Skip to content

skill-examples.md

Latest commit

 

History

History
366 lines (271 loc) · 8.97 KB

skill-examples.md

File metadata and controls

366 lines (271 loc) · 8.97 KB
Raw

CodePrism Skill - Examples

This document provides practical examples of using the CodePrism skill with AI agents.

Example 1: Understanding a Complex Function

Scenario: You want to understand a complex payment processing function.

Prompt:

Use codeprism to understand the process_payment function in payments/processor.py. Explain what it does, who calls it, what it calls, and any security considerations.

Expected Agent Response:

  1. Calls search_symbol to locate the function
  2. Calls get_context(payments/processor.py, process_payment, depth=2)
  3. Provides:
    • Function signature and purpose
    • List of direct callers (which parts of the app use it)
    • List of direct callees (what dependencies it has)
    • Types and parameters
    • Any security-related calls

Tokens used: ~250 tokens (vs ~2,000 tokens for reading the file)

Example 2: Refactoring Impact Analysis

Scenario: You want to rename a core utility function.

Prompt:

I want to rename the format_date function in utils/helpers.py. Use codeprism to check the impact before I start. Tell me what will break.

Expected Agent Response:

  1. Calls get_impact(utils/helpers.py, format_date)
  2. Reports:
    • Severity (HIGH/MEDIUM/LOW)
    • Number of direct dependents
    • List of affected files
    • Affected test files
    • Transitive dependents (files that depend on dependents)

Tokens used: ~300 tokens (vs ~5,000 tokens for manual analysis)

Example 3: Security Check Before Commit

Scenario: You wrote a new authentication function and want to check for security issues.

Prompt:

I just wrote a new authentication function in auth/login.py. Use codeprism to scan it for security issues before I commit.

Expected Agent Response:

  1. Calls scan_file(auth/login.py) or scan_diff with changes
  2. Reports:
    • Status (PASS/WARN/BLOCK)
    • Any secrets detected
    • Any injection vulnerabilities
    • Any weak cryptography
    • Any unsafe dependencies

If BLOCK:

  • Agent refuses to commit
  • Explains the specific issues
  • Suggests fixes

If PASS:

  • Agent approves commit
  • Reports any warnings

Example 4: Finding All Authentication Functions

Scenario: You need to audit all authentication-related code.

Prompt:

Find all functions in this codebase that handle authentication. Use codeprism to search for them.

Expected Agent Response:

  1. Calls search_symbol("auth", kind="function")
  2. Presents:
    • List of matching functions
    • Their locations
    • Brief descriptions
    • Ranking by relevance

Tokens used: ~200 tokens (vs grepping through all files)

Example 5: Understanding Module Dependencies

Scenario: You want to know what a module depends on.

Prompt:

Use codeprism to analyze the dependencies of the auth module. What does it import and what depends on it?

Expected Agent Response:

  1. Calls get_dependencies(auth/login.py) for imports
  2. Calls get_dependents(auth/login.py) for dependents
  3. Reports:
    • Internal dependencies (other project files)
    • External dependencies (libraries)
    • Files that depend on this module
    • Dependency graph overview

Example 6: Data Flow Analysis

Scenario: You want to trace where user input flows.

Prompt:

Trace where data from the handle_request function flows. Use codeprism to find all the sinks.

Expected Agent Response:

  1. Calls get_data_flow(api/handlers.py, handle_request)
  2. Reports:
    • Data sources
    • Intermediate processing steps
    • Final sinks (database, API calls, file writes)
    • Security-critical paths

Example 7: Module-Level Understanding

Scenario: You're exploring a new codebase and want to understand a module quickly.

Prompt:

Use codeprism to give me a high-level overview of the core module. What does it do and what are its main components?

Expected Agent Response:

  1. Calls get_module_summary(codeprism/core/__init__.py)
  2. Reports:
    • Module purpose (narrative summary)
    • Public API (main functions/classes)
    • Key classes
    • Dependencies
    • Complexity score

Tokens used: ~150 tokens (vs ~1,000 tokens for reading the file)

Example 8: Team Refactoring Coordination

Scenario: Your team is planning a large refactoring and needs to coordinate.

