OCI Skills — Administração Oracle Cloud
adibirzu/oci-skills
Administração OCI segura e reutilizável para agentes de IA, sem dados sensíveis
Ver no GitHubSkill pack tenancy-agnostic para administração Oracle Cloud Infrastructure com nove domínios (IAM, Security, Observability, Networking, Compute, Cost, Log Analytics, Data Safe, Eventos/Functions). Implementa preflight de tenancy, dry-run/confirm e redação de segredos para operações seguras em agentes de código como Claude e Gemini.
Instalação
npx convez-ai adibirzu/oci-skills1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Automação de tarefas administrativas OCI (IAM, políticas, compartimentos)
- Governance e compliance em tenancies OCI (Cloud Guard, auditoria, CIS scanning)
- Orquestração de infraestrutura e ciclo de vida de projetos em OCI
- Consultas e análise de logs com OCI Log Analytics
- Monitoramento, custos e alertas em ambientes OCI
Veredito

Skill pack maduro e bem estruturado para administração OCI em escala, com atenção genuína a segurança (preflight, redação, dry-run). Excelente para equipes que usam agentes de IA para IaC e governance, mas requer conhecimento prévio de OCI e CLI.
README
A tradução para o português deste README ainda está na fila.
oci-skills — OCI Administrator skill pack
A tenancy-agnostic Oracle Cloud Infrastructure (OCI) administration skill pack for AI coding agents. One safety-first knowledge core, nine admin domain skills, packaged for Claude Code, Codex, Gemini CLI, and Antigravity.
Built to be reused in any tenancy. It ships no OCIDs, IPs, keys, or tenancy data — only generic command patterns and
<PLACEHOLDER>tokens you resolve at runtime from your own environment.
New here? → docs/QUICKSTART.md — install, bind a named context, preflight, and run the read-only "what's going on?" loop in five minutes.
Why
OCI administration knowledge tends to get copy-pasted across scripts: the same
oci CLI auth negotiation, the same "check the service limit first", the same
"is the WAF rule in OBSERVE or BLOCK?" gotchas. This pack centralizes those into
one reusable core plus nine domain skills, with a hard rule that nothing
sensitive is ever printed or committed.
Domains
| Plugin | Covers |
|---|---|
| oci-iam-admin | Users, groups, dynamic groups, policies (least-privilege review), compartments, budgets, quotas, service limits, tags, Identity Domains. |
| oci-security-compliance | Cloud Guard, Vault/KMS, Security Zones, WAF, Audit, CIS / ISO-42001 / sovereignty scanning, IAM policy review, secret redaction. |
| oci-observability-db | Monitoring & alarms, Logging, Log Analytics, APM (traces/RUM), Notifications, Service Connector Hub, Database Management, Operations Insights, Autonomous DB. |
| oci-networking-compute | VCN, subnets, NSGs, route tables, gateways, load balancers, OKE, compute instances, OCIR. |
| oci-cost | Cost & usage reporting (Usage API: spend by service/compartment/region/tag), budgets (limit vs actual vs forecast), cost-tracking tags, guardrail recommendations. |
| oci-log-analytics | OCI Log Analytics (Logan): the OCL query language, a read-only query helper, sources/parsers/fields/entities/log groups, detections (incl. Sigma→OCL), saved/scheduled searches, dashboards, content migration. |
| oci-resource-manager | Resource Manager (managed Terraform): stacks, plan/apply/destroy jobs, job logs/state, drift detection, state import, variables, and schema.yaml stack packaging. |
| oci-data-safe | Data Safe: target-database registration (ADB + cloud DB), private endpoints, Security/User Assessment, Activity Auditing, Data Discovery, Data Masking. |
| oci-events-functions | Event-driven & serverless: OCI Functions (deploy/invoke/config), Events rules (eventType → FAAS/ONS/STREAMING), Notifications/ONS, Service Connector Hub fan-out, Streaming transport. |
| oci-project | Project lifecycle orchestrator (above the nine domains): bootstrap/scaffold a project (compartment + scoped IAM + network + budget + tags), project status/health, deploy/release (ORM/OKE), and gated teardown — scoped to one project compartment via a named context. |
Scope & related. This pack is the default entry point for OCI tenancy administration — broad infrastructure and control-plane work across nine domains, gated by the safety core. It is complementary to the official oracle/skills collection, which goes deep on a few capabilities. Catch the request here (tenancy preflight + redaction + destructive-op guard), then hand off: deep OKE day-2 (GVA, Multus, troubleshooting) →
oci/oke; OCI Generative AI / Enterprise AI →oci/enterprise-ai; inside an Oracle Database (SQL/PL/SQL, RMAN, AWR/ASH, migrations, Data Guard) →db/. We own the OCI services around the database (DBM, OPSI, Data Safe, ADB provisioning). Full routing contract — coverage matrix, hand-off rules, shared conventions — in references/oracle-skills-alignment.md.
