Index of every page on this site: llms.txt
All skills
acedergren avatar

/compute-management

@641a79b
Run it yourself

Use when the user asks to "launch OCI compute", "choose an OCI shape", "debug compute capacity", "configure instance principals", or "optimize OCI instance cost".

  • 2 files
  • 9.2 KB
  • Updated 2 months ago
  • GitHub
Use this Skill: https://skilld.dev/gh/acedergren/agentic-tools/compute-management

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈46 tokens for metadata: the name and description. ≈2k when used: this file. ≈203 more on demand in 1 file.

Description uses 2.3% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈183 characters, or 2.3%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

OCI Compute 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 "launch OCI compute", "choose an OCI shape", "debug compute capacity", "configure instance principals", or "optimize OCI instance cost".

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

When the compute task involves private operator access, Managed SSH, or the Bastion plugin, also load oci/managed-bastion-access.

NEVER Do This

NEVER launch instances without checking service limits first

oci limits resource-availability get \
  --service-name compute \
  --limit-name "standard-e4-core-count" \
  --compartment-id <ocid> \
  --availability-domain <ad>

Many "out of capacity" launch failures are quota or service-limit problems rather than host-capacity problems. Check limits BEFORE launching to get accurate error messages.

NEVER use console serial connection as primary access

  • Creates security audit findings (bypasses SSH key controls)
  • Use only for boot troubleshooting when SSH fails
  • Delete connection immediately after troubleshooting

NEVER mix regional and AD-specific resources in templates

  • Breaks portability when moving between regions
  • Use AD-agnostic designs: spread via fault domains, not hardcoded ADs

NEVER terminate without deciding the boot volume's fate

# Default: boot volume is deleted (--preserve-boot-volume defaults to false)
oci compute instance terminate --instance-id <id> --force
# Keep it deliberately (and track it — it keeps billing)
oci compute instance terminate --instance-id <id> --preserve-boot-volume true --force

NEVER enable public IP on production instances

  • Use bastion service or private endpoints for access
  • Treat public exposure as a security and incident-response cost driver, not only as a network setting.

NEVER use fixed shapes when Flex covers the same specs

  • Fixed shapes (e.g., VM.Standard2.1) are often MORE expensive than Flex equivalents
  • Always compare Flex pricing first before choosing a fixed shape

Capacity Error Decision Tree

"Out of host capacity for shape X"?
│
├─ Check service limits FIRST
│  └─ oci limits resource-availability get
│     ├─ available = 0 → Request limit increase (NOT a capacity issue)
│     └─ available > 0 → True capacity issue, continue below
│
├─ Same shape, different AD?
│  └─ Query the tenancy's ADs and try each viable AD in the region
│
├─ Different shape, same series?
│  └─ E4 failed → try E5 (newer gen, often more available)
│  └─ Standard failed → try Optimized or DenseIO variants
│
├─ Different architecture?
│  └─ AMD → ARM (A1.Flex often has capacity when Intel/AMD full)
│
└─ All ADs exhausted?
   └─ Create capacity reservation (guarantees future launches)

Terraform plans do not prove compute capacity or quota availability. Before applying compute resources, check live service limits, regional availability, shape availability, and compartment quotas; then treat "out of host capacity" separately from service-limit exhaustion.

Shape Selection: Cost vs Performance

Budget-critical:

  • Compare Arm flexible shapes against current AMD/Intel alternatives in the Oracle price list.
  • Test ARM64 compatibility first; not all Docker images, compiled binaries, agents, or vendor packages support Arm.

General purpose:

  • Prefer flexible shapes when they meet the workload requirement so CPU and memory can be right-sized separately.
  • Start from measured CPU, memory, network, and storage I/O needs; scale based on metrics, not guesses.

Memory-intensive:

  • Compare flexible shape memory ratios against the workload's actual resident set, cache, and JVM/DB memory needs.
  • Verify the current memory-per-OCPU limits in Oracle shape documentation before promising a target ratio.

