All skills

Use when the user asks to "find OCI skills", "route Oracle Cloud work", "install the OCI skill pack", "review OCI skill ownership", or "separate Oracle skills".

Use this Skill: https://skilld.dev/gh/acedergren/agentic-tools/oci

This session only. Nothing lands on disk.

iam-identity-managementSKILL.md

≈2.6k tokens on demand. Your agent reads this file only when SKILL.md points to it.

OCI IAM and Identity Management - Expert Knowledge

Do NOT load this skill when

Do not load this skill for unrelated general programming, non-Oracle cloud work, or questions covered by a narrower sibling skill.

When to Use

Load this skill for: the user asks to "write OCI IAM policy", "debug OCI 403", "Terraform apply gets 403", "configure dynamic groups", "use identity domains", or "fix IDCS federation".

Prefer this skill only for its named domain. For broader OCI architecture triage, start with oci/best-practices as the router.

NEVER Do This

NEVER use overly broad policies in production

# WRONG - grants admin to everyone, instant security audit failure
Allow any-user to manage all-resources in tenancy

# RIGHT - explicit group, specific resource, specific compartment
Allow group 'Default'/'AppDevelopers' to manage instance-family in compartment AppDev

Policy conditions support only = and != (plus any {} / all {}). There is no =~; that is MQL. Patterns use slashes: /dev-*/, /*-prod/, /*web*/, and matching is case-insensitive. A quoted 'dev-*' is a literal string, not a wildcard.

NEVER attach a policy below the compartment it grants on

# WRONG - policy attached to A:B:C cannot reference A (outside its scope)
Policy location: Compartment A:B:C
"Allow group X to read buckets in compartment A"   # rejected when you create the policy

# RIGHT - attach at or above the target compartment
Policy location: Compartment A (or the tenancy)
"Allow group X to read buckets in compartment A"

# Attached higher up? Use the compartment path relative to the attachment point:
Policy location: tenancy
"Allow group X to read buckets in compartment A:B"

Policies inherit downward: a grant on A also applies to A:B and A:B:C. The attachment location also decides who can edit the policy — attach it where the right admins own it.

NEVER use any-user in production policies

  • Grants access to ALL future users, including compromised accounts
  • Always fails SOC2/HIPAA/CIS audits
  • Use explicit group membership: Allow group DataReaders to read buckets in compartment SharedData

NEVER grant access to instances using user principal syntax

# WRONG - instances are NOT users
Allow user <instance-ocid> to read buckets in compartment X

# RIGHT - use dynamic groups for instances
Allow dynamic-group app-instances to read buckets in compartment X

NEVER hardcode resource OCIDs in dynamic group rules

# WRONG - breaks when instance is replaced
ALL {instance.id = 'ocid1.instance.oc1.phx.xxxxx'}

# RIGHT - use compartment or defined-tag matching (survives instance replacement)
ALL {instance.compartment.id = '<compartment-ocid>'}
ANY {tag.Operations.Environment.value = 'production'}

Dynamic group rules match defined tags as tag.<namespace>.<key>.value. There is no instance.freeform-tags.* variable.

NEVER forget that a condition variable that doesn't apply to a request denies it

# This grants nothing for ListUsers/ListGroups: target.group.name is not
# present on those requests, so the condition evaluates false.
Allow group GroupAdmins to use groups in tenancy where target.group.name != 'Administrators'

# Add the list permission in a separate, unconditioned statement:
Allow group GroupAdmins to inspect groups in tenancy

target.compartment.name/.id cannot be used to filter List operations either.

IAM Permission Troubleshooting

"404 NotAuthorizedOrNotFound"

This error is intentionally ambiguous — OCI returns 404 whether the resource doesn't exist OR the caller lacks inspect permission. This prevents enumeration attacks.

