Skip to content

aiskillstore/marketplace / skills/2389-research/building-multiagent-systems

building-multiagent-systems

Verified

This skill should be used when designing or implementing systems with multiple AI agents that coordinate to accomplish tasks. Triggers on "multi-agent", "orchestrator", "sub-agent", "coordination", "delegation", "parallel agents", "sequential pipeline", "fan-out", "map-reduce", "spawn agents", "agent hierarchy".

skill-report.json

View on GitHub
{
  "schema_version": "2.0",
  "meta": {
    "generated_at": "2026-07-04T17:37:33.518Z",
    "slug": "2389-research-building-multiagent-systems",
    "source_url": "https://github.com/2389-research/claude-plugins/tree/main/building-multiagent-systems/skills",
    "source_ref": "main",
    "model": "codex",
    "analysis_version": "3.0.0",
    "source_type": "community",
    "content_hash": "ca0ba2fc4876e3629a29642a762c28ea61d0cb19faebe8eb709b980bb4cb2dc5",
    "tree_hash": "0fee8e736a6f73bc7aed54117c97c8110ac9c900a17ce3c44cbcace7e19e7ea5",
    "provenance": {
      "model_requested": "gpt-5.5:high",
      "model_effective": "codex:gpt-5.5:high",
      "fallback_chain": [
        {
          "agent": "codex",
          "model": "gpt-5.5",
          "effort": "high",
          "outcome": "success"
        }
      ],
      "analysis_version": "3.0.0"
    }
  },
  "skill": {
    "name": "building-multiagent-systems",
    "description": "Guides architecture and implementation of coordinated multi-agent systems with patterns for delegation, tool coordination, lifecycle management, and production hardening.",
    "summary": "Design multi-agent systems with coordination patterns, safety boundaries, and production controls.",
    "icon": "📦",
    "version": "1.0.0",
    "author": "2389-research",
    "license": "MIT",
    "tags": [
      "multi-agent systems",
      "agent orchestration",
      "tool coordination",
      "production hardening",
      "system architecture"
    ],
    "supported_tools": [
      "claude",
      "codex",
      "claude-code"
    ],
    "risk_factors": [
      "external_commands"
    ],
    "category": "coding"
  },
  "security_audit": {
    "risk_level": "medium",
    "is_blocked": false,
    "safe_to_publish": true,
    "analysis_status": "ok",
    "summary": "All 31 static findings are false positives caused by Markdown formatting, TypeScript template literals, ordinary variable names, and documentation keywords. No prompt injection or credential exfiltration intent was found. One medium semantic risk remains because the self-modification example can encourage destructive repository rollback without explicit user approval.",
    "remediation": [
      {
        "issue": "Self-modification guidance includes a hard-reset rollback example.",
        "suggestion": "Replace hard reset with a user-approved rollback plan, preserve uncommitted work, and require explicit confirmation before destructive git operations.",
        "severity": "medium"
      },
      {
        "issue": "Documentation discusses YOLO orchestration and self-modifying agents.",
        "suggestion": "State that production systems should default to safety-first or hybrid approval for shell, write, and repository operations.",
        "severity": "low"
      }
    ],
    "risk_factor_evidence": [
      {
        "factor": "external_commands",
        "evidence": [
          {
            "file": "references/four-layer-architecture.md",
            "line_start": 41,
            "line_end": 56
          },
          {
            "file": "references/maker-pattern.md",
            "line_start": 33,
            "line_end": 72
          },
          {
            "file": "references/maker-pattern.md",
            "line_start": 72,
            "line_end": 83
          },
          {
            "file": "references/production-hardening.md",
            "line_start": 166,
            "line_end": 191
          },
          {
            "file": "references/tool-coordination.md",
            "line_start": 155,
            "line_end": 170
          },
          {
            "file": "SKILL.md",
            "line_start": 38,
            "line_end": 38
          },
          {
            "file": "SKILL.md",
            "line_start": 45,
            "line_end": 45
          },
          {
            "file": "SKILL.md",
            "line_start": 63,
            "line_end": 63
          },
          {
            "file": "SKILL.md",
            "line_start": 88,
            "line_end": 88
          },