Architecture
A request enters through the router (oci-administrator), is routed by intent
to one of nine domain skills, and every CLI call funnels through one shared
safety core (scripts/common.sh) before it ever reaches the tenancy. The same
core is installed, unchanged, into each agent harness.
flowchart TD
U([User / agent request]) --> R{{"oci-administrator<br/>router skill"}}
R -->|route by intent| D
subgraph D[Nine domain skills]
direction LR
IAM[oci-iam-admin]
SEC[oci-security-compliance]
OBS[oci-observability-db]
NET[oci-networking-compute]
COST[oci-cost]
LOG[oci-log-analytics]
ORM[oci-resource-manager]
DS[oci-data-safe]
EF[oci-events-functions]
end
D --> CORE
subgraph CORE["Shared safety core — scripts/common.sh"]
direction LR
CLI["oci_cli<br/>one auth path"]
MUT["run_mutating / confirm<br/>gate mutations"]
RED["redact.py<br/>mask OCIDs / IPs / secrets"]
CTX["oci_context.py<br/>named contexts"]
end
HOOK[["PreToolUse guard hook<br/>blocks delete / terminate / destroy"]] -. guards .-> CLI
CORE --> OCI[("OCI tenancy<br/>via OCI CLI / SDK")]
R -. installed into .-> H
subgraph H[Harness adapters]
direction LR
C[Claude Code]
CX[Codex]
G[Gemini CLI]
AG[Antigravity]
end
Progressive disclosure keeps it simple: an agent reads the router, then one
domain SKILL.md, then that domain's references/*.md only if it needs depth — it
never loads all nine domains at once. Each layer is one short file.
Safety model
Every plugin inherits the same contract (see
references/tenancy-safety.md):
- Preflight — confirm which tenancy/compartment before any change
(
scripts/oci_preflight.sh), shown by name, never raw OCID. - Read before write, idempotent (
409 Conflict= "already exists"). - Destructive ops gated by
confirm/run_mutating; honorOCI_SKILLS_DRY_RUN=trueandOCI_SKILLS_FORCE=true. - Never emit secrets —
scripts/redact.pymasks OCIDs, IPs, fingerprints, install keys, private-key blocks, and token blobs (used as a CI gate). - One auth path —
oci_cliinscripts/common.shnegotiates config / instance / resource / OKE-workload / security-token auth.
Install
As a Claude Code plugin (slash commands + safety hook)
/plugin marketplace add adibirzu/oci-skills # or: /plugin marketplace add ~/dev/oci-skills
/plugin install oci-administrator@oci-skills
This gives you the slash commands below plus a PreToolUse hook that blocks
destructive oci commands until they are preflighted and confirmed:
| Command | Does |
|---|---|
/oci-administrator:context | Manage named contexts (name → profile + compartment + region). |
/oci-administrator:preflight | Confirm the target tenancy/compartment by name (read-only). |
/oci-administrator:audit | Read-only IAM posture snapshot. |
/oci-administrator:cost | Read-only cost, usage & budget summary. |
/oci-administrator:logan | Read-only Log Analytics (OCL) query with a time window. |
/oci-administrator:orm | Read-only Resource Manager overview (stacks + latest job). |
/oci-administrator:datasafe | Read-only Data Safe overview (targets + assessment). |
/oci-administrator:kb | Search the KB for a known fix. |
/oci-administrator:troubleshoot | KB-first, route to domain, propose a gated fix. |
The copy-install paths below additionally deliver the auto-triggering knowledge skills and the multi-harness adapters (Codex, Gemini, Antigravity).
One line (recommended)
Installs into every agent harness it detects (Claude Code, Codex, Gemini CLI, Antigravity):
curl -fsSL https://raw.githubusercontent.com/adibirzu/oci-skills/main/bootstrap.sh | bash
Pick specific harnesses, or pin a fork/branch:
curl -fsSL https://raw.githubusercontent.com/adibirzu/oci-skills/main/bootstrap.sh | bash -s -- claude codex
OCI_SKILLS_REF=main curl -fsSL https://raw.githubusercontent.com/adibirzu/oci-skills/main/bootstrap.sh | bash
bootstrap.sh clones (or fast-forwards) the repo into
~/.local/share/oci-skills, then runs the installer. Re-run it any time to
update. Requires git.
