lawve-ai/awesome-legal-skills/agent-authority-charter-builder-arkadiy-miteiko

Claude Code and agent skill by lawve-ai/awesome-legal-skills

Trust tier 2
  • community
LLM review pending
Source check pending

Checking that the SKILL.md file is still in the repository.

Creates an Agent Authority Charter for enterprise or regulated AI agents before deployment. Use this Skill when a user needs to define what an AI agent is allowed to do, who delegated authority to it, what actions are permitted or prohibited, when human approval is required, what evidence must be preserved, and how the agent can be suspended, revoked, or escalated.

Install

npx vskill@latest install lawve-ai/awesome-legal-skills/agent-authority-charter-builder-arkadiy-miteiko

SKILL.md

The same SKILL.md was published earlier as lawve-ai/awesome-legal-skills/agent-authority-charter-builder-arkadiy-miteiko.

Agent Authority Charter Builder

Purpose

This Skill helps legal, compliance, risk, product, operations, and AI governance teams create an Agent Authority Charter before an AI agent is deployed into an enterprise or regulated workflow.

The purpose is not to decide whether an AI system is generally ethical, safe, or compliant. The purpose is narrower and more operational:

Define what institutional authority the agent has before it acts.

The Skill converts a proposed AI agent use case into a structured, reviewable governance artifact covering:

  • agent identity;
  • principal and delegation source;
  • operational scope;
  • permitted actions;
  • prohibited actions;
  • human approval rules;
  • escalation triggers;
  • evidence and audit requirements;
  • suspension, revocation, and kill-switch conditions;
  • deployment-readiness determination.

The final output should be suitable for review by legal, compliance, risk, security, business, audit, and technical stakeholders.

Core Principle

An AI agent should not be governed only by what it can technically do.

It must be governed by what the institution has authorized it to do.

The charter must answer:

  1. Who authorized the agent?
  2. What is the agent allowed to do?
  3. What is the agent forbidden from doing?
  4. What conditions must be satisfied before action?
  5. What human approvals are required?
  6. What evidence must be preserved?
  7. When must the agent stop and escalate?
  8. Who can suspend, revoke, or amend the authority?
  9. What happens if the agent exceeds scope?

When to Use This Skill

Use this Skill when the user asks to:

  • define the authority of an AI agent;
  • prepare an AI agent for enterprise deployment;
  • document agent permissions;
  • create a governance artifact for an AI workflow;
  • prepare for legal, compliance, risk, or audit review of an agent;
  • distinguish between assistant, copilot, workflow automation, and agent authority;
  • prepare a regulator-facing or internal AI pilot;
  • define human approval, escalation, or kill-switch requirements;
  • assess whether an agent should be permitted to take actions inside business systems;
  • create an authority model for autonomous or semi-autonomous AI;
  • review whether an AI agent can safely affect records, customers, transactions, workflows, or external commitments.

Use this Skill even if the user phrases the request informally, such as:

  • “Can this agent approve things?”
  • “What should we document before deploying this agent?”
  • “Help me define what this AI is allowed to do.”
  • “We need a governance charter for an internal agent.”
  • “What permissions should this agent have?”
  • “How do we make this agent safe for compliance review?”
  • “What needs human approval?”
  • “Where should we put escalation rules?”
  • “Can this AI act on behalf of the company?”

When Not to Use This Skill

Do not use this Skill to:

  • provide legal advice or a legal opinion;
  • approve an agent for deployment;
  • classify an AI system under a specific statute unless the user provides the relevant legal framework;
  • draft a full enterprise AI policy;
  • perform cybersecurity testing;
  • replace legal, compliance, risk, privacy, security, or regulator review;
  • create technical access-control code;
  • make final determinations about regulatory compliance;
  • authorize an AI agent to operate.

If the user asks for legal conclusions, state that the output is a governance drafting aid and should be reviewed by qualified counsel and the appropriate institutional authority.