404 NotAuthorizedOrNotFound?
│
├─ Does the resource definitely exist?
│  ├─ YES → Permission issue
│  │  └─ Does caller have at least 'inspect' on that resource type?
│  │  │  └─ Is policy at or above the target compartment?
│  │  └─ Does a condition reference a variable the operation doesn't carry?
│  └─ NO → Verify OCID, compartment, region
│
├─ Using dynamic group / instance principal?
│  └─ oci compute instance get --instance-id <ocid>  (check compartment + tags)
│  └─ Does instance's compartment/tags match the dynamic group rule?
│
└─ Cross-compartment access?
   └─ Groups live in an identity domain, not a compartment: the policy must be
      attached at or above the *target* compartment (or in the tenancy)

"403 NotAuthorized"

Caller is identified but explicitly lacks permission.

Common causes:

  1. Wrong verb: Policy grants read but action requires use or manage
  2. Wrong resource-type: Granted instance-family but accessing volume-family
  3. Condition doesn't match: a quoted 'prod-*' is literal; use /prod-*/
  4. Identity domain prefix missing: a group in a non-default domain must be written 'DomainName'/'GroupName'
  5. Propagation: new or edited policies can take a short time to take effect; re-test before rewriting

Verb hierarchy (each includes those below it):

inspect < read < use < manage

Terraform and Resource Manager 403s

Before changing HCL, identify the principal and scope:

Symptom Check first
Terraform local apply gets 403 API key user, session-token profile, identity-domain group mapping, and target resource policy
Terraform on Compute gets 403 Instance dynamic group membership and policy on target resource family
Resource Manager stack/job gets 403 orm-* permissions plus target service permissions in the target compartment
Resource Manager dynamic group cannot create VCN Whether the principal is actually a Resource Manager/user context, and whether manage virtual-network-family is granted where the VCN is created
Identity-domain group can log in but cannot apply Group mapping, exact group name, policy subject, compartment policy location
ZPR policy or security attributes cannot be managed ZPR IAM permissions plus policy location and compartment scope
Bastion session creation is denied Bastion/session permissions plus target resource/network permissions

Policy Syntax Gotchas

Resource Type Families (Often Confused)

Family Includes Common Mistake
instance-family instances, instance-images, instance-console-connection, console-histories, app-catalog-listing, volume-attachments (attach only) Does NOT include VNICs/subnets (virtual-network-family) or volumes
volume-family volumes, volume-attachments, volume-backups, … Separate from instance-family
object-family objectstorage-namespaces, buckets, objects Objects are a separate resource type from buckets
database-family Base Database / Exadata DB systems, homes, databases, backups Does NOT cover Autonomous DB — use autonomous-database-family (autonomous-databases, autonomous-backups)

LaunchInstance also needs use vnics, use subnets and use network-security-groups in the network compartment.

Conditions (WHERE clause)

# Defined-tag based (freeform tags are NOT usable in policy conditions)
where target.resource.tag.Operations.Environment = 'production'
where request.principal.group.tag.Operations.Project = 'alpha'

# Pattern match: slashes, not quotes, not =~
where target.bucket.name = /logs-*/

# Request properties
where request.operation = 'LaunchInstance'
where request.region = 'ARN'          # 3-letter region key (ARN = Stockholm)

# Combined conditions
where all {target.resource.tag.Operations.Environment = 'prod', request.permission != 'BUCKET_DELETE'}
where any {request.user.name = 'alice', request.user.name = 'bob'}

Tag-conditioned manage does not cover List or Create (the resource has no tag yet): add a separate inspect statement and create in an untagged scope, or tag via tag defaults.

Location Syntax

in compartment <name>                  # Compartment (and, by inheritance, its subcompartments)
in compartment id <compartment-ocid>   # Same, by OCID
in compartment A:B                     # Path, when the policy is attached above A
in tenancy                             # Whole tenancy

Dynamic Group Patterns

By compartment (most common — covers all current and future instances):

ALL {instance.compartment.id = '<compartment-ocid>'}

By defined tag (flexible — survives instance replacement):

ANY {tag.Operations.App.value = 'webserver'}

Restrictive AND rule (production workloads):

ALL {instance.compartment.id = '<comp-ocid>', tag.Operations.Environment.value = 'production'}

Non-instance resources (functions, etc.) match with resource.type, resource.id, resource.compartment.id.