Prompt:

We're planning to refactor the user model. Use codeprism to check the impact and tell me which teams and files will be affected.

Expected Agent Response:

  1. Calls get_impact(models/user.py, User)
  2. Reports:
    • Severity assessment
    • List of affected files grouped by module
    • Test files that need updates
    • Suggested coordination points

Example 9: Finding Similar Functions

Scenario: You want to find all validation functions.

Prompt:

Find all functions that validate input. Use codeprism to search for validation-related functions.

Expected Agent Response:

  1. Calls search_symbol("validate", kind="function")
  2. Presents:
    • All validation functions
    • Their locations
    • Brief descriptions
    • Can group by validation type

Example 10: Session-Aware Long Task

Scenario: A long multi-step task where the agent should remember what it has read.

Prompt:

I need to refactor the payment system. Use codeprism throughout:
1. First, check the impact of changing process_payment
2. Then understand who calls it
3. Then understand what it calls
4. Then make the changes
5. Finally, scan for security issues before writing

Use session tracking to avoid re-reading the same context.

Expected Agent Response:

  1. Calls get_context and record_read for each query
  2. Uses get_session_context to check what's already been read
  3. Avoids re-fetching the same information
  4. Calls scan_diff before writes
  5. Calls record_write to track changes

Benefits:

  • Token-efficient (no repeated reads)
  • Faster (no re-querying)
  • Consistent state tracking

Example 11: Cross-Language Analysis

Scenario: Your project has Python and TypeScript code.

Prompt:

Find all functions that call the Python API from the TypeScript frontend. Use codeprism to trace the cross-language calls.

Expected Agent Response:

  1. Searches for API functions in Python
  2. Searches for callers in TypeScript
  3. Reports:
    • Cross-language call graph
    • Entry points
    • Integration points

Example 12: Performance Investigation

Scenario: A function is slow and you want to understand why.

Prompt:

The export_data function is slow. Use codeprism to analyze what it calls and identify potential bottlenecks.

Expected Agent Response:

  1. Calls get_context(export_data, depth=2)
  2. Analyzes callees for:
    • Database operations
    • File I/O
    • Network calls
    • Complex computations
  3. Reports:
    • Likely bottlenecks
    • Suggested optimization points

Tips for Effective Prompts

Be Specific About What You Want

Good:

Use codeprism to find who calls the process_payment function in the payments module.

Bad:

Use codeprism to tell me about process_payment.

Ask for Analysis, Not Just Data

Good:

Use codeprism to analyze the impact of changing this function and tell me what tests I need to update.

Bad:

Use codeprism to get the impact of this function.

Use CodePrism for Structural Questions

Good for:

  • Understanding relationships (callers, callees)
  • Impact analysis
  • Dependency analysis
  • Security scanning
  • Symbol search

Not for:

  • Exact implementation details (read the file after getting context)
  • Line-by-line code review
  • Business logic verification

Combine with File Reading When Needed

Good:

Use codeprism to get the context for this function, then read the file if you need implementation details.

Common Patterns

Pattern 1: Quick Overview

Use codeprism to give me a quick overview of [module/file].

Pattern 2: Deep Dive

Use codeprism to get context for [function] with depth=2, then read the file for implementation details.

Pattern 3: Before Refactoring

Use codeprism to check the impact of changing [symbol] before I start refactoring.

Pattern 4: Security Check

Use codeprism to scan [file] for security issues before I commit.

Pattern 5: Find Relationships

Use codeprism to find all functions that [condition] in this codebase.

Token Savings Summary

Task Without CodePrism With CodePrism Savings
Understand function ~2,000 tokens ~250 tokens 87.5%
Find callers ~3,000 tokens ~300 tokens 90%
Impact analysis ~5,000 tokens ~350 tokens 93%
Security scan ~1,000 tokens ~150 tokens 85%
Module summary ~1,000 tokens ~150 tokens 85%

Average: 88% token reduction

Next Steps