Live Cost Check

Do not quote static shape prices from this skill. Before estimating cost:

  1. Open the current Oracle Cloud price list for the target region, currency, and subscription model.
  2. Confirm the shape family, architecture, OCPU/ECPU terms, memory billing unit, and any Free Tier eligibility in current Oracle docs.
  3. Calculate with live unit rates: monthly_compute = hours * ((ocpus * live_ocpu_rate) + (gb_memory * live_memory_rate))
  4. Add non-compute resources separately: boot volume, block volumes, public IPs, load balancers, backups, images, network transfer, and support commitments.
  5. State the date and source of the price lookup whenever giving numbers.

Instance Principal Authentication (Production)

When an instance needs to call OCI APIs, NEVER put user credentials on the instance.

# 1. Create dynamic group
oci iam dynamic-group create \
  --name "app-instances" \
  --matching-rule "instance.compartment.id = '<compartment-ocid>'"

# 2. Grant permissions (IAM policy)
# "Allow dynamic-group app-instances to read object-family in compartment X"

# 3. Code uses instance principal — no credentials needed:
signer = oci.auth.signers.InstancePrincipalsSecurityTokenSigner()
client = oci.object_storage.ObjectStorageClient(config={}, signer=signer)

Benefits: No credential rotation, no secrets to manage, automatic token refresh.

OCI-Specific Gotchas

Availability Domain names are tenant-specific

  • Your AD: fMgC:US-ASHBURN-AD-1
  • Another tenant: ErKW:US-ASHBURN-AD-1
  • MUST query your own tenant: oci iam availability-domain list
  • Hardcoding AD names from examples breaks cross-tenant portability

Boot Volume backups do NOT include instance config

  • Backup captures disk only — NOT shape, networking, or metadata
  • Custom images also capture only the boot disk (plus image capabilities), not shape, VNICs or NSGs
  • For DR: rebuild instances from Terraform, restoring boot/block volumes from backups or replicas

Instance Metadata Service: use v2 and disable the legacy endpoints

  • v1: http://169.254.169.254/opc/v1/ (legacy, no header required — SSRF-prone)
  • v2: http://169.254.169.254/opc/v2/ requires the header Authorization: Bearer Oracle
  • Disable v1: oci compute instance update --instance-id <id> --instance-options '{"areLegacyImdsEndpointsDisabled": true}' (check that Oracle Cloud Agent and your tooling already use v2 first)

Last verified: 2026-09-30 (OCI CLI 3.94.1; docs.oracle.com Instance Metadata Service, Resource Billing for Stopped Instances)

Progressive Loading Reference

Load references/oci-compute-shapes-reference.md for links to the official shape, launch, and instance-principal pages (it is a source map, not a spec table). Query live shapes with oci compute shape list -c <compartment-ocid>.

Do NOT load for quick cost comparisons, capacity troubleshooting, or shape selection — this file covers those.

Load ../infrastructure-as-code/references/oci-terraform-auth-matrix.md when Terraform runs on OCI Compute and should use instance principals rather than local API keys.

Load ../infrastructure-as-code/references/oci-terraform-realms-regions.md when shape availability, government regions, FIPS, or realm-specific endpoints may affect Terraform.

Load ../managed-bastion-access/references/managed-bastion-reference.md when diagnosing Managed SSH, Oracle Cloud Agent, Bastion plugin, Ubuntu/Ampere plugin issues, or private instance access.

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

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No third-party reports yet.

Signed by skilld at 641a79b. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub last month.

Steadyupdated 2 months ago
version
2.0.0
aliases
[
  "oci-compute",
  "compute-shapes"
]
domains
[
  "oci",
  "compute"
]
Other metadata
keywords
[
  "OCI",
  "Oracle Cloud",
  "Compute",
  "instance",
  "shape",
  "OCPU",
  "ECPU",
  "boot volume",
  "capacity",
  "instance principal",
  "Terraform",
  "quota",
  "service limits",
  "Managed SSH",
  "Bastion plugin"
]

README badge

README badge for acedergren/agentic-tools/compute-management