jr-limits

Keeps the Mac awake while Claude works, shared by every session on the machine, and warns before the next sleep.

A coloured project badge and its branch, on the right above the prompt, so you always know which window you are typing in. Hidden until you type /jr-badge. A `badge: LABEL #hex` line in CLAUDE.md sets your own.

A toast 30 seconds before the prompt cache goes cold, and on /jr-cache-clock a quiet countdown above the prompt: how long the cache stays warm after Claude's last reply, going green to red, and the last cache hit rate.

Nudges you toward small atomic commits: when uncommitted changes pass a threshold, a band and a toast suggest a split by folder.

The context window in one quiet row above the prompt: a mini bar coloured by category, the fill and the window, turning into an alert when compaction is near. /jr-context hides or shows it, /jr-context details opens the full bar with its legend.

A decision journal per project: /jr-decisions <decision> or a prompt starting with "decision:" records one, /jr-decisions lists them, /jr-decisions export writes docs/decisions.md.

Token usage tracking, model cost attribution per agent, budget alerts, and optimization recommendations

24 skills · 1 command · 1 agent · hooks

OpenTelemetry observability for Claude Code sessions. Captures tool calls, LLM invocations, token usage, and errors as OTel traces. Send telemetry to Dash0 or any OpenTelemetry-compatible backend.

3 skills

The official CCXT MCP server plus the ccxt-mcp usage skill — market data, balances and opt-in trading across 100+ crypto exchanges and prediction markets, with keys kept local.

1 MCP server

Agent skills for Qdrant vector search: scaling, performance optimization, search quality, monitoring, deployment, model migration, version upgrades, and SDK usage

33 skills

Persistent memory for AI coding agents -- captures tool usage, compresses via LLM, injects context into future sessions

17 skills · 1 MCP server · hooks

Ultra-compressed communication mode. Cuts token usage 20-50% (75% upper bound) by dropping filler, articles, pleasantries, and hedging while keeping full technical accuracy. Derived from Matt Pocock's MIT-licensed caveman with: (1) 3 stdlib Python tools (deterministic compressor, token-savings estimator with $/Mtok cost extrapolation, lint that detects banned vocab with code-block + exception-zone whitelisting), (2) 3 references citing 7-8 sources (compression principles, when caveman backfires, companion tooling), (3) cs-caveman-mode persona agent + /cs-caveman slash command. Matt's persistence rules + auto-clarity exception preserved verbatim per MIT.

1 skill · 1 command · 1 agent