Shared agents, skills, and the cycle MCP for nyaarium projects. Includes testability assessment, quality review, code analysis, debugging instrumentation, team collaboration agents, and a looping step-runner MCP that drives an agent through named, reusable procedures.
One-shot subagent for use with Agent, Task, or runSubagent. Performs deep code analysis and architectural investigation. Returns structured findings with specific file references and actionable insights.
One-shot subagent for use with Agent, Task, or runSubagent. Audits codebases for SOC 2, GDPR, and CPRA compliance gaps. Evaluates access control, data classification, audit trails, data subject rights, retention, ghost data, and code auditability. Recommends the highest-priority gap to close next.
One-shot subagent for use with Agent, Task, or runSubagent. Identifies missing or incomplete architectural patterns in a codebase. Names paradigms, assesses completion levels, and recommends which pattern to extract or complete first.
One-shot subagent for use with Agent, Task, or runSubagent. Evaluates code quality and identifies improvement opportunities. Recommends one prioritized refactoring based on dependency order and impact.
One-shot subagent for use with Agent, Task, or runSubagent. Executes systematic refactoring through incremental migration and continuous verification. Maintains buildable codebase at every step and returns structured progress reports.
Caveman speech plus anti-repeat rules for terse agent output. Drop filler, name repeated processes instead of restating. Load and use every response.
Coding guidelines for all agents. Always load this skill before you edit a single line of code.
Orchestrates iterative compliance remediation using specialized subagents. Covers SOC 2, GDPR, and CPRA. Manages assessment, prioritization, execution, and verification workflows for closing real compliance gaps and making controls auditable in one place rather than scattered across the codebase.
Debugging guidelines. Always load this skill before any debugging work. If you catch yourself saying "Fixed", "Found the bug", or any completion phrase, load this immediately and correct course.
Identifies missing or incomplete architectural patterns in a codebase and proposes framework-level solutions. Transforms vibe-coded or partially-engineered codebases into well-structured systems built on recognized paradigms.
Admin access level
Server config contains admin-level keywords
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Skills, agents, and the cycle MCP I use across nyaarium projects.
Add plugin to .claude/settings.json
{
"enabledPlugins": {
"nyaaskills@atelier-nyaarium": true
},
"extraKnownMarketplaces": {
"atelier-nyaarium": {
"source": {
"source": "github",
"repo": "atelier-nyaarium/claude-marketplace"
},
"autoUpdate": true
}
}
}
A looping step-runner that drives an agent through a named, reusable procedure (a "cycle")
defined by a markdown file. Progress is tracked in a JSON sidecar next to the plan
(<plan>.cycle.json), so the tools never touch the document the author is editing.
The MCP server is declared in .mcp.json and runs from source via
bun run ${CLAUDE_PLUGIN_ROOT}/src/cycle-mcp.ts. Its only runtime dependencies are
@modelcontextprotocol/sdk and zod; run bun install in the plugin directory once so they
resolve (the same way the switchboard plugin installs its deps).
Definitions live in cycles/<name>.md: front matter steps: [...] and one ## <step> section per
step. Resolution order for the library directory:
NYAASKILLS_CYCLES_DIR - explicit override.CLAUDE_PLUGIN_ROOT - set by Claude Code; cycles/ sits at the plugin root.cycles/ next to src/, for running from source.Ships with plan-refinement and audited-implementation.
To run your own cycles, point NYAASKILLS_CYCLES_DIR at a directory of <name>.md runbooks (same
format as above). This overrides the shipped library rather than merging with it, so copy the
built-in definitions into your directory to keep them. They also make the best starting templates:
cp ~/.claude/plugins/cache/atelier-nyaarium/nyaaskills/*/cycles/*.md your-cycles-dir/
cycleList(...) then reports whatever lives in that directory.
| Tool | Description | Mutable |
|---|---|---|
cycleStartPlan | Initialize a cycle on a plan file; optional phases tracks the plan's phases | Yes |
cycleStartItems | Initialize a cycle over an explicit work queue (items mode) | Yes |
cycleAppendItems | Append items to a running items queue (noDup, mtime-guarded) | Yes |
cycleNext | Advance one step (confirm-then-advance via completed) | Yes |
cycleCheckpoint | End-of-lap decision: done / loop / critical-stop | Yes |
cycleStatus | Report cycle position (step, lap, status) | No |
cycleGoto | Jump to a step, or reopen a finished cycle | Yes |
cycleList | List available cycle definitions | No |
Laps are 1-indexed and open-ended: the agent keeps looping and ends with done when another lap
would add only minimal gains.
Both start tools take includeSteps to run a subset of the cycle's steps: omit it and the tool
returns a confirm bounce listing the full suite for the user to trim, pass ["all"] for everything,
or name the steps to keep. cycleStartPlan also takes phases (labels matching the plan's ##
headers) to track plan phases one per lap, with each phase's section body persisted for crash
recovery.
bun install
bun test # bun:test, runs the cycle suite
bun run lint # biome ci + tsc --noEmit
bun run lint:fix # biome auto-fix
npx claudepluginhub atelier-nyaarium/nyaaskills --plugin nyaaskillsCross-team communication, devcontainer orchestration, and tool proxying for agent teams. Provides crosstalk, dispatch, connector, and evie-bot bridge tools via MCP.
Create and validate production-grade agent skills with 100-point marketplace grading
Skill evaluation and benchmarking - test skill effectiveness with behavioral eval cases, grade results, and track quality improvements
No description provided.
Collection of agent skills
Representation Synthesis workflow for auditing agent skills in Claude Code.
🧪 Master Test Architect — Master Test Architect