Human Approval Boundary Mapper

Automation without oversight leads to irreversible failures. This skill solves the problem of "invisible risk" in delegated workflows by mapping exactly where an agent can act alone and where a human must intervene. It eliminates the ambiguity of tool permissions and prevents catastrophic errors like duplicate payments or unauthorized data deletion.

What it does

• Delegation mapping categorizes every workflow action into auto-execute, prepare-only, human-approval, or human-execution zones.
• Consequence classification ranks actions from low to critical to determine the required strength of human gates.
• Evidence verification uses a strict fact-based model to confirm tool access and completion rather than assuming success.
• Approval gate design creates specific, occurrence-based checkpoints that prevent a single approval from granting permanent authority.
• Recovery planning defines explicit rollback steps and exception handling for high-stakes actions.

How it works

1. Inventory actions by decomposing the workflow into granular steps, tools, and side effects.
2. Assign control states based on confirmed permissions and consequence levels.
3. Generate the Control Pack which includes the delegation map, tool-scope matrix, and boundary decision ledger.
4. Audit readiness against a checklist of missing evidence, unresolved blockers, and safety risks.

Frameworks & tools

This tool is platform-agnostic. It works for any stack involving CRM, billing systems, CI/CD pipelines, APIs, or internal databases where delegated agents require strict operational boundaries.

Why this beats prompting it yourself

Generic prompts often ignore the distinction between "intent to execute" and "verified completion." This skill forces a rigorous evidence-based audit that prevents scope leakage and ensures that critical actions like payments or deletions never happen without observable human authorization.

Use cases

• Defining safe boundaries for an agent managing customer billing and refunds.
• Mapping approval gates for automated social media publishing workflows.
• Auditing tool permissions before deploying a multi-agent system to production.
• Establishing recovery paths for automated database migrations or infrastructure changes.

Known limitations

This skill performs analysis and mapping based on provided documentation; it does not directly configure software permissions or execute code. Missing evidence will result in blocked states.