Skip to content
Back to skills

Flux Safe

ASecurity

What makes an object wrong only under a reconciler (Flux) — a Job with ttlSecondsAfterFinished, a Job with no spec hash in its name, force on a kind that holds data — the `flux-safe` lint, what the bundle build fixes for you, and the per-tree budget it records.

  • 10 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added November 13, 2026
ai-agentsgo

Security analysis

A100/100

Scanned November 13, 2026

npx -y skills add reliant-labs/forge --skill flux-safe --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Flux Safe?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Flux Safe
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/reliant-labs-flux-safe/badge)](https://www.skillsdirectory.com/skills/reliant-labs-flux-safe)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: deploy/flux-safe
description: What makes an object wrong only under a reconciler (Flux) — a Job with ttlSecondsAfterFinished, a Job with no spec hash in its name, force on a kind that holds data — the `flux-safe` lint, what the bundle build fixes for you, and the per-tree budget it records.
---

# Flux-safe bundles

`forge env deploy` applies once and exits. Flux re-applies the same bytes on every
interval and prunes what the bundle stops carrying, so some objects that are fine
under `kubectl apply` are wrong under it. `forge lint` (`flux-safe`) and the bundle
build check them.

## Refused (you must fix it)

| Rule | Why it is wrong under Flux | Fix |
|---|---|---|
| `flux-job-ttl` — `ttlSecondsAfterFinished` on a Job | the TTL controller deletes the finished Job, the next reconcile creates it again: a migration re-runs every period | remove it. A superseded Job goes away by its spec-hash name plus prune, never by a timer (Flux's inventory prune; a direct apply keeps the newest 3 finished Jobs of each name and deletes the older ones after the apply) |
| `flux-job-unhashed` — a Job name not ending in `-<6+ hex>` | a Job's spec is immutable, so after any change every reconcile fails instead of creating a new Job | name it `<name>-<hash of its spec>`; a forge `kind = "job"` workload already is |
| `flux-force-on-data` — `kustomize.toolkit.fluxcd.io/force: enabled` on a PVC, PV, StatefulSet, Secret, Namespace or CRD | Flux would delete and re-create it on an immutable-field change, losing its data | remove it; change the field through a migration |

These are **errors** for an object in a delivery unit a reconciler applies and
**warnings** for one forge applies itself, so a build that works today keeps
working. That is the whole env when it declares no `lifecycle`, no `applier` and no
`hub_converges`; only the listed units when it declares `hub_converges`; and
nothing when it declares `lifecycle = "local"`/`"ephemeral"`. A warning is the
to-do list for the move onto Flux: fix it before listing the unit in
`hub_converges`, which is when it becomes an error.

## Done for you (nothing to change)

- A Deployment's `strategy` (`RollingUpdate`, 25%/25%) and `progressDeadlineSeconds`
  (600) are rendered explicitly, for forge workloads and for a Deployment you wrote
  in `forge.Manifests`. Values you wrote are kept; `Recreate` never gains
  `rollingUpdate`.
- A `StorageClass`, `RuntimeClass`, `PriorityClass`, `ClusterRoleBinding` or
  `RoleBinding` gets `kustomize.toolkit.fluxcd.io/force: enabled`, so Flux can
  recreate it on an immutable-field change as direct apply does. A `Service`, a
  Deployment and a Job are not stamped (a Service releases its LoadBalancer
  address); annotate those yourself if you mean it.

- A StatefulSet with `volumeClaimTemplates` gets a pre-rollout **resize Job** (a
  hash-named, TTL-free Job in the unit's `pre` tree, so it passes the rules above) that
  grows its claims when the declared size grew. Flux dry-runs every object of a tree
  before applying any, so without it one raised size would wedge the whole unit:
  `deploy/statefulset-volumes`.

## Budgets

Each tree in the bundle records `budget_seconds`: the largest `progressDeadlineSeconds`
(rollout trees) or `activeDeadlineSeconds` (Job trees, 600 s if unstated) among the
objects it is judged on, plus 120 s. Flux itself only waits one short attempt; the
control plane fails a tree when this budget is spent. Declare `activeDeadlineSeconds`
on a slow migration so its tree is given the time it needs.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

View all in ai-agents →

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.