Modal Probe

The problem

AI agents suffer from short-term memory. You spend the first 10 minutes of every session re-explaining your architecture, folder structure, and linting rules only for the agent to hallucinate a pattern that violates your codebase conventions.

What it does

• Scans manifest files to detect exact versions of your runtime, framework, and package manager.
• Maps your directory structure to depth 3 with inferred descriptions for each major path.
• Extracts coding conventions, including indentation, naming schemes, and error-handling patterns, from representative source files.
• Compiles testing protocols and CLI commands for building, linting, and deployment into a single reference.
• Generates a dense PROJECT_CONTEXT.md file with a one-line reload directive for future sessions.

Why this beats prompting it yourself

Manual documentation is tedious and quickly becomes outdated. This skill programmatically scrapes your actual source code and config files to ensure the context is factual rather than aspirational. It condenses complex repo knowledge into a token-efficient format that prevents the agent from "rediscovering" your stack on every chat.

Use cases

• Onboarding an AI agent to a legacy codebase for the first time.
• Updating agent context after a major framework migration or directory restructure.
• Standardizing AI behavior across a team to ensure every agent follows the same forbidden patterns and lint rules.
• Shrinking token usage by providing a dense, structured summary instead of raw file dumps.

Known limitations

The skill scans to a depth of 3 directories. It does not analyze nested monorepos deeply unless invoked within the specific package root.