          {
            "file": "SKILL.md",
            "line_start": 101,
            "line_end": 101
          },
          {
            "file": "SKILL.md",
            "line_start": 107,
            "line_end": 113
          },
          {
            "file": "SKILL.md",
            "line_start": 113,
            "line_end": 127
          },
          {
            "file": "SKILL.md",
            "line_start": 127,
            "line_end": 181
          },
          {
            "file": "SKILL.md",
            "line_start": 181,
            "line_end": 182
          },
          {
            "file": "SKILL.md",
            "line_start": 182,
            "line_end": 183
          },
          {
            "file": "SKILL.md",
            "line_start": 183,
            "line_end": 184
          },
          {
            "file": "SKILL.md",
            "line_start": 184,
            "line_end": 185
          }
        ]
      }
    ],
    "static_findings": [
      {
        "id": "external_commands:references/four-layer-architecture.md:41:ruby-shell-backtick-execution",
        "category": "external_commands",
        "severity": "medium",
        "pattern": "Ruby/shell backtick execution",
        "file": "references/four-layer-architecture.md",
        "line_start": 41,
        "line_end": 56,
        "snippet": "const analysis = await llm.generate(`Analyze this code: ${code}`);"
      },
      {
        "id": "external_commands:references/maker-pattern.md:33:ruby-shell-backtick-execution",
        "category": "external_commands",
        "severity": "medium",
        "pattern": "Ruby/shell backtick execution",
        "file": "references/maker-pattern.md",
        "line_start": 33,
        "line_end": 72,
        "snippet": "name: `parser-vote-${i}`"
      },
      {
        "id": "external_commands:references/maker-pattern.md:72:ruby-shell-backtick-execution",
        "category": "external_commands",
        "severity": "medium",
        "pattern": "Ruby/shell backtick execution",
        "file": "references/maker-pattern.md",
        "line_start": 72,
        "line_end": 83,
        "snippet": "subtask.description + `\\nPrevious failed: ${result}`,"
      },
      {
        "id": "blocker:references/maker-pattern.md:68:system-reconnaissance",
        "category": "blocker",
        "severity": "low",
        "pattern": "System reconnaissance",
        "file": "references/maker-pattern.md",
        "line_start": 68,
        "line_end": 68,
        "snippet": "const isValid = await validateResult(result, subtask.constraints);"
      },
      {
        "id": "blocker:references/maker-pattern.md:90:network-reconnaissance",
        "category": "blocker",
        "severity": "low",
        "pattern": "Network reconnaissance",
        "file": "references/maker-pattern.md",
        "line_start": 90,
        "line_end": 91,
        "snippet": "- Cost is secondary to correctness"
      },
      {
        "id": "external_commands:references/production-hardening.md:166:ruby-shell-backtick-execution",
        "category": "external_commands",
        "severity": "medium",
        "pattern": "Ruby/shell backtick execution",
        "file": "references/production-hardening.md",
        "line_start": 166,
        "line_end": 191,
        "snippet": "await gitBranch(`fix-bug-${Date.now()}`);"
      },
      {
        "id": "sensitive:references/production-hardening.md:113:sqlite-database-file",
        "category": "sensitive",
        "severity": "medium",
        "pattern": "SQLite database file",
        "file": "references/production-hardening.md",
        "line_start": 113,
        "line_end": 113,
        "snippet": "await this.db.insert({ ...task, id, status: 'pending', claimedBy: null });"
      },
      {
        "id": "sensitive:references/production-hardening.md:119:sqlite-database-file",
        "category": "sensitive",
        "severity": "medium",
        "pattern": "SQLite database file",
        "file": "references/production-hardening.md",
        "line_start": 119,
        "line_end": 119,
        "snippet": "const task = await this.db.findOneAndUpdate("
      },
      {
        "id": "sensitive:references/production-hardening.md:127:sqlite-database-file",
        "category": "sensitive",
        "severity": "medium",
        "pattern": "SQLite database file",
        "file": "references/production-hardening.md",
        "line_start": 127,
        "line_end": 127,
        "snippet": "await this.db.update("
      },
      {
