qwen-delegation

Governance & Quality

Hooks adapt context based on the active agent. pensive tracks usage frequency and failure rates to identify unstable workflows. imbue enforces test-driven development via a PreToolUse hook that verifies the existence of test files before allowing implementation writes. For complex tasks, imbue:rigorous-reasoning requires step-by-step logic checks before tool execution.

Security & Session Management

leyline manages OAuth flows for GitHub, GitLab, and AWS with local token caching. conserve implements permission checks, automatically approving safe commands like ls while blocking high-risk operations like rm -rf /. sanctum isolates named sessions for debugging, feature work, and PR reviews. Quality gates in /create-skill and /create-command halt execution if the project has failing tests.

Resilience & Collaboration

/update-plugins recommends updates based on plugin stability metrics and now includes module auditing to detect orphaned or missing skill references. /fix-workflow attempts to repair failed runs by analyzing previous errors. For strategic decisions, /attune:war-room uses a Type 1/2 reversibility framework to route choices to appropriate expert subagents, with war-room-checkpoint enabling embedded escalation at decision points during implementation.

Cross-Session State (Claude Code 2.1.16+)

attune, spec-kit, and sanctum integrate with the native Claude Code Tasks system. Task creation occurs on-demand, and state persists across sessions via CLAUDE_CODE_TASK_LIST_ID. The war-room-checkpoint skill integrates with commands like /do-issue, /pr-review, /fix-pr, and /architecture-review to trigger expert deliberation when high-stakes decisions emerge during workflows. Ambiguity detection prompts for user decisions when task boundaries are unclear. Versions prior to 2.1.16 use file-based state by default.

Commands automate multi-step processes to reduce manual intervention. /prepare-pr validates branch scope, runs configured linters, and verifies a clean git state before drafting a pull request. /full-review audits syntax, logic, and security in a single pass. /speckit-specify requires a written specification phase before generating code. To maintain context, /catchup reads recent git history, and /attune:init detects project types (Python, Node) to scaffold configuration files.

Claude Code Plugin Commands

```bash

# 1. Add the marketplace

/plugin marketplace add athola/claude-night-market

# 2. Install plugins

/plugin install sanctum@claude-night-market # Git workflows

/plugin install pensive@claude-night-market # Code review

/plugin install spec-kit@claude-night-market # Spec-driven dev

# 3. Use

/prepare-pr # Prepare a pull request

/full-review # Run code review

Skill(sanctum:git-workspace-review) # Invoke a skill

```

npx skills

```bash

# Install the entire marketplace

npx skills add athola/claude-night-market

# Or install specific plugins

npx skills add athola/claude-night-market/sanctum # Git workflows

npx skills add athola/claude-night-market/pensive # Code review

npx skills add athola/claude-night-market/conserve # Resource optimization

```

Post-Installation Setup (Claude Code 2.1.16+)

Initialize plugins with Setup hooks:

```bash

# One-time initialization

claude --init

# Weekly maintenance

claude --maintenance

```

> Note: If the Skill tool is unavailable, read skill files at `plugins/{plugin}/skills/{skill-name}/SKILL.