Claude Hemat

Provides Q&A, casual chat, and simple tasks using Haiku 4.5; spawns Sonnet 4.5 or Opus 4.6 for analysis, reasoning, and complex tasks as needed.

Security Scan

The skill's instructions, requirements, and behavior are internally consistent with its stated purpose (model selection and escalation); it is an instruction-only skill that asks for no credentials or installs.

Purpose & Capability

The name/description (use Haiku 4.5 by default, escalate to Sonnet 4.5 or Opus 4.6 for harder tasks) matches the SKILL.md. No unrelated environment variables, binaries, or installs are requested.

Instruction Scope

SKILL.md contains clear escalation rules and exact session_spawn calls to invoke other models. It does not instruct reading system files, environment secrets, or external endpoints beyond calling models. One runtime assumption: the agent platform must implement a sessions_spawn API and provide the named models (anthropic/claude-sonnet-4-5, anthropic/claude-opus-4-6); if those are unavailable the instructions won't work as written.

Install Mechanism

No install spec or code files are present (instruction-only). Nothing will be written to disk during installation.

Credentials

The skill declares no required environment variables, credentials, or config paths — proportionate for a purely instructional model-routing skill.

Persistence & Privilege

always is false and there are no special persistence requests. The skill allows normal autonomous invocation (disable-model-invocation:false by default), which is the platform default. Be aware that autonomous invocation combined with the ability to spawn additional models increases runtime capability (and potential data exposure) but is expected behavior for a skill that routes to other models.

Assessment

This skill appears coherent and low-risk: it only contains rules for when to use a default model vs escalate to stronger models and does not request credentials or install code. Before using, verify that your platform supports the sessions_spawn mechanism and the specific model identifiers referenced (anthropic/claude-sonnet-4-5, anthropic/claude-opus-4-6); if those models are billed or external, spawning them may incur cost and will expose the user prompt/data to those models. Note SKILL.md allows "file lookups" but does not define which files — avoid providing sensitive files unless you trust the runtime and destination model. If you prefer the agent not to call this skill autonomously, set disable-model-invocation or restrict skill permissions per your platform.

Introduction

1. Haiku 4.5 is the default model. Only spawn Sonnet 4.5 or Opus 4.6 when the task actually needs either of them
2. Check Model Rules section to decide which model to use!
3. Use sessions_spawn to use more advanced models

sessions_spawn(
  message: "<the full task description>",
  model: "anthropic/claude-sonnet-4-5",
  label: "<short task label>"
)

sessions_spawn(
  message: "<the full task description>",
  model: "anthropic/claude-opus-4-6",
  label: "<short task label>"
)

Model Rules

Haiku 4.5

1. Simple Q&A - What, When, Who, Where
2. Casual chat - No reasoning needed
3. Quick lookups
4. File lookups
5. Simple tasks - May but not limited to repetitive tasks
6. Cron Jobs, if it needs reasoning, THEN ESCALATE TO THE NEXT ADVANCED MODEL
7. Basically everything that doesn't need any thinking
8. Provide concise output, just plain answer, no explaining
9. DO NOT CODE WITH HAIKU 4.5
10. DO NOT ANALYZE USING HAIKU 4.5
11. DO NOT ATTEMPT ANY REASONING USING HAIKU 4.5
12. If you think the request does not fall into point 1-6, THEN ESCALATE TO THE NEXT ADVANCED MODEL
13. If you think you will violate point 8-10, THEN ESCALATE TO THE NEXT ADVANCED MODEL

Sonnet 4.5

1. Analysis - Why, How
2. Code
3. Planning
4. Reasoning
5. Comparisons
6. Reporting
7. If you think the request is pretty critical to the user, THEN ESCALATE TO THE NEXT ADVANCED MODEL

Opus 4.6

1. Deep research
2. Critical decisions
3. Extreme complex reasoning
4. Extreme complex planning
5. Detailed explanation

Other Notes

1. When the user asks you to use a specific model, use it
2. Always put which model is used IN EVERY OUTPUTS
3. After you are done with more advanced models (Sonnet 4.5 or Opus 4.6), revert back to Haiku 4.5 as the default model