        "id": "sensitive:references/production-hardening.md:135:sqlite-database-file",
        "category": "sensitive",
        "severity": "medium",
        "pattern": "SQLite database file",
        "file": "references/production-hardening.md",
        "line_start": 135,
        "line_end": 135,
        "snippet": "await this.db.updateMany("
      },
      {
        "id": "blocker:references/production-hardening.md:112:system-reconnaissance",
        "category": "blocker",
        "severity": "low",
        "pattern": "System reconnaissance",
        "file": "references/production-hardening.md",
        "line_start": 112,
        "line_end": 112,
        "snippet": "const id = generateId();"
      },
      {
        "id": "blocker:references/production-hardening.md:279:system-reconnaissance",
        "category": "blocker",
        "severity": "low",
        "pattern": "System reconnaissance",
        "file": "references/production-hardening.md",
        "line_start": 279,
        "line_end": 279,
        "snippet": "release(): void {"
      },
      {
        "id": "external_commands:references/tool-coordination.md:155:ruby-shell-backtick-execution",
        "category": "external_commands",
        "severity": "medium",
        "pattern": "Ruby/shell backtick execution",
        "file": "references/tool-coordination.md",
        "line_start": 155,
        "line_end": 170,
        "snippet": "`Review ${filePath} focusing on: ${focusAreas.join(', ')}`"
      },
      {
        "id": "sensitive:references/tool-coordination.md:121:certificate-key-files",
        "category": "sensitive",
        "severity": "high",
        "pattern": "Certificate/key files",
        "file": "references/tool-coordination.md",
        "line_start": 121,
        "line_end": 121,
        "snippet": "for (const key of this.cache.keys()) {"
      },
      {
        "id": "blocker:references/tool-coordination.md:51:system-reconnaissance",
        "category": "blocker",
        "severity": "low",
        "pattern": "System reconnaissance",
        "file": "references/tool-coordination.md",
        "line_start": 51,
        "line_end": 51,
        "snippet": "release(resource: string, agentId: string): void {"
      },
      {
        "id": "blocker:references/tool-coordination.md:85:system-reconnaissance",
        "category": "blocker",
        "severity": "low",
        "pattern": "System reconnaissance",
        "file": "references/tool-coordination.md",
        "line_start": 85,
        "line_end": 85,
        "snippet": "private refill(): void {"
      },
      {
        "id": "blocker:references/tool-coordination.md:120:system-reconnaissance",
        "category": "blocker",
        "severity": "low",
        "pattern": "System reconnaissance",
        "file": "references/tool-coordination.md",
        "line_start": 120,
        "line_end": 120,
        "snippet": "invalidate(pattern: string): void {"
      },
      {
        "id": "external_commands:SKILL.md:38:ruby-shell-backtick-execution",
        "category": "external_commands",
        "severity": "medium",
        "pattern": "Ruby/shell backtick execution",
        "file": "SKILL.md",
        "line_start": 38,
        "line_end": 38,
        "snippet": "See `references/four-layer-architecture.md` for detailed implementation with code examples."
      },
      {
        "id": "external_commands:SKILL.md:45:ruby-shell-backtick-execution",
        "category": "external_commands",
        "severity": "medium",
        "pattern": "Ruby/shell backtick execution",
        "file": "SKILL.md",
        "line_start": 45,
        "line_end": 45,
        "snippet": "| **Hierarchical IDs** | Encode delegation hierarchy (e.g., `session.1.2`) for cost aggregation |"
      },
      {
        "id": "external_commands:SKILL.md:63:ruby-shell-backtick-execution",
        "category": "external_commands",
        "severity": "medium",
        "pattern": "Ruby/shell backtick execution",
        "file": "SKILL.md",
        "line_start": 63,
        "line_end": 63,
        "snippet": "See `references/coordination-patterns.md` for detailed implementations."
      },
      {
        "id": "external_commands:SKILL.md:88:ruby-shell-backtick-execution",
        "category": "external_commands",
        "severity": "medium",
        "pattern": "Ruby/shell backtick execution",
        "file": "SKILL.md",
        "line_start": 88,
        "line_end": 88,
        "snippet": "See `references/maker-pattern.md` for full implementation with medical diagnosis example."