Required Inputs

Collect as many of the following inputs as possible. If the user does not provide enough information, proceed with reasonable assumptions but mark missing items as “To be confirmed.”

1. Organization Context
  • Organization name
  • Industry
  • Jurisdiction or operating region
  • Regulated or non-regulated environment
  • Business unit
  • Relevant internal policies
  • Relevant control environment
  • Relevant regulatory obligations, if known
2. Agent Identity
  • Agent name
  • Agent purpose
  • Agent type
  • Business owner
  • Technical owner
  • Compliance owner
  • Legal owner, if applicable
  • Deployment environment
  • Systems accessed

Possible agent types include:

  • Informational Assistant
  • Decision Support Copilot
  • Workflow Preparation Agent
  • Bounded Execution Agent
  • Consequential Execution Agent
  • Observer / Governance Agent
  • Escalation / Control Agent
  • Remediation Agent
3. Principal and Delegated Authority

Identify who or what delegates authority to the agent.

Authority may come from:

  • named executive owner;
  • business process owner;
  • legal department;
  • compliance department;
  • risk function;
  • internal AI governance committee;
  • board-approved policy;
  • contract;
  • regulator-approved pilot;
  • system owner;
  • workflow-specific operating procedure.

If the source of authority is unclear, mark the charter as not deployment-ready.

4. Operational Scope

Identify:

  • business process or workflow;
  • systems the agent may access;
  • tools the agent may use;
  • data the agent may read;
  • data the agent may write, change, or modify;
  • transactions or records the agent may affect;
  • internal users the agent may serve;
  • external parties the agent may interact with;
  • geographic limits;
  • customer, product, transaction, or risk limits;
  • time limits or pilot limits.
5. Permitted Actions

Define what the agent may do:

  • without human approval;
  • only with human approval;
  • only during a pilot;
  • only below a threshold;
  • only for a specific class of users, cases, customers, or transactions.

Examples of permitted actions:

  • retrieve information;
  • summarize documents;
  • classify records;
  • prepare recommendations;
  • draft communications;
  • update internal notes;
  • initiate a workflow;
  • approve low-risk cases within a threshold;
  • reject incomplete submissions;
  • escalate exceptions;
  • notify a human reviewer;
  • generate an evidence pack.
6. Prohibited Actions

Define what the agent may not do.

Examples of prohibited actions:

  • bind the organization contractually;
  • approve credit, insurance, hiring, benefits, medical, legal, financial, or disciplinary decisions without review;
  • make external statements on behalf of the organization;
  • override human decisions;
  • alter audit logs;
  • delete records;
  • access unauthorized systems;
  • create new authority for itself;
  • act outside its defined workflow;
  • continue operating after suspension or revocation;
  • make decisions involving protected characteristics unless specifically authorized and legally reviewed;
  • act where the policy basis is ambiguous.
7. Action Tiers

Classify each agent action into one of the following authority tiers.

Tier 0 — Observe Only

The agent may read, summarize, classify, or flag information. It may not alter records, trigger workflows, make decisions, communicate externally, or create business consequences.

Tier 1 — Prepare

The agent may draft, structure, recommend, or queue actions for human review. It may not execute the action without human approval.

Tier 2 — Execute Bounded Internal Action

The agent may perform low-risk internal actions within clearly defined thresholds. The action must be logged and reversible where possible.

Tier 3 — Execute Consequential Action with Human Approval

The agent may prepare or initiate consequential actions only after explicit human approval. The approval must be preserved as part of the evidence record.

Tier 4 — Prohibited or Reserved Authority

The agent may not perform these actions. They require human, legal, compliance, board, regulator, or other institutional authority.

Drafting Workflow

Follow this process when creating the charter.

Step 1 — Identify the Agent’s Institutional Role

Determine whether the agent is assisting a human or exercising delegated authority.

