Agent delegation
Skill ZC0325/sol-native-guardrails/plugins/sol-native-guardrails/skills/agent-delegation
Five narrowly scoped Codex engineering skills that preserve native model judgment.
npx -y skills add ZC0325/sol-native-guardrails --skill agent-delegationAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 25 days oldThe repository was created 25 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Orchestrate user-authorized subagents for multiple independent tasks, bounded research, or independent review when parallelism or context isolation materially improves the outcome. Use only when the platform permits delegation and shared-state conflicts can be controlled.
SKILL.md
1.9 KB, 309 tokens by cl100k_base, as published. Nobody here has run it
Agent Delegation
Delegate only when isolation creates more value than coordination cost. The primary Codex agent remains responsible for integration and verification.
Entry Check
Confirm all of the following:
- Delegation is within the user's request or explicit authorization.
- The platform and current policy allow subagents.
- Each task has a bounded outcome and can be understood with task-local context.
- Parallel agents will not edit the same files, compete for the same mutable resource, or depend on unresolved shared decisions.
If these conditions are not met, work inline or sequence the tasks.
Workflow
- Divide work by independent problem domain, not by arbitrary file count.
- Give each agent the goal, relevant evidence, scope, constraints, allowed side effects, verification expectation, and required output.
- Avoid leaking the expected conclusion when the agent is being used for independent review or validation.
- Use no more agents than the task justifies. Keep tightly coupled implementation under one owner.
- Review returned artifacts and diffs; do not trust a success summary by itself.
- Resolve conflicts and run integrated verification before reporting completion.
Boundaries
- Do not mandate a subagent for every task or a fixed number of review passes.
- Do not downgrade model capability automatically; match the role to risk and available options.
- Do not require a worktree unless Git isolation is actually needed and authorized.
- Do not parallelize exploratory debugging before independent problem domains are known.
- Do not delegate decisions that require missing user authority.
What ships with it: 1 file
223 B alongside SKILL.md
agents/
- openai.yaml223 B