Skip to content
Get packs

Guide · Agent skill

How to use the Access Review Assistant agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

The Access Review Assistant is an agent skill, a folder of instructions that teaches an AI agent a structured process for auditing who has access to systems such as AWS IAM, GitHub organizations and Okta. This guide explains what the skill does, how to install it, how to ask for help with natural requests, and what safety limits to respect. By the end you will be able to run a planned access review with the agent while keeping every revocation decision in human hands.

Access Review Assistantsickn33/agentic-awesome-skills · MIT

Quick start

  1. 1Download and unzip the skill. Download the access-review ZIP from this page and unzip it so the folder contains SKILL.md and references/details.md.
  2. 2Place it where the agent can find it. Copy the folder to ~/.claude/skills/ for personal use or to .claude/skills/ inside a project. You can also upload it in the Claude apps under Settings, Capabilities, Skills.
  3. 3Prepare read-only access. Create read-only, scoped credentials for each system in scope, such as GitHub CLI access, AWS read-only credentials or a read-only Okta token.
  4. 4Define the review scope. Ask the agent to help you confirm the accounts, review period, owners and deadlines, using the skill's scope stage.
  5. 5Collect inventory privately. Run the inventory scripts yourself after reading them, and keep output in a restricted folder.
  6. 6Have managers certify access. Send flagged items to managers to approve, modify or revoke, and track non-responses against the deadline.
  7. 7Document and report. Record exceptions with justification, confirm changes with system owners, and ask the agent to draft completion metrics for your audit file.

What the access-review skill makes the agent do

The skill sets out a five-stage access review workflow: scope, extract, review, remediate and report. In the scope stage the agent helps define systems in scope, name review owners and set deadlines. The extract stage covers pulling current access data, correlating identities across platforms and flagging inactive, over-privileged or orphaned accounts.

The review stage has managers certify each user's access as approve, modify or revoke, with privileged users reviewed first and non-responses escalated after the deadline. The remediate and report stages cover documenting exceptions, confirming changes and producing completion metrics and evidence for audits.

  • Suggested frequencies: privileged access and service accounts quarterly, standard access semi-annually, API keys monthly.
  • Suggested SLAs: revocations within 5 business days of a decision, modifications within 10 business days.
  • The skill itself labels these cadences and SLAs as examples, not compliance certification.

Install the access-review skill

Download the ZIP for the skill from this directory page and unzip it. The skill folder contains SKILL.md, the main instructions, and references/details.md, which holds extended reference material.

For a personal setup, place the unzipped folder in ~/.claude/skills/. To limit the skill to one project, place it in that project's .claude/skills/ folder instead. Alternatively, in the Claude apps, open Settings, then Capabilities, then Skills, and upload the skill there.

  • Personal use: ~/.claude/skills/access-review
  • Single project: .claude/skills/access-review inside the project

Requirements and what to prepare before a review

The skill itself needs no privileged tooling, but the scripts it includes need specific access. The GitHub script requires the GitHub CLI (gh) with read-only organization access. The AWS report-parsing example requires Python 3 and AWS read-only credentials for IAM reports. The Okta section requires an Okta API token scoped to read-only.

Before any data is collected, the skill asks you to confirm the exact tenant or accounts, the read-only credentials, the review period and the authorized reviewers. Reports should be stored privately with a defined retention period.

  • GitHub CLI (gh) with read-only organization access, for the GitHub script
  • Python 3, for the AWS report-parsing example
  • AWS read-only credentials, for IAM reports
  • Okta API token scoped read-only, for the Okta section

Trigger the skill with natural requests

You do not need special syntax. Describe the task in plain language and mention access reviews, certifications or the system you care about. The agent should pick up the skill from the description, which covers periodic access reviews, certifications and recertification workflows.

Good starting requests name the system, the goal and the output you want. Give the agent the context it needs, such as whether you are planning a cycle, checking one system or drafting a report.

  • Help me set up a quarterly access certification for our GitHub organization.
  • Walk me through an access review of our AWS IAM users with a list of accounts to flag for manager review.
  • Draft an audit-ready summary of our last Okta access review, including completion metrics.
  • Create a checklist for the semi-annual review of standard access across our identity providers.

