A 400+ line bash script that clones the later half (or quarter) of a Claude Code conversation to reduce token usage. Demonstrates advanced bash craftsmanship: awk-based JSONL processing with UUID remapping, thinking block stripping, session-config line filtering, and cross-platform UUID generation (uuidgen, /proc/sys/kernel/random/uuid, openssl, fallback). Handles Windows paths, pre-generates UUIDs for performance, scales down token counts, and appends reference messages. Includes a companion test suite (test-half-clone.sh) with 19 tests covering edge cases like double tagging, progress messages, and quarter mode. This is the most reusable component in the repo - it could be extracted and adapted for any JSONL-based conversation system.

A two-line status bar script for Claude Code that displays model name, current directory, git branch, uncommitted file count, sync status with origin, and a visual context usage bar. Parses Claude Code's JSON metadata from stdin, calculates context percentage from transcript files, and supports 10 color themes. Handles edge cases gracefully (no git repo, no transcript, first-run baseline estimate). The git sync detection using FETCH_HEAD timestamps and ahead/behind counts is particularly well-implemented. Reusable as a template for any Claude Code status line customization.

A detailed skill definition for fetching Reddit content via its JSON API. Documents a novel technique (the DuckDuckGo-hop unlock) for bypassing Reddit's hard blocks on automated access: navigating Playwright through a DuckDuckGo search result redirect to set a session cookie, then accessing .json URLs directly. Includes fallback methods (Playwright DOM scraping, Claude for Chrome, curl as last resort), rate limiting guidance, and complete JSON parsing examples. The skill is well-researched and practical, solving a real problem that many developers face when trying to use Reddit data programmatically.

A bash script that shallow-mirrors all non-fork repos of specified GitHub accounts for fast local full-text search with ripgrep. Uses xargs with 16 parallel processes for efficient syncing, handles incremental updates (skips if synced within the last hour), and supports --force. Solves a real problem: GitHub's gh search code cannot reliably search private repos, and the web-only code search has indexing lag. The script is self-contained, well-documented, and could be adapted for any multi-repo search workflow.

A comprehensive test suite for the half-clone-conversation.sh script with 19 tests covering normal operation (even/odd message counts, minimum messages), error cases (single message), output validation (null parentUuid, session ID remapping, tag presence), and edge cases (double tagging, thinking block stripping, progress messages with nested user types, session config line stripping, quarter mode with token division). Creates isolated test environments with mock conversations and validates both file contents and history entries. Demonstrates good bash testing practices with setup/cleanup, colored output, and exit code handling.