      },
      {
        "id": "external_commands:SKILL.md:101:ruby-shell-backtick-execution",
        "category": "external_commands",
        "severity": "medium",
        "pattern": "Ruby/shell backtick execution",
        "file": "SKILL.md",
        "line_start": 101,
        "line_end": 101,
        "snippet": "See `references/tool-coordination.md` for implementations."
      },
      {
        "id": "external_commands:SKILL.md:107:ruby-shell-backtick-execution",
        "category": "external_commands",
        "severity": "medium",
        "pattern": "Ruby/shell backtick execution",
        "file": "SKILL.md",
        "line_start": 107,
        "line_end": 113,
        "snippet": "```"
      },
      {
        "id": "external_commands:SKILL.md:113:ruby-shell-backtick-execution",
        "category": "external_commands",
        "severity": "medium",
        "pattern": "Ruby/shell backtick execution",
        "file": "SKILL.md",
        "line_start": 113,
        "line_end": 127,
        "snippet": "```"
      },
      {
        "id": "external_commands:SKILL.md:127:ruby-shell-backtick-execution",
        "category": "external_commands",
        "severity": "medium",
        "pattern": "Ruby/shell backtick execution",
        "file": "SKILL.md",
        "line_start": 127,
        "line_end": 181,
        "snippet": "See `references/production-hardening.md` for detailed implementations."
      },
      {
        "id": "external_commands:SKILL.md:181:ruby-shell-backtick-execution",
        "category": "external_commands",
        "severity": "medium",
        "pattern": "Ruby/shell backtick execution",
        "file": "SKILL.md",
        "line_start": 181,
        "line_end": 182,
        "snippet": "| `references/four-layer-architecture.md` | Four-layer stack, deterministic boundary, schema-first t"
      },
      {
        "id": "external_commands:SKILL.md:182:ruby-shell-backtick-execution",
        "category": "external_commands",
        "severity": "medium",
        "pattern": "Ruby/shell backtick execution",
        "file": "SKILL.md",
        "line_start": 182,
        "line_end": 183,
        "snippet": "| `references/coordination-patterns.md` | Seven coordination patterns with code |"
      },
      {
        "id": "external_commands:SKILL.md:183:ruby-shell-backtick-execution",
        "category": "external_commands",
        "severity": "medium",
        "pattern": "Ruby/shell backtick execution",
        "file": "SKILL.md",
        "line_start": 183,
        "line_end": 184,
        "snippet": "| `references/maker-pattern.md` | MAKER implementation, voting, medical diagnosis example |"
      },
      {
        "id": "external_commands:SKILL.md:184:ruby-shell-backtick-execution",
        "category": "external_commands",
        "severity": "medium",
        "pattern": "Ruby/shell backtick execution",
        "file": "SKILL.md",
        "line_start": 184,
        "line_end": 185,
        "snippet": "| `references/tool-coordination.md` | Permission inheritance, locking, rate limiting, caching |"
      },
      {
        "id": "blocker:SKILL.md:46:system-reconnaissance",
        "category": "blocker",
        "severity": "low",
        "pattern": "System reconnaissance",
        "file": "SKILL.md",
        "line_start": 46,
        "line_end": 46,
        "snippet": "| **Agent State Machines** | Explicit states (idle → thinking → tool_execution → stopped) with inval"
      },
      {
        "id": "blocker:SKILL.md:149:system-reconnaissance",
        "category": "blocker",
        "severity": "low",
        "pattern": "System reconnaissance",
        "file": "SKILL.md",
        "line_start": 149,
        "line_end": 150,
        "snippet": "6. **Choose orchestration model** - YOLO, Safety-First, or Hybrid"
      }
    ],
    "finding_verdicts": [
      {
        "id": "external_commands:references/four-layer-architecture.md:41:ruby-shell-backtick-execution",
        "verdict": "false_positive",
        "confidence": 0.96,
        "reason": "This is a TypeScript template literal inside a Markdown code example about LLM placement. It is not Ruby shell backtick execution and no package code executes it."