Ask:

  • Is the agent producing information only?
  • Is it changing records?
  • Is it triggering workflows?
  • Is it approving or rejecting something?
  • Is it communicating externally?
  • Is it creating legal, financial, operational, customer, employee, patient, citizen, market, or reputational consequence?

Classify the agent as:

  • Informational Assistant
  • Decision Support Copilot
  • Workflow Preparation Agent
  • Bounded Execution Agent
  • Consequential Execution Agent
  • Observer / Governance Agent
  • Escalation / Control Agent
Step 2 — Define the Delegation Chain

The charter should never say “the agent is authorized” without identifying who or what authorized it.

Document:

  • principal delegating authority;
  • source of delegation;
  • approving body or person;
  • effective date;
  • expiration or review date;
  • conditions of authority;
  • unresolved authority gaps.

If authority is unclear, state:

“Not ready — authority source not established.”

Step 3 — Separate Capability from Authority

The agent may be technically capable of many actions. Only some actions are institutionally authorized.

Create a table with three columns:

Technical CapabilityAuthorized UseRequired Control

Example:

Technical CapabilityAuthorized UseRequired Control
Send emailDraft only, no autonomous sendHuman approval required
Update CRMAdd internal note onlyLog entry required
Approve refundUp to approved threshold onlyEvidence pack required
Deny claimNot authorized unless specifically approvedHuman decision required
Step 4 — Assign Authority Tiers

Classify each action into Tier 0 through Tier 4.

Use conservative classification when the facts are unclear.

Any action involving legal, financial, medical, employment, credit, insurance, public-sector, customer-harm, or external commitment consequence should default to Tier 3 or Tier 4 unless the user provides a specific approved threshold and authority source.

Step 5 — Define Human Approval Rules

For each action requiring human approval, define:

  • approver role;
  • approval method;
  • approval record;
  • required rationale;
  • timeout rule;
  • fallback if approver is unavailable;
  • whether approval applies to one action, one case, a batch, or a time-bound operating period.
Step 6 — Define Escalation Logic

The charter must specify when the agent must stop and escalate.

Escalation triggers may include:

  • missing authority;
  • ambiguous policy basis;
  • conflicting instructions;
  • unusual transaction size;
  • customer harm risk;
  • legal or regulatory consequence;
  • financial threshold exceeded;
  • sensitive personal data;
  • protected class or discrimination risk;
  • evidence gap;
  • model uncertainty;
  • tool failure;
  • contradictory records;
  • suspected fraud;
  • security concern;
  • request to override a control;
  • repeated failed attempts;
  • novel fact pattern;
  • external commitment risk;
  • pending litigation or complaint;
  • regulator inquiry;
  • adverse customer outcome.

Write escalation rules in operational language.

Good example:

“The agent must escalate if the requested refund exceeds the approved threshold, if the customer account has an open dispute, or if the agent cannot identify a policy basis for the action.”

Weak example:

“The agent should escalate when the matter is risky.”

Step 7 — Define Evidence Requirements

For each action tier, define minimum evidence.

Suggested baseline:

TierMinimum Evidence
Tier 0Timestamp, request, source records, generated output
Tier 1Timestamp, request, source records, draft or recommendation, reviewer identity
Tier 2Timestamp, authority source, constraints applied, action taken, affected record, audit log
Tier 3Tier 2 evidence plus human approval, approval rationale, escalation history
Tier 4Prohibited action log, denial reason, escalation record

Evidence should include:

This is the start of the file. Read the full SKILL.md on GitHub.

Security

  • Security scanPassed, 100 of 100Pattern checks on the skill's files.
  • Model reviewNot runA model review is scheduled for the next rescan.

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

Repository stars
689
Repository forks
88
7-day trend score
33

Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/lawve-ai/awesome-legal-skills/agent-authority-charter-builder-arkadiy-miteiko/badge)](https://verified-skill.com/skills/lawve-ai/awesome-legal-skills/agent-authority-charter-builder-arkadiy-miteiko)

Something wrong with this skill or its page? Tell us.