From memory-access
Operate strict orchestrator/subagent workflows with enforceable contracts, lock management, and fresh-context handoff execution. Use when planning or running agent swarms, validating assignment/result packets, preventing lock overlap, reconciling task ledgers, or recovering blocked multi-agent runs.
How this skill is triggered — by the user, by Claude, or both
Slash command
/memory-access:multi-agent-operator-guideThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Run strict multi-agent execution with machine validation, not prompt text alone.
agents/openai.yamlassets/handoff-template.jsonexamples/01-assignment-invalid-dependency-id.jsonexamples/01-assignment-invalid-dependency-id.output.jsonexamples/01-assignment-invalid-empty-normalized-lock.jsonexamples/01-assignment-invalid-empty-normalized-lock.output.jsonexamples/01-assignment-valid.jsonexamples/01-assignment-valid.output.jsonexamples/02-locks-conflict.jsonexamples/02-locks-conflict.output.jsonexamples/02-locks-valid.jsonexamples/02-locks-valid.output.jsonexamples/03-result-invalid-empty-acceptance.jsonexamples/03-result-invalid-empty-acceptance.output.jsonexamples/03-result-invalid-secret-note.jsonexamples/03-result-invalid-secret-note.output.jsonexamples/03-result-valid.jsonexamples/03-result-valid.output.jsonexamples/04-reconcile-valid.jsonexamples/04-reconcile-valid.output.jsonRun strict multi-agent execution with machine validation, not prompt text alone.
Use this skill when you need deterministic orchestration across multiple agents, especially when tasks run in parallel and require lock-safe writes, append-only worklogs, and schema-validated handoffs.
memory-access:orchestrator) with global objective and constraintsscripts/validate_packet.py)scripts/check_lock_overlap.py)scripts/build_dispatch_prompt.py and spawn any agent type via Task toolscripts/validate_result.py)scripts/reconcile_ledger.py)If validator scripts are unavailable, stop and mark run blocked with MISSING_VALIDATOR.
These two payloads are the minimum required control-plane artifacts for a first valid run.
{
"schema_version": "1.0.0",
"run_id": "3f56dc4d-35cf-4f97-925c-0b04a6fe8bf4",
"packet_type": "assignment",
"global_objective": "Implement endpoint tests",
"task": {
"task_id": "T-12",
"title": "Add endpoint tests",
"type": "parallelizable",
"dependencies": [],
"lock_scope": ["tests/test_api.py"],
"forbidden_scope": ["src/"],
"acceptance_criteria": ["All endpoint tests pass"],
"worklog_path": "worklogs/T-12.jsonl",
"timeout_seconds": 1200,
"heartbeat_interval_seconds": 120,
"priority": "high"
},
"active_locks": [
{"task_id": "T-9", "resource": "src/api.py", "active": true}
],
"context_package": [
{"kind": "file", "value": "tests/test_api.py"},
{"kind": "constraint", "value": "Do not edit src/api.py"}
],
"required_output_schema": "subagent_result_v1"
}
{
"schema_version": "1.0.0",
"run_id": "3f56dc4d-35cf-4f97-925c-0b04a6fe8bf4",
"task_id": "T-12",
"status": "done",
"changes": [
{"resource": "tests/test_api.py", "action": "edit", "evidence": "Added endpoint assertions"}
],
"acceptance_check": [
{"criterion": "All endpoint tests pass", "status": "pass", "evidence": "pytest tests/test_api.py"}
],
"worklog_path": "worklogs/T-12.jsonl",
"notes_for_orchestrator": ["No conflicts, ready for merge"]
}
references/contracts.md (source of truth)references/error-codes.mdreferences/state-machine.mdreferences/locking.mdThe orchestrator role is a proper agent (agents/orchestrator.md) — its contract is the system prompt when spawned. The subagent contract (references/subagent-directive.md) is a directive preamble that the orchestrator injects into the prompt of any agent it dispatches, ensuring lock/scope/result compliance regardless of agent type.
Read only additional references as needed:
references/fresh-context.md for new-session handoff/reset protocolreferences/security.md for untrusted context handlingreferences/slo.md for quality targets and operational metricsassets/handoff-template.json and fill run metadata.scripts/validate_packet.py.scripts/check_lock_overlap.py before dispatch.scripts/validate_result.py.scripts/reconcile_ledger.py before marking done.examples/README.md for a complete runnable walkthrough with expected validator outputs.Borrowed from wshobson/agents conductor patterns:
todo, in_progress, blocked, terminal) and do not skip validation gates.schema_version: 1.0.0 on all machine payloads.references/error-codes.md.done without acceptance evidence.notes_for_orchestrator without secret-pattern filtering.npx claudepluginhub emmahyde/memory-access --plugin memory-accessManages multi-agent orchestration using contracts, AgentDB briefing, 4 fault tolerance layers (retry, fallback, classification, checkpointing), and context transfer protocols. Use for coordinating parallel agents or spawning sub-agents.
Guides proactive use of /orchestrate for complex multi-agent workflows with sequential/parallel execution, conditionals, retries, visualization, checkpoints, and automation.
Coordinates multiple AI agents through a single orchestrator: decomposes tasks, routes to specialists, prevents duplication, and enforces quality gates with heartbeat monitoring.