      },
      {
        "id": "external_commands:references/maker-pattern.md:33:ruby-shell-backtick-execution",
        "verdict": "false_positive",
        "confidence": 0.97,
        "reason": "The backticks create a TypeScript string for a sample agent name in documentation. There is no shell command, subprocess call, or executable skill script."
      },
      {
        "id": "external_commands:references/maker-pattern.md:72:ruby-shell-backtick-execution",
        "verdict": "false_positive",
        "confidence": 0.96,
        "reason": "The flagged text is a TypeScript template literal used to add retry context in a Markdown example. It does not invoke a shell or execute external commands."
      },
      {
        "id": "blocker:references/maker-pattern.md:68:system-reconnaissance",
        "verdict": "false_positive",
        "confidence": 0.94,
        "reason": "The line validates a subtask result against constraints in a sample algorithm. It does not collect host, network, or environment reconnaissance data."
      },
      {
        "id": "blocker:references/maker-pattern.md:90:network-reconnaissance",
        "verdict": "false_positive",
        "confidence": 0.95,
        "reason": "The line is prose about cost and correctness trade-offs for the MAKER pattern. It contains no network scanning, probing, or discovery behavior."
      },
      {
        "id": "external_commands:references/production-hardening.md:166:ruby-shell-backtick-execution",
        "verdict": "false_positive",
        "confidence": 0.91,
        "reason": "The flagged backticks are a TypeScript template literal in a documentation example for branch naming. The static pattern does not show shell backtick execution in runnable package code."
      },
      {
        "id": "sensitive:references/production-hardening.md:113:sqlite-database-file",
        "verdict": "false_positive",
        "confidence": 0.96,
        "reason": "The snippet uses an abstract this.db API in a Markdown example. It does not include or expose a SQLite database file or credential material."
      },
      {
        "id": "sensitive:references/production-hardening.md:119:sqlite-database-file",
        "verdict": "false_positive",
        "confidence": 0.96,
        "reason": "This is an illustrative database update call in documentation. There is no database file, connection string, or sensitive data artifact in the skill package."
      },
      {
        "id": "sensitive:references/production-hardening.md:127:sqlite-database-file",
        "verdict": "false_positive",
        "confidence": 0.96,
        "reason": "The line is a generic this.db.update call in a TypeScript sample. It does not reference a SQLite file or bundled sensitive data."
      },
      {
        "id": "sensitive:references/production-hardening.md:135:sqlite-database-file",
        "verdict": "false_positive",
        "confidence": 0.96,
        "reason": "The snippet releases stale tasks through an abstract database API. It is documentation only and does not contain a SQLite database file."
      },
      {
        "id": "blocker:references/production-hardening.md:112:system-reconnaissance",
        "verdict": "false_positive",
        "confidence": 0.95,
        "reason": "generateId() creates an identifier for a sample task store. It does not enumerate system properties, users, files, or networks."
      },
      {
        "id": "blocker:references/production-hardening.md:279:system-reconnaissance",
        "verdict": "false_positive",
        "confidence": 0.95,
        "reason": "release() is a semaphore method in a concurrency primitive example. It performs no reconnaissance and has no access to host or network metadata."
      },
      {
        "id": "external_commands:references/tool-coordination.md:155:ruby-shell-backtick-execution",
        "verdict": "false_positive",
        "confidence": 0.95,
        "reason": "The line is a TypeScript prompt string passed to a reviewer agent in documentation. It is not shell backtick execution or a command execution sink."
      },
      {
        "id": "sensitive:references/tool-coordination.md:121:certificate-key-files",
        "verdict": "false_positive",
        "confidence": 0.97,
        "reason": "The variable named key is a cache key in a Map iteration. It is not a certificate, private key, or secret file path."
      },
      {
        "id": "blocker:references/tool-coordination.md:51:system-reconnaissance",
        "verdict": "false_positive",
        "confidence": 0.95,
        "reason": "release() unlocks a named resource in a sample ResourceLock class. It does not inspect the host system or gather reconnaissance data."