Run the GitHub inventory script safely

The included GitHub script is a partial organization inventory. It lists members, owners, outside collaborators, repository collaborators, team members, pending invitations and deploy keys, writing each result to files in a dated folder. It sets a restrictive file umask and uses the gh CLI for read calls.

Replace the placeholder organization name before running it. Read the script first, because the skill notes that branch and ruleset enforcement needs a separate authorized read and that a 403 or 404 response is unknown, not proof a branch is unprotected. The skill also notes that site_admin means GitHub site administration, not organization ownership.

  • Set ORG to your organization name before running.
  • Output goes to ./access-review/github/ in a year-month folder.
  • Owner data comes from the admin role filter, not the site_admin field.
gh api orgs/$ORG/members --paginate \ --jq '.[] | [.login, .site_admin] | @csv' \ > "$OUTPUT_DIR/org-members.csv"

Use the AWS and Okta guidance carefully

The AWS example reads an IAM credential report through a read-only client and flags console users with a password but no active MFA. It returns candidates for review rather than making changes. The skill says a missing value means unknown, not zero or inactive, and that policy-name matching is only a hint for effective privilege.

For Okta, the skill asks you to use a secret-aware SDK or client against the approved Okta domain and not to put SSWS tokens in curl arguments. It also requires pagination, since a single limit=200 response is incomplete, and warns that factor enrollment does not prove that MFA is required by the sign-on policy.

  • AWS: an inactivity or stale-key result never revokes a key on its own.
  • Okta: keep names and email inventories restricted.
  • Identity matching: use verified immutable IDs, not display names.
def review_console_mfa(report_path): candidates = [] with open(report_path, newline="", encoding="utf-8") as source: for row in csv.DictReader(source): if row.get("password_enabled") == "true" and row.get("mfa_active") == "false": candidates.append({"user": row["user"], "finding": "review_console_mfa"}) return candidates # Private review data, not automatic disable actions.

Safety rules to follow

The skill is explicit that nothing should be revoked automatically. Owner approval and a recovery or break-glass plan are required before modifying access. Inactivity heuristics never authorize revoking production service accounts, and the skill notes that a wrong revocation can cause outages.

The scripts make network calls to your identity providers, so run them only with credentials you are authorized to use. The skill suggests keeping evidence in an approved, access-controlled store rather than a public issue or an unrestricted folder.

  • Use read-only, scoped credentials only.
  • Do not publish personal access inventories in public issues or pages.
  • Treat script output as private review data that a human must act on.

What people use it for

  • Planning a quarterly certification of privileged access across cloud and code platforms.
  • Building a checklist and sample inventory script for a GitHub organization's members, teams and outside collaborators.
  • Flagging console users without MFA in an AWS credential report for manager review.
  • Drafting an audit-ready summary of completed review decisions and on-time percentages.
  • Structuring an Okta review that lists users, factors, app assignments and admin roles.

Frequently asked questions

How do I install the access-review agent skill?

Download the ZIP from this page and unzip it. Place the folder in ~/.claude/skills/ for personal use or in a project's .claude/skills/ folder, or upload it in the Claude apps under Settings, Capabilities, Skills.

Is the access-review skill free to use?

The skill is published under the MIT licence, as stated in its metadata. Check the licence file linked from the skill's source page for the full terms.

What do I need to run the access-review scripts?

The GitHub script needs the GitHub CLI with read-only organization access. The AWS example needs Python 3 and read-only AWS credentials for IAM reports. The Okta section needs an API token scoped to read-only.

Does the skill revoke access automatically?

No. The skill states that nothing should be revoked automatically, and that owner approval and a recovery plan are required before access is modified. Its scripts return findings for human review.

Why does the GitHub script say site_admin is not an owner indicator?

The skill notes that site_admin refers to GitHub site administration, not organization ownership. Owners are collected with the admin role filter, so read the output carefully before drawing conclusions.

Why does the skill not recognise my users when I match by display name?

The skill says identity joins need verified immutable IDs, because matching by display name gives wrong results. Map accounts by their IDs before comparing systems.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

This skill suits security, IT, platform and compliance teams that want a structured review process, a checklist and read-only sample scripts for AWS, GitHub or Okta. It is a poor fit if you want unattended access changes or a legally validated certification, because the skill itself warns against automatic revocation and treats its cadences as examples.

Download the skill