Duration: 25 minutes
Skill: Extract project conventions, architecture rules, and working instructions from pairs of existing artifacts (issues + diffs, specs + code, PRs + reviews) using the AI text-triangle principle — where any two related texts can generate the third.
- The text-triangle principle: if two related texts produce a third, any two can reconstruct the missing one
- Extracting meta-knowledge from issue + commit diff pairs
- Building project instructions incrementally, commit by commit
- Reverse engineering coding conventions, architecture decisions, and team workflow
- Creating reusable instruction files from reverse-engineered knowledge
- Manifest-driven workflow for tracking progress across large commit histories
- Research phase: AI-powered classification of commit prefixes to filter signal from noise
- From conventions to SDLC instructions: generating actionable workflow files for AI agents
You can take a project with zero documentation, find completed issues and their commit diffs, and feed them to an AI model to extract project conventions, coding standards, architecture decisions, and workflow rules — producing usable instruction files that capture how the team actually works.
- 070 — Custom Instructions
- 060 — Version Control with Git
- 105 — MCP GitHub Integration (optional, recommended)
- 160 — Bulk File Processing with AI
- VS Code with GitHub Copilot (Agent Mode enabled)
- Access to a Git repository with completed issues and commits
- Basic understanding of diffs and commit history
- Python 3.10+ installed (for bulk extraction script in Step 7)
- You joined a new project with no documentation or stale documentation
- You want to understand how the team actually works (not how they say they work)
- You need to create onboarding instructions but no one has time to write them
- You want to extract coding conventions from real code changes, not guess at them
- You are preparing an instruction file and need grounded evidence of actual practices