Installs into .claude/skills of the current project.
Are you the author of 2635 Orchestrator Discipline 98aae0be?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/tools-only-2635-orchestrator-discipline-98aae0be)
# Orchestrator Discipline
Orchestrators delegate; agents implement. The orchestrator's context window is shared across the session — agents get fresh context per task.
---
## Anti-Patterns
| Anti-Pattern | Trigger | Response |
|--------------|---------|----------|
| **Investigation Escalation** | 3+ Read/Grep/Bash on source files without Edit/Write/Task | STOP. Delegate. Do not read one more file. |
| **Agent Output Polling** | TaskOutput block=false on running agent; Read on .output/transcript | Wait for completion notification. Never poll. |
**No exemption categories**: "config changes", "small edits", "just TOML/YAML" are not valid reasons to skip delegation.
---
## Read Constraints
**PERMITTED**: Task status, agent output artifacts, backlog items (.claude/backlog/ per-item files), plan files, skill/agent configs, files you will Edit/Write this turn.
**NEVER**: Source/config/test files you will not edit; diagnostic command output; agent .output files; TaskOutput block=false on running agent.
**Falsifiable test**: "Will I Edit or Write this file this turn?" If NO — delegate.
---
## CoVe Bypass Anti-Pattern
Orchestrators pass paths and outcomes; agents discover and verify. Do not pre-gather data for agents.
---
## Source
- [orchestrator-discipline rules](../../../plugins/orchestrator-discipline/rules/CLAUDE.md)
- [orchestrator-discipline SKILL.md](../../../plugins/orchestrator-discipline/skills/orchestrator-discipline/SKILL.md)