      },
      {
        "id": "blocker:references/tool-coordination.md:85:system-reconnaissance",
        "verdict": "false_positive",
        "confidence": 0.95,
        "reason": "refill() updates token bucket counters in a rate limiter example. It does not perform network or system discovery."
      },
      {
        "id": "blocker:references/tool-coordination.md:120:system-reconnaissance",
        "verdict": "false_positive",
        "confidence": 0.94,
        "reason": "invalidate(pattern) removes matching entries from an in-memory result cache. It does not enumerate files, services, users, or network hosts."
      },
      {
        "id": "external_commands:SKILL.md:38:ruby-shell-backtick-execution",
        "verdict": "false_positive",
        "confidence": 0.98,
        "reason": "This is Markdown inline code naming a reference file. It is not executable syntax and cannot run an external command."
      },
      {
        "id": "external_commands:SKILL.md:45:ruby-shell-backtick-execution",
        "verdict": "false_positive",
        "confidence": 0.98,
        "reason": "The backticks format an example hierarchical ID in a Markdown table. There is no shell expression or command execution behavior."
      },
      {
        "id": "external_commands:SKILL.md:63:ruby-shell-backtick-execution",
        "verdict": "false_positive",
        "confidence": 0.98,
        "reason": "This is Markdown inline code for a reference file path. It is documentation navigation, not an executable command."
      },
      {
        "id": "external_commands:SKILL.md:88:ruby-shell-backtick-execution",
        "verdict": "false_positive",
        "confidence": 0.98,
        "reason": "This is Markdown inline code pointing to a reference document. It does not execute and contains no shell command."
      },
      {
        "id": "external_commands:SKILL.md:101:ruby-shell-backtick-execution",
        "verdict": "false_positive",
        "confidence": 0.98,
        "reason": "The line uses Markdown inline code to reference a documentation file. It is not Ruby or shell backtick execution."
      },
      {
        "id": "external_commands:SKILL.md:107:ruby-shell-backtick-execution",
        "verdict": "false_positive",
        "confidence": 0.97,
        "reason": "The flagged marker opens a plain Markdown code block for lifecycle steps. It is not executable code and contains no shell command."
      },
      {
        "id": "external_commands:SKILL.md:113:ruby-shell-backtick-execution",
        "verdict": "false_positive",
        "confidence": 0.97,
        "reason": "The flagged marker closes a Markdown code block. It has no command semantics and is not runnable package content."
      },
      {
        "id": "external_commands:SKILL.md:127:ruby-shell-backtick-execution",
        "verdict": "false_positive",
        "confidence": 0.98,
        "reason": "This is Markdown inline code naming a production-hardening reference. It is a documentation pointer, not command execution."
      },
      {
        "id": "external_commands:SKILL.md:181:ruby-shell-backtick-execution",
        "verdict": "false_positive",
        "confidence": 0.98,
        "reason": "The line formats a file path inside the reference table. Markdown inline code does not execute external commands."
      },
      {
        "id": "external_commands:SKILL.md:182:ruby-shell-backtick-execution",
        "verdict": "false_positive",
        "confidence": 0.98,
        "reason": "The finding is a Markdown-formatted reference file path. It is documentation text and not a shell execution pattern."
      },
      {
        "id": "external_commands:SKILL.md:183:ruby-shell-backtick-execution",
        "verdict": "false_positive",
        "confidence": 0.98,
        "reason": "This is a Markdown table entry that names a reference document. It cannot execute and has no command interpreter context."
      },
      {
        "id": "external_commands:SKILL.md:184:ruby-shell-backtick-execution",
        "verdict": "false_positive",
        "confidence": 0.98,
        "reason": "The backticks are Markdown formatting around a file path. They are not shell backticks and introduce no command execution risk."
      },
      {
        "id": "blocker:SKILL.md:46:system-reconnaissance",
        "verdict": "false_positive",
        "confidence": 0.94,
        "reason": "The line describes agent lifecycle states in a Markdown table. It does not probe or enumerate system information."
