Kbediako/evergreen-codex-skills / skills/native-subagents-first
native-subagents-first
VerifiedCoordinate native Codex subagents for requested parallel work or independent execution and review scopes. Use for ownership, worker selection, and lifecycle management.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 5.5 KB |
| agents/openai.yaml | config | 243 B |
| references/astra-routing.md | markdown | 2.7 KB |
| references/briefs-and-ledger.md | markdown | 5.3 KB |
| references/read-only-boundaries.md | markdown | 2.1 KB |
references/astra-routing.md
View on GitHub# GPT-6 Astra role routing
Within this suite, use GPT-6 Astra and vary only reasoning effort. If a higher-priority instruction requires another model, treat that as a different worker condition outside this routing table.
## Selection
| Role | Effort | Choose when | Do not choose merely because |
| --- | --- | --- | --- |
| Scout | `medium` | The question is narrow, read-only, and independently checkable | The repository is large |
| Worker | `high` | Implementation or review needs complex tracing, checks, or edge-case judgment | A task edits more than one file |
| Smart worker | `xhigh` | Ambiguity spans boundaries, competing designs remain, or a prior scoped repair failed | More time is available |
| Deep worker | `max` | The work is exceptionally difficult, novel, high-impact, or needs extensive alternative exploration | It is labeled "important" |
Keep `medium` as the normal floor for this suite. Escalate one step only when the brief names the uncertainty or quality boundary the extra effort should resolve. If the higher-effort treatment does not change accepted outcomes, return to the lower role.
Use custom roles only when their resolved model and effort match the assignment; a role's configured values can override explicit spawn values. Otherwise omit `agent_type` and set `model: "gpt-6-astra"` plus `reasoning_effort` explicitly.
## Context inheritance
- Use `fork_turns: "none"` for a fresh, focused assignment. Restate task-local permissions, safety constraints, exact scope, and acceptance.
- Use a positive bounded history count when recent decisions are required and can be identified.
- Use full history only when the whole conversation is necessary and inheriting the parent model/effort is acceptable.
- Do not pass model or effort overrides with a full-history fork.
Treat a child's model, effort, custom role, tool surface, runtime version, and prompt as part of its worker condition. Hold those constant when evaluating a skill change. A different model or effort opens a new comparison condition.
## Ultra and validation
When the active Codex runtime exposes `ultra`, reserve it for high-stakes parent coordination. Do not configure routine children at `ultra`; the coordinator should own fanout and synthesis.
Qualify this routing with representative local outcomes. Benchmark scores, token counts, and a single rollout do not establish that higher effort or more agents improve a task.
Sources:
- [GPT-6 Astra model and reasoning efforts](https://developers.openai.com/api/docs/models/gpt-6-astra)
- [GPT-6 Astra prompting and migration guidance](https://developers.openai.com/api/docs/guides/latest-model?model=gpt-6-astra#prompting-best-practices)
- [Codex subagent configuration](https://learn.chatgpt.com/docs/agent-configuration/subagents)
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |