From agent-platform-pro
Multi-squad orchestration: task decomposition, resource optimization, routing, error recovery, cross-squad dependencies. Use when planning or managing large epics.
How this skill is triggered — by the user, by Claude, or both
Slash command
/agent-platform-pro:epic-orchestratorThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Epic Orchestrator is the orchestration skill for large-scale projects that run for days or weeks, require 5-50+ sub-agents, and involve the coordinated effort of multiple independent squads. **Benchmark-proven: without structured decomposition rules, models produce vague timelines with no dependency DAGs, token budgets, or failure recovery strategies.**
Epic Orchestrator is the orchestration skill for large-scale projects that run for days or weeks, require 5-50+ sub-agents, and involve the coordinated effort of multiple independent squads. Benchmark-proven: without structured decomposition rules, models produce vague timelines with no dependency DAGs, token budgets, or failure recovery strategies.
Core Principle: FOR SMALL TASKS, USE SUBAGENT-ORCHESTRATOR; FOR LARGE TASKS, USE EPIC ORCHESTRATOR.
Seven rules that are never violated:
USE WHEN:
DO NOT USE (use subagent-orchestrator instead):
NEVER USE (delegate to human instead):
grill first).opportunity-solver + validation-designer first).research-first to discover domain context first).Every run follows a strict 10-phase sequence. The full per-phase protocol — with YAML schemas and worked examples for each phase — is in REFERENCE.md.
| Phase | Name | Key output |
|---|---|---|
| 0 | Project Ingestion & Scoping | Validated brief, success criteria, constraints — no agents spawned |
| 1 | Epic Decomposition | Epic/story/task hierarchy, dependency DAG, critical path |
| 2 | Squad Design | Role assignments, handoff contracts, token budgets |
| 3 | Resource Planning | Timeline, bottleneck analysis, parallelism map |
| 4 | Squad Spawn & Monitoring Init | Live squads, alert rules, human checkpoints |
| 5 | Mid-Flight Steering | Task injection, rebalancing, scope cuts |
| 6 | Error Recovery & Resilience | 3-tier retry → fallback → escalate per failure type |
| 7 | Result Consolidation & Verification | Integration tests, consistency checks, quality gates |
| 8 | Post-Mortem & Learning | Planned-vs-actual, extracted patterns, skill updates |
| 9 | Closing & Archiving | Packaged deliverables, benchmarks, final summary |
Every plan must be expressed in the mandatory YAML schema and must pass the bad/good output contrast check. See REFERENCE.md for the full schema, a concrete bad/good example pair, and model-specific calibration notes.
Stop immediately and rerun decomposition if any of these appear mid-flight:
| Excuse | Reality |
|---|---|
| "This project is too simple, epic decomposition is overkill" | What starts simple today easily scales to 3 epics and 15 stories tomorrow. Plan first. |
| "We will resolve inter-squad dependencies as we go" | Resolving dependencies ad-hoc is a direct route to project deadlock. Map them in a DAG. |
| "Estimating token budgets is a waste of time" | 70% of projects launched without token budgeting stall due to context exhaustion. |
| "Let's run 10 squads simultaneously to speed up delivery" | The coordination cost of 10 concurrent squads consumes twice the output speed of 5 squads. |
| "We will conduct the post-mortem once the project finishes" | At the end of the project, exhaustion sets in and critical lessons are forgotten. Run post-mortems after each epic. |
| "Let's give all squads access to the entire repository" | Spawning 50 sub-agents with 100K token contexts consumes 5M tokens immediately. Budget will run out. |
Escalate immediately when:
Inform at checkpoints:
Recommended chaining after completion: squad-builder → task-lifecycle-manager → agent-teammate → skill-marketplace → post-mortem → skill_manage.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub haj1t/senior-dev-squad-skills --plugin agent-platform-pro