dispatching-parallel-agents

1# Dispatching Parallel Agents
2 
3## Purpose
4 
5Use parallel children for two or more ad-hoc, bounded tasks only when their
6independence is already credible and concurrency is worth the coordination.
7Parallelism changes elapsed time; it does not relax evidence, ownership, or
8completion requirements.
9 
10## Dispatch Gate
11 
12Dispatch only when all are true:
13 
14- there are `2+` distinct tasks or evidence questions;
15- one task's result is not needed to define or start another;
16- the tasks do not require overlapping writes or the same mutable resource;
17- a shared root cause is unlikely based on current evidence;
18- each result can be returned with bounded evidence and unresolved unknowns;
19- available concurrency and expected work justify coordination cost.
20 
21Keep work inline when decomposition is still exploratory, failures may share a
22root cause, full-system context is required, or tasks are too small to repay the
23handoff. A written implementation plan routes to
24`aegis:subagent-driven-development`, not this ad-hoc workflow.
25 
26## Context And Ownership
27 
28Choose the live host's context inheritance mode deliberately. Give each child
29the minimum sufficient instructions and evidence; do not rely on accidental
30inheritance or assume that context must never be inherited.
31 
32Each task packet states:
33 
34- goal and question to resolve;
35- allowed files, systems, and side effects;
36- known facts and evidence locations;
37- constraints, non-goals, and stop condition;
38…