Testing Dynamic Group Membership

# 1. Get instance details (compartment + tags)
oci compute instance get --instance-id <instance-ocid>

# 2. Check dynamic group rule
oci iam dynamic-group get --dynamic-group-id <group-ocid>

# 3. Verify rule matches — if rule is "instance.compartment.id = X",
#    confirm the instance's compartment_id field equals X

# 4. Test from the instance (SSH in and run):
oci os ns get  # Works only if instance principal is correctly configured

Authentication Methods

Method Use Case Key Constraint
API Key Local dev, CI/CD outside OCI Manual rotation required
Instance Principal Apps on OCI compute Only works on OCI compute
Resource Principal OCI Functions, Data Flow Limited to specific services
Session Token oci session authenticate (browser/federated login) Short-lived; refresh with oci session refresh

Identity Domain Gotchas

  • Groups live in identity domains. For the Default domain you may write group Admins; for any other domain you must write group 'DomainName'/'GroupName' (and dynamic-group 'DomainName'/'DgName'). Oracle recommends prefixing 'Default'/ anyway for readability.
  • User can log in but sees nothing → the domain group has no policy, or the policy omits the domain prefix.
  • External IdP (SAML/OIDC) users get OCI access through domain group membership (JIT provisioning or SCIM sync); check the user's group membership inside the domain before touching policies.
  • Legacy tenancies not yet migrated to identity domains used IDCS federation with explicit IDCS→OCI group mappings; that model does not apply to identity-domain tenancies.

Compartment Hierarchy Design

# WRONG: Flat structure — no IAM boundary between dev/prod
Tenancy
├─ Application1  (mix of dev/test/prod resources)
└─ SharedServices

# RIGHT: Environment-based hierarchy — clear blast radius, cost reporting
Tenancy
├─ Production
│  ├─ App1
│  └─ App2
├─ Development
│  ├─ App1
│  └─ App2
└─ SharedServices
   ├─ Networking
   └─ Security

Progressive Loading Reference

Load references/oci-iam-policies-reference.md for the general condition-variable table, pattern syntax, and known pitfalls. It links, rather than copies, the per-service verb/permission tables — open the service's policy reference page for those.

Last verified: 2026-09-30 (docs.oracle.com: Policy Syntax/Subjects, Conditions, General Variables, Policy Inheritance, Managing Dynamic Groups, Core Services and ADB policy references)

Load ../infrastructure-as-code/references/oci-terraform-auth-matrix.md when Terraform, OCI DevOps, Resource Manager, Compute instance principals, resource principals, OKE workload identity, Cloud Shell, or CI/CD federation affect the caller.

Load ../zpr-security/references/zpr-reference.md when the IAM task is specifically about who may configure ZPR, security attributes, or ZPR policies.

Load ../managed-bastion-access/references/managed-bastion-reference.md when the IAM task is specifically about who may create bastions, create sessions, update allowlists, or connect through OCI Bastion.

Arguments

$ARGUMENTS: Optional user-provided target, path, environment, symptom, or constraint. When empty, infer the narrowest safe scope from the current repository context and ask only if multiple high-impact choices remain.

Source: SKILL.md on GitHub

1 warning3mo3 checks · Risk MEDIUM
  • Gen Agent Trust Hub3mo

    An expert-level OCI utility pack that includes advanced scripts for presentation management and database administration. Security concerns include runtime compilation of C shims for system call interception and a powerful SQL script for tenancy-wide security remediation. The skill also ingests untrusted data from document archives and web headers, creating a surface for indirect prompt injection.

  • Socket3mo

    No alerts

  • Snyk3mo

    Risk: LOW · No issues

Signed by skilld at d0e87b8. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 2 months ago.

Steadyupdated 4 months ago
version
1.0.0
aliases
[
  "oci-skills",
  "oracle-skills",
  "oci-skill-pack"
]
domains
[
  "oci",
  "oracle",
  "skill-pack"
]
Other metadata
keywords
[
  "OCI",
  "Oracle Cloud",
  "Oracle",
  "skill pack",
  "skill routing",
  "separation of duties",
  "ownership",
  "architecture",
  "operations",
  "manifest"
]

README badge

README badge for acedergren/agentic-tools/oci