Safe Long-Run Mode GPT5.4 + Claude

Operate long-running AI tasks safely across GPT-5.4 and Claude by using model selection rules, phased execution, checkpoints, resumable workflows, API thrott...

Security Scan

This is an instruction-only guidance skill for running long tasks across GPT-5.4 and Claude; its requirements and actions are coherent with the stated purpose and it does not request extra credentials or install anything.

Purpose & Capability

The skill is purely procedural guidance (phases, checkpoints, model routing) and matches the name/description. It references external systems (Azure, Graph, Orgo, browser automation) only as contexts where long-run practices apply; it does not require unrelated binaries, env vars, or installs.

Instruction Scope

The instructions tell the agent to write artifacts to disk, use subagents, validate auth, and interact with external APIs/GUI automation. Those operations are appropriate for long-running workflows, but they are broad: follow-on actions (auth/access to APIs, saving screenshots, writing files) will depend on the agent/platform connectors. The skill itself does not supply code or explicit endpoints.

Install Mechanism

No install spec and no code files — lowest-risk pattern. Nothing will be written to disk by an installer and there are no download URLs to evaluate.

Credentials

The skill declares no required environment variables or credentials, which is proportionate. However, it prescribes interactions with cloud control planes and external APIs; those will require platform-supplied credentials at runtime. Users should ensure any credentials granted to the agent are minimal-scope and appropriate.

Persistence & Privilege

Defaults are used (always:false, agent-invocable:true). The skill does not request permanent presence, nor does it modify other skills' configurations. Autonomous invocation is allowed by platform default and is not a concern on its own.

Assessment

This skill is a set of best-practice instructions — it does not install software or ask for secrets, so it's internally coherent. Before enabling it: (1) confirm where the agent will save checkpoints and ensure those storage locations are secure and have appropriate access controls; (2) restrict any cloud/API credentials the agent uses to least privilege and review auditing/expiration policies; (3) ensure subagent/autonomous runs are acceptable in your environment (monitor and timeout long tasks); and (4) watch for any agent actions that actually perform destructive cloud operations — the skill advises safe patterns, but execution still depends on the platform and connectors. If you want stronger assurance, ask for a concrete runbook or a list of connectors the agent will use and a short example of how checkpoints/artifacts are stored.

Use this skill for tasks that may run long, span multiple systems, or risk losing progress if interrupted.

Core rule

Do not run long tasks as one monolithic attempt. Split into phases, write checkpoints, and keep the work resumable.

Model selection

Use GPT-5.4 for:

• coding
• docs
• research
• file-heavy transformations
• multi-agent delegated work
• repetitive build tasks
• long internal work where cost and throughput matter

Use Claude for:

• strategic judgment
• sensitive decisions
• nuanced synthesis
• client-facing polish
• brand voice refinement
• high-trust orchestration

Default to GPT-5.4 first. Escalate to Claude only when the task actually benefits from higher-quality judgment or tone.

Operating procedure

1. Scope before acting

Before starting, decide:

• what the final deliverable is
• which systems/tools will be touched
• what can fail or throttle
• what must be saved after each phase

2. Break work into phases

Use phases such as:

1. gather / inspect
2. plan / write brief
3. execute / edit / build
4. validate
5. deploy or report

At the end of each phase, write artifacts to disk.

3. Always checkpoint

For long tasks, save progress in files:

• draft outputs
• notes
• reports
• partial results
• tracker entries
• checkpoint summaries

Prefer a resumable workspace state over a perfect one-shot run.

4. Isolate long work

Use subagents when:

• the task will take more than a few tool calls
• multiple files/systems are involved
• external APIs are involved
• failure should not pollute the main session
• specialized work can be delegated cleanly

5. Throttle external systems

When interacting with Azure, Graph, Orgo, messaging providers, registries, websites, or any external API:

• batch reads when possible
• avoid tight polling loops
• serialize risky writes
• respect retry/backoff
• avoid one-item burst loops when a bulk operation is possible

6. Prefer resumability over perfection

The goal is not "never fail." The goal is: if interrupted, resume with minimal loss.

System-specific guidance

Azure / cloud control planes

• validate auth first
• create foundational resources first
• verify after each layer
• log resource names/IDs
• do not chain long destructive commands blindly

Browser / Orgo / GUI automation

• use explicit goals and stop conditions
• capture screenshots at checkpoints
• bound retry counts
• save artifacts locally
• prefer API/CLI over GUI when equivalent exists

Coding / documentation work

• create a brief/spec first for complex tasks
• write files in chunks
• validate after each major change
• leave notes for resume if work is unfinished

What to tell the user

When relevant, explain that safe long-run mode means:

• cheapest adequate model
• phased execution
• saved checkpoints
• subagent isolation
• controlled API usage
• resumable progress

Failure handling

If a long task is interrupted:

1. summarize completed phases
2. point to saved artifacts
3. identify exact next step
4. resume from checkpoint rather than restarting

References

• Read references/checklist.md for a reusable pre-flight checklist and model routing matrix.