adam-s

adam-s/agent-tuning

Settings file Claude Code

Using recursion to acheive predictable agent output.

adam-s/testing-claude-agent

Instructions file

Claude Code instructions for adam-s/testing-claude-agent, covering claude-agent-eval, what this is, quick commands and structure.

adam-s/testing-claude-agent

Agent

You are a builder agent. Your job is to write working code that meets the requirements.

adam-s/rubiks-cube-mcp

Skill Claude CodeCodex

Run a bounded live audit of the harness against a real LLM provider over multiple seeds — captures solving behavior, costs, and divergence from the cube oracle. Bounded by wall time AND dollar budget — never unbounded. Use when the user asks to "audit", "live test", or "check whether the harness still solves with a…

adam-s/rubiks-cube-mcp

Skill Claude CodeCodex

Reference and recipes for the dashboard's HTTP control plane on :8000 — workspaces, prompts, agents, runs, evolution loops, websockets. Use when the user wants to drive the dashboard from curl / a script / Python without going through the UI; or asks how to launch a run, create a prompt, start an evolution loop, or…

adam-s/rubiks-cube-mcp

Skill Claude CodeCodex

Launch a Jony-Ive-voiced product designer (Opus) to critique the dashboard UI/UX from inside the project's existing aesthetic constraint. Reads the snapshot sweep + the components and returns a single essay, not a checklist. Use when the user asks for "design critique", "Jony Ive review", "UI review", or says the…

adam-s/rubiks-cube-mcp

Skill Claude CodeCodex

Iteratively diagnose and fix a UI bug by driving Playwright against the running dashboard, fusing Python and browser logs into one timeline, and patching until the reproducer runs clean. Use when the user describes a UI problem ("dialog won't open", "the cube flickers when scrubbing", "tag input loses focus"…

adam-s/rubiks-cube-mcp

Skill Claude CodeCodex

Drive the running dashboard from a chat LLM via the mcpcontrol MCP server — create prompts, launch runs, run evolution loops, and "play piano" on the open dashboard tab (focus a run, switch view, scrub the timeline) all from native MCP tool calls. Use when the user asks to install mcpcontrol in a chat client (Claude…

adam-s/rubiks-cube-mcp

Skill Claude CodeCodex

Launch an adversarial mutation-testing agent (Opus) that injects targeted regressions into production code, runs the test suite, and reports which mutations SURVIVED — surviving mutations are direct evidence of test-coverage gaps. Use when the user says "trickster", "mutation test", "break the code", "can the tests…

adam-s/rubiks-cube-mcp

Skill Claude CodeCodex

Launch an adversarial bug-hunting code reviewer (Opus) to find real bugs, correctness issues, concurrency hazards, and resource-abuse risks in this project — not style nits. Use when the user asks for a "red team", "bug hunt", "adversarial review", or at checkpoints during long coding tasks.

adam-s/rubiks-cube-mcp

Skill Claude CodeCodex

Launch an adversarial reviewer (Opus) to find bugs IN THE TEST SUITE — tautological assertions, weak oracles, mock/fake lies, fixture drift, coverage gaps, in-process-vs-subprocess divergence. Use when the user says "red team the tests", "audit the tests", "are our tests any good", or after a large batch of tests.

adam-s/rubiks-cube-mcp

Skill Claude CodeCodex

Walk one cold-state user journey end-to-end against the running dashboard, find every functional bug AND intuitiveness gap, fix the functional ones with destructive freedom, and pin a regression so each fix is durable. Use when the user says "user journey", "walk through X", "qa pass on Y", "is the dashboard ready for…

adam-s/rubiks-cube-mcp

MCP server Claude CodeCodexCursor

MCP server "mcp-control" as configured in adam-s/rubiks-cube-mcp. Runs locally from the mcp_control Python package. Needs 1 environment variable to run.