      },
      {
        "id": "blocker:SKILL.md:149:system-reconnaissance",
        "verdict": "false_positive",
        "confidence": 0.9,
        "reason": "The line lists orchestration modes, including safety-first and hybrid options. It is architectural guidance, not a system reconnaissance instruction."
      }
    ],
    "semantic_findings": [
      {
        "title": "Unsafe Self-Modification Rollback Guidance",
        "description": "The production-hardening reference describes sub-agents modifying code and includes a rollback path that hard-resets the repository to main after failed validation. This can prompt agents to discard workspace changes without explicit user approval.",
        "severity": "medium",
        "locations": [
          {
            "file": "references/production-hardening.md",
            "line_start": 143,
            "line_end": 191
          }
        ],
        "confidence": 0.82,
        "confidence_reasoning": "The guidance is documentation, not executable code, but it directly presents self-modification flow with a hard-reset rollback. The risk depends on a downstream agent following the example without additional approval."
      }
    ],
    "critical_findings": [],
    "high_findings": [],
    "medium_findings": [
      {
        "title": "Unsafe Self-Modification Rollback Guidance",
        "description": "The production-hardening reference describes sub-agents modifying code and includes a rollback path that hard-resets the repository to main after failed validation. This can prompt agents to discard workspace changes without explicit user approval.",
        "locations": [
          {
            "file": "references/production-hardening.md",
            "line_start": 143,
            "line_end": 191
          }
        ],
        "review_kind": "security",
        "source_category": "semantic",
        "source_severity": "medium",
        "confidence": 0.82,
        "confidence_reasoning": "The guidance is documentation, not executable code, but it directly presents self-modification flow with a hard-reset rollback. The risk depends on a downstream agent following the example without additional approval."
      }
    ],
    "low_findings": [],
    "dangerous_patterns": [],
    "files_scanned": 6,
    "total_lines": 1067,
    "audit_model": "codex",
    "audited_at": "2026-07-04T17:37:33.518Z",
    "agent_auto_install_policy": "allowed",
    "manual_install_policy": "allowed"
  },
  "content": {
    "user_title": "Design Reliable Multi-Agent Systems",
    "value_statement": "Multi-agent projects often fail from unclear orchestration, weak boundaries, and unmanaged tool access. This skill provides architecture patterns, safety controls, and production checklists for coordinated agents.",
    "seo_keywords": [
      "Claude",
      "Codex",
      "Claude Code",
      "multi-agent systems",
      "agent orchestration",
      "sub-agent coordination",
      "tool coordination",
      "agent architecture",
      "production agent systems",
      "LLM workflows"
    ],
    "actual_capabilities": [
      "Guides discovery questions for scale, state, tool access, and coordination constraints.",
      "Explains a four-layer architecture with deterministic tools below the reasoning layer.",
      "Compares seven coordination patterns, including fan-out, pipelines, delegation, queues, map-reduce, peer review, and MAKER.",
      "Provides implementation guidance for permission inheritance, locks, rate limits, caching, and sub-agents as tools.",
      "Covers production controls such as cascading stop, orphan detection, checkpointing, cost tracking, and self-modification review."
    ],
    "limitations": [
      "It provides guidance and examples, not a complete runnable framework.",
      "The code examples need adaptation to the target language, runtime, and security model.",
      "High-stakes medical, legal, or financial use needs domain review and external validation.",
      "Cost and reliability estimates are illustrative and should be tested with real workloads."
    ],
    "use_cases": [
      {
        "title": "Plan a New Agent Platform",
        "description": "Define orchestration layers, agent roles, tool boundaries, and lifecycle rules before implementation starts.",
        "target_user": "Engineering leads"
      },
      {
        "title": "Harden Existing Agent Workflows",
        "description": "Add cleanup, checkpointing, rate limits, cost tracking, and permission controls to an active multi-agent system.",
        "target_user": "Platform engineers"
      },
      {
        "title": "Choose Coordination Patterns",
        "description": "Compare parallel, sequential, recursive, queue-based, and voting approaches for a specific product workflow.",
        "target_user": "Product-focused developers"
      }
    ],
    "prompt_templates": [
      {
        "title": "Start With Discovery",
        "scenario": "Beginner planning session",
        "prompt": "Help me design a multi-agent system. Ask the required discovery questions first, then recommend an initial architecture."
