v3.13.0

Release v3.13.0 — Agent Spawn Protocol & Multi-Agent Hierarchy Hardening

We are excited to announce v3.13.0, introducing the canonical Agent Spawn Protocol, the Universal TypeName='self' Toolset Inheritance Pattern, hardened multi-agent hierarchy boundaries, and robust tooling differentiation across all agent roles.

---

🤖 Agent Spawn Protocol & Universal TypeName='self' Pattern

• Universal Toolset Inheritance:
  • Standardized TypeName='self' across all pipeline dispatch points (overseer, conductor, tech-lead, red-team-lead, and audit coordinators).
  • Guarantees that spawned subagents inherit full runtime toolsets (file write/edit, shell execution, subagent dispatch) rather than default read-only tool provisions.
  • Role differentiation and domain boundaries remain strictly enforced via the Role descriptor and system prompts referencing .agents/agents/{role}.md.
• Sovereign Subagent Awareness:
  • Added §6 Sovereign Subagent Awareness in agent-protocols, directing coordinators and tech-leads to proactively decompose and delegate multi-domain cards across parallel specialist builders rather than serializing work.

---

🏛️ Pipeline Hierarchy Hardening & Boundary Enforcement

• Overseer Isolation:
  • Enforced strict hierarchy boundaries: Overseer spawns only Conductor and Red Team Lead.
  • Added hard guards prohibiting Overseer from directly dispatching Tech-Leads/Builders, giving implementation instructions, or embedding dual conversation IDs.
• Tech-Lead Delegation Mandate:
  • Strengthened Tech-Lead role boundaries in workflow-team.md to explicitly forbid writing primary feature domain code; mandates parallel delegation to specialized builders (backend-engineer, frontend-engineer, test-automation-engineer).
  • Added instruction-override guards preventing Tech-Leads from flattening delegation when instructed to self-implement.
• Scoped System Prompt Templates:
  • Scoped prompt templates in workflow-team.md to Conductor & Tech-Lead use only, preventing role bleed across supervisory layers.

---

🛠️ Tooling Differentiation & Reactive Lifecycle Clarity

• Subagent Status vs Background Tasks:
  • Resolved model hallucinations by differentiating manage_subagents (for checking subagent lifecycle states) from manage_task (reserved strictly for background processes/shell jobs).
  • Added negative instructions preventing Tech-Leads and Red Team Leads from invoking manage_task or setting watchdog timers on subagents.
• Reactive Wakeups:
  • Clarified that subagent completion messages trigger automatic reactive wakeups, removing unnecessary polling loops and redundant timer calls.
• Ecosystem Alignment:
  • Synchronized spawn specifications and checklist standards across /audit, /perf-optimize, /security-audit, overseer.md, and conductor.md.

---

📦 Summary of Commits included in v3.13.0

• 3498490 docs(protocols): clarify subagent monitoring tools across skills and coordinators
• 36eefb8 fix(agents): prevent tech-lead manage_task tool confusion after builder dispatch
• fd7c358 fix(agents): reconcile stale spawn protocol standards in overseer and conductor
• 01b4c4e fix(pipeline): standardize Universal TypeName='self' spawn pattern across all tiers
• 260b2ce fix(team): prevent overseer from issuing implementation commands and enforce tech-lead delegation
• 7d23c62 fix(team): enforce hierarchy boundaries to prevent overseer from bypassing conductor
• 565e42e refactor(workflows): reconcile ad-hoc subagent spawn rationale in audit pipelines
• 9c4fe9d fix(team): restore named subagent delegation and parallel builder dispatch
• ffd2427 feat(protocols): introduce Agent Spawn Protocol and sovereign subagent awareness

---

📦 Installation & Upgrade

Upgrade your local project using the selective CLI installer:

npx awesome-agv@latest

Full Changelog: v3.12.0...v3.13.0