent skill to your local

pnpm dlx add-skill https://github.com/tianjianjiang/smith/tree/HEAD/smith-ctx-cursor

Skill Files

Browse the full folder contents for smith-ctx-cursor.

Download Skill

Loading file tree…

smith-ctx-cursor/SKILL.md

Skill Metadata

Name
smith-ctx-cursor
Description
Cursor context management with /summarize command, @ mentions for file inclusion, and @codebase discovery. Use when operating in Cursor IDE or when context exceeds 60%. Activate for context optimization in Cursor sessions.

Cursor Context Management

<metadata>
  • Load if: Using Cursor, context >60%
  • Prerequisites: @smith-ctx/SKILL.md
</metadata>

CRITICAL: Context Commands (Primacy Zone)

<required>

Agent prompts for context status, then recommends /summarize at 60-70%.

Automatic summarization uses smaller flash model → vague summaries, lost details. Recommend manual /summarize before auto-trigger.

After any summarization:

  1. Re-state critical context (task, files, decisions)
  2. Use @ mentions to restore files: @auth/middleware.ts @auth/tokens.ts
</required>

/summarize - Manual Summarization

When to use: Before automatic trigger, between task phases

<required>

Recommendation format:

/summarize

After summarizing, re-add critical files:
@auth/middleware.ts @auth/tokens.ts

Verify summary preserved:

  • Task goals
  • File locations
  • Design decisions
  • Next steps
</required> <forbidden>
  • Claiming to execute /summarize directly
  • Relying solely on automatic summarization
  • Continuing without verifying summary quality
</forbidden>

@ Mentions - Force-Include Files

Syntax: @filename or @path/to/file.ext

<required>

Use @ mentions for:

  • After summarization (restore critical files)
  • Before implementation (load files to modify)
  • Large files >600 lines (@codebase truncates at 250)

Patterns:

@auth/middleware.ts @auth/tokens.ts    # Multiple files
@src/auth.ts @tests/auth.test.ts       # File + tests
</required> <forbidden>
  • @codebase for files >600 lines (only loads first 250)
  • @ mention entire directories
  • @ mention files you won't use
</forbidden>

@codebase - Discovery Only

Use for: Finding files, understanding patterns Limitation: Returns summaries, not complete code

<required>

Workflow:

  1. @codebase authentication middleware → Find files
  2. @auth/middleware.ts → Load full file
  3. Implement with complete context
</required>

.cursorrules - Persistent Instructions

Location: .cursorrules in project root

<required>

Put in .cursorrules (project-specific):

  • Technology stack
  • Project structure
  • Reference to @AGENTS.md

Put in skill files (personal/global):

  • Universal principles
  • Language style
  • Git workflow

Example:

# Project Rules
Follow standards from @AGENTS.md

## Stack
- Framework: NestJS
- Database: PostgreSQL
- Testing: Jest
</required> <related>
  • @smith-ctx/SKILL.md - Universal context strategies
</related>

ACTION (Recency Zone)

<required>

Proactive context management:

  1. Prompt for context status
  2. At 60-70%: Recommend /summarize before auto-trigger
  3. After summarization: Remind to re-add files with @ mentions

Agent RECOMMENDS - user executes the command.

File size strategy:

  • <250 lines: @codebase works
  • 250-600 lines: Use @ mention

  • 600 lines: Always @ mention

</required>