      },
      {
        "title": "Select a Pattern",
        "scenario": "Coordination choice",
        "prompt": "Compare fan-out, sequential pipeline, recursive delegation, work queue, map-reduce, peer collaboration, and MAKER for my workflow."
      },
      {
        "title": "Design Tool Boundaries",
        "scenario": "Implementation planning",
        "prompt": "Design the tool coordination model for this agent system, including permissions, locks, rate limits, caching, and cleanup rules."
      },
      {
        "title": "Review Production Readiness",
        "scenario": "Advanced audit",
        "prompt": "Audit this multi-agent design for production readiness. Check cascading stop, orphan detection, checkpoints, cost tracking, and self-modification controls."
      }
    ],
    "output_examples": [
      {
        "input": "I need agents to review many pull request files in parallel.",
        "output": "Recommended a batched fan-out and fan-in design with specialist reviewers, timeouts, cost tracking, and cascading cleanup."
      },
      {
        "input": "My agent workflow loses progress when a session crashes.",
        "output": "Outlined checkpoint triggers, a project-level task store, stale task release, and recovery steps for claimed work."
      },
      {
        "input": "Several agents need access to the same rate-limited API.",
        "output": "Suggested shared token buckets, permission inheritance, cache rules, and lock ownership checks for coordinated access."
      }
    ],
    "best_practices": [
      "Ask the discovery questions before selecting a coordination pattern.",
      "Keep tools deterministic and place LLM calls in orchestration or reasoning layers.",
      "Add lifecycle cleanup, permission limits, timeouts, and cost tracking before production use."
    ],
    "anti_patterns": [
      "Spawning unbounded agents without batching, depth limits, or cost controls.",
      "Letting child agents inherit write or shell permissions they do not need.",
      "Using self-modifying agents without tests, branch isolation, approval gates, and rollback planning."
    ],
    "faq": [
      {
        "question": "Does this skill provide runnable code?",
        "answer": "No. It provides architecture guidance and illustrative TypeScript examples that need project-specific implementation."
      },
      {
        "question": "Which tools can use this skill?",
        "answer": "The metadata lists Claude, Codex, and Claude Code as supported tools."
      },
      {
        "question": "When should I use MAKER?",
        "answer": "Use it only when tasks decompose cleanly, validation is deterministic, and correctness matters more than speed or cost."
      },
      {
        "question": "Does it cover production operations?",
        "answer": "Yes. It covers cascading stop, orphan detection, checkpointing, cost tracking, task stores, and coordination primitives."
      },
      {
        "question": "Can it help with tool permissions?",
        "answer": "Yes. It explains permission inheritance, read-only sub-agents, lock ownership, rate limiting, and cache invalidation."
      },
      {
        "question": "Is it safe for high-stakes domains?",
        "answer": "It can help structure designs, but high-stakes deployments need domain experts, validation, monitoring, and approval controls."
      }
    ]
  },
  "file_structure": [
    {
      "name": "references",
      "type": "dir",
      "path": "references",
      "children": [
        {
          "name": "coordination-patterns.md",
          "type": "file",
          "path": "references/coordination-patterns.md",
          "lines": 175
        },
        {
          "name": "four-layer-architecture.md",
          "type": "file",
          "path": "references/four-layer-architecture.md",
          "lines": 93
        },
        {
          "name": "maker-pattern.md",
          "type": "file",
          "path": "references/maker-pattern.md",
          "lines": 148
        },
        {
          "name": "production-hardening.md",
          "type": "file",
          "path": "references/production-hardening.md",
          "lines": 289
        },
        {
          "name": "tool-coordination.md",
          "type": "file",
          "path": "references/tool-coordination.md",
          "lines": 176
        }
      ]
    },
    {
      "name": "SKILL.md",
      "type": "file",
      "path": "SKILL.md",
      "lines": 186
    }
  ]
}
writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 322kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 118kUpdated Claude CodeCodexVerifiedInstall