From a clone
git clone https://github.com/adibirzu/oci-skills.git && cd oci-skills
make install # install into every detected harness
make list # show detected harnesses
make install-claude # or a single harness: -claude / -codex / -gemini / -antigravity
make dry-run # preview, copy nothing
# equivalently, the installer directly:
./install.sh # every detected harness
./install.sh claude codex # pick specific ones
DRY_RUN=true ./install.sh # preview
Install targets (override with env vars — see install.sh header):
| Harness | Destination | Adapter |
|---|---|---|
| Claude Code | ~/.claude/skills/oci-administrator/ | SKILL.md |
| Codex | ~/.codex/skills/oci-administrator/ | harness/codex/agents/openai.yaml |
| Gemini CLI | ~/.gemini/extensions/oci-skills/ | harness/gemini/ |
| Antigravity | ~/.antigravity/skills/oci-administrator/ | harness/antigravity/AGENTS.md |
Requirements
bash(3.2+, so the macOS system/bin/bashworks), plus the OCI CLI andjqon PATH for the shell scripts.- Python 3.10+ and the
ociSDK (pip install oci) foriam_audit.py. - A configured
~/.oci/configprofile, or instance/resource/workload-identity auth when running inside OCI.
Repository layout
.claude-plugin/ # Claude Code plugin manifest + marketplace entry
plugin.json marketplace.json
SKILL.md # Claude Code entrypoint (router)
AGENTS.md # Codex / Antigravity entrypoint (mirror)
commands/ # Claude Code slash commands (context/preflight/audit/cost/logan/orm/datasafe/project/kb/troubleshoot)
hooks/ # PreToolUse guard that blocks destructive oci commands
hooks.json guard_destructive.py
references/ # domain + safety knowledge (progressive disclosure)
tenancy-safety.md agent-safety.md oci-error-catalog.md
oracle-docs.md # verified docs.oracle.com source-of-truth index
oracle-skills-alignment.md # routing contract vs the official oracle/skills repo (deep OKE / GenAI / in-DB)
solution-authoring.md # Stage 0 design: requirement → guardrailed architecture → blueprint
project-workflow.md # oci-project lifecycle index → project-phase{1-4}-*.md
project-phase1-bootstrap.md project-phase2-status.md
project-phase3-deploy.md project-phase4-teardown.md
mcp-gateway.md # the non-official oci-mcp-gateway (optional read surface; not an Oracle product)
helper-conventions.md KB.md named-contexts.md
credential-management.md
iam-tenancy.md security-compliance.md observability-db.md networking-compute.md
cost-management.md log-analytics.md resource-manager.md data-safe.md events-functions.md
scripts/ # shared core
common.sh oci_context.py oci_preflight.sh oci_cost.sh oci_logan.sh oci_orm.sh oci_datasafe.sh
oci_project.sh oci_cli_help.py redact.py iam_audit.py kb_lookup.py check_doc_links.py
skills/ # eleven auto-discoverable skills (router + nine domains + project orchestrator)
oci-administrator/ oci-iam-admin/ oci-security-compliance/
oci-observability-db/ oci-networking-compute/ oci-cost/ oci-log-analytics/
oci-resource-manager/ oci-data-safe/ oci-events-functions/ oci-project/
harness/ # per-harness adapters (codex / gemini / antigravity)
evals/evals.json # trigger + behavior evals
bootstrap.sh # one-line remote installer (curl | bash)
install.sh # multi-harness installer
Makefile # make install / list / dry-run
Quick start
# 0. (Optional) register a friendly context so you never paste OCIDs again
./scripts/oci_context.py add dev --profile DEFAULT --region eu-frankfurt-1 \
--compartment <COMPARTMENT_OCID> --description "sandbox"
eval "$(./scripts/oci_context.py use dev)" # sets profile/region/compartment
# 1. Confirm you are pointed at the right tenancy (read-only)
./scripts/oci_preflight.sh -c "${OCI_SKILLS_COMPARTMENT:-<COMPARTMENT_OCID>}"
# 2. Read-only IAM posture snapshot
python3 ./scripts/iam_audit.py --profile "${OCI_CLI_PROFILE:-DEFAULT}"
# 3. Look up a known fix before debugging
python3 ./scripts/kb_lookup.py "kubectl unauthorized oke"
Security & contributing
This is a public repository. Never add real OCIDs, IPs, fingerprints,
tenancy namespaces, datakeys, or secrets — CI runs gitleaks and
redact.py --check. See CONTRIBUTING.md.
License
MIT.