A streaming, position-tracking tokenizer for 1C event-log files (.lgf/.lgp) that handles S-expression-like records. Designed to never throw on truncated or malformed trailing records from concurrently writing cluster processes. Demonstrates non-obvious data engineering skill: it skips BOM, handles header lines, and can recover from incomplete records without failing. This is reusable in any project that needs to parse semi-structured log files where corruption is expected.

A utility that walks only persisted (non-derived, non-transient) containment references in an EMF model tree, avoiding expensive computed containments that could trigger lazy model loading. Uses a stack-based depth-first iterator to prevent StackOverflow on deep trees. This is a reusable pattern for safely navigating large EMF models without side effects.

A fluent builder for JSON Schema (Draft 2020-12) used to construct tool input and output schemas declaratively. Supports strings, integers, booleans, enums, objects, arrays, required properties, and descriptions. Enforces type safety and provides clear error messages at schema construction time. Extracted into a new project, it could serve any application needing programmatic JSON Schema construction without heavyweight libraries.

A utility that enforces correct EDT BM transaction boundaries: read tasks use IBmModel.executeReadonlyTask, write tasks use IBmModel.execute, and a special executeAndRollback for modeling operations. Prevents a class of bugs where writes accidentally perform read operations or vice versa. The rollback variant is used for operations that inspect but do not modify state. This pattern is reusable in any EMF-based Eclipse plugin where transaction safety is critical.

Implements a tool that serves Markdown guides for other tools, loaded from bundled resources. Guides are advertised via MCP resources/list as 'guide://<toolName>' URIs, and error messages in other tools point callers to this tool for detailed documentation. This pattern decouples tool descriptions (always loaded) from detailed usage instructions (fetched on demand), reducing initial payload size and enabling richer documentation.

A PII redaction utility that generates deterministic, non-reversible pseudonyms using HMAC-SHA256. Supports two modes: countable strings (e.g., names) produce stable pseudonyms keyed on rank; uncountable strings produce a flat mask. This is reusable in any application needing user-data pseudonymization before export or logging, with a configurable secret.

A per-file mutex using ReentrantLock with reference-counted map eviction. Ensures that concurrent comparison operations writing to the same file do not interleave. The eviction policy automatically removes locks from the map when no longer referenced, preventing memory leaks. This is a reusable concurrency pattern for projects that need fine-grained file-level locking without external dependencies.

A builder for YAML frontmatter blocks prepended to Markdown tool responses. Safely escapes strings, numbers, reserved words, and numeric-looking strings (e.g., 'NaN', 'true') to prevent YAML interpretation errors. This is reusable in any application that generates Markdown with structured metadata headers.