NYC
skills/letta-ai/skills/merge-diff-arc-agi-task

merge-diff-arc-agi-task

SKILL.md

Merge-Diff ARC-AGI Task

This skill provides guidance for tasks that combine git operations with ARC-AGI style pattern recognition problems. These tasks typically involve fetching git bundles, merging branches with conflicting implementations, analyzing example input-output pairs to discover transformation patterns, and implementing correct algorithmic solutions.

When to Use This Skill

Apply this skill when the task involves:

  • Fetching and merging git bundles containing different code implementations
  • Analyzing grid-based input-output examples to discover transformation rules
  • Implementing algorithms that transform 2D grids based on discovered patterns
  • Resolving merge conflicts between competing implementations

Recommended Workflow

Phase 1: Environment Setup

Before starting the main workflow, complete all environment configuration:

  1. Configure git identity - Set user.name and user.email before any merge operations to avoid "Committer identity unknown" errors
  2. Verify Python availability - Check whether python or python3 is the correct command
  3. Initialize git repository if not already done
  4. Fetch all required bundles before analyzing any code

Phase 2: Analysis Before Implementation

Analyze thoroughly before writing any code:

  1. Test existing implementations first - Before merging or writing new code, test what each branch's implementation actually does
  2. Parse and examine all examples - Read the examples.json (or equivalent) file once and analyze all input-output pairs systematically
  3. Document the pattern formally - Write out the mathematical relationship explicitly (e.g., "For each cell (i,j), output[i][j] = sequence[(i+j) mod n]")
  4. Verify the pattern against ALL examples - Confirm the discovered pattern holds for every example, not just a subset

Phase 3: Pattern Recognition Strategies

When analyzing grid transformation patterns:

  1. Identify unique values - Extract non-zero or distinct values from the input
  2. Look for positional relationships - Check if output depends on row index, column index, or combinations like (i+j), (i-j), etc.
  3. Check for tiling or repetition - Determine if the output is a tiled or repeated version of a smaller pattern
  4. Consider modular arithmetic - Many patterns use modulo operations based on grid dimensions or sequence lengths
  5. Test edge cases mentally - What happens with all zeros? Single values? Non-square grids?

Phase 4: Implementation

  1. Create a reusable test harness early - Write a single comprehensive test function that can be called repeatedly instead of writing test code multiple times
  2. Implement with clear variable names - Make the algorithm's intent obvious
  3. Add input validation at boundaries - Handle empty grids, all-zero inputs, and edge cases
  4. Handle potential collisions - If using dictionary or position-based mapping, consider what happens if two values map to the same key

Phase 5: Merge and Verify

  1. Only start the merge after understanding the required implementation - Avoid starting a merge prematurely and needing to abort
  2. Use a complete verification script - Test with all provided examples and check expected outputs
  3. Run the official test suite - Ensure all tests pass before considering the task complete

Common Pitfalls to Avoid

Git-Related Mistakes

  • Starting a merge before configuring git user identity
  • Beginning the merge before understanding what implementation is needed
  • Forgetting to stage and commit after resolving conflicts

Python-Related Mistakes

  • Using python when python3 is required (or vice versa)
  • Not testing the existing implementations before writing new code

Pattern Analysis Mistakes

  • Stream-of-consciousness reasoning instead of formal pattern documentation
  • Spot-checking only a few examples instead of verifying all
  • Missing edge cases like empty sequences or all-zero inputs
  • Not considering what happens when multiple values map to the same position

Code Organization Mistakes

  • Writing test code multiple times in different forms instead of creating a reusable function
  • Reading data files multiple times instead of parsing once
  • Incomplete or truncated verification scripts

Verification Checklist

Before marking the task complete, verify:

  • Git user identity configured
  • All bundles fetched successfully
  • Pattern documented with explicit mathematical formula
  • Pattern verified against ALL examples (not just some)
  • Edge cases handled (empty input, all zeros, etc.)
  • All tests pass
  • Merge committed successfully
Weekly Installs
18
Repository
letta-ai/skills
First Seen
Jan 24, 2026
Installed on
claude-code13
codex12
gemini-cli12
opencode12
antigravity11
cursor9