docs(blog): add distribution post batch

blog/distribution-batch-2026-08.md

@@ -0,0 +1,172 @@
+# Clawdlinux distribution batch
+
+Anonymous drafts for developer and platform-engineering channels.
+
+Use the visual listed with each post. Keep the author field empty.
+Do not claim customer adoption, air-gap certification, or complete end-to-end attestation.
+
+## Post 1: The boundary question
+
+AI agents are moving from demos into production systems.
+
+The hard question is not whether an agent can call a tool.
+
+It is what happens when the tool call is wrong.
+
+Who is the agent acting as?
+What can it reach?
+Who approves the action?
+What record can an auditor verify later?
+
+Kubernetes gives teams the place to answer those questions.
+It does not answer them by itself.
+
+Clawdlinux is an open Kubernetes governance layer for AI agent workloads.
+It adds runtime-neutral workload controls, egress policy generation, and offline audit verification primitives.
+
+The current repository is honest about its limits.
+Some controls are generated but depend on cluster enforcement.
+The full signed artifact path is still a release gate.
+
+That is a better starting point than pretending the problem is solved.
+
+Repo: https://github.com/Clawdlinux/agentic-operator-core
+
+Visual: https://raw.githubusercontent.com/Clawdlinux/agentic-operator-core/main/assets/landing-hero.png
+
+## Post 2: Agent egress is a practical control
+
+Prompt instructions are not a network boundary.
+
+An agent can be told not to call an endpoint.
+A compromised tool can still try.
+
+A declared egress policy gives the cluster a second control point.
+
+The Clawdlinux operator can generate default-deny and allow-list policy objects for managed workloads.
+Cilium FQDN policy templates are available when the cluster supports them.
+
+The policy is not magic.
+The CNI must enforce it.
+The selectors must be correct.
+The workload must declare the destinations it needs.
+
+That is the practical model for Kubernetes AI agent egress policy:
+make the allowed boundary explicit, then verify the cluster enforces it.
+
+Docs: https://clawdlinux.org/products/operator
+
+Visual: Mermaid diagram showing AgentWorkload -> admission -> policy -> sandboxed pod -> allowed endpoint.
+
+## Post 3: Offline audit evidence
+
+A log line is not automatically evidence.
+
+If an auditor can edit the record without detection, the record is weak.
+
+The Clawdlinux operator repository includes HMAC hash-chain and JSONL verification primitives.
+The `audit-verify` path can verify exported records without a live control plane.
+
+That matters for air-gapped AI agent deployment.
+The auditor may not have access to the production cluster.
+The verifier still needs a stable artifact and a clear failure result.
+
+The current code does not claim that every run already emits a complete signed artifact.
+That connection remains a release gate.
+
+The useful design question is simple:
+what exact file does the auditor receive, and what command proves it was not changed?
+
+Repo: https://github.com/Clawdlinux/agentic-operator-core
+
+Visual: before/after terminal capture of a clean verification and a deliberate tamper failure.
+
+## Post 4: Runtime-neutral means less lock-in
+
+A governance layer should not require one agent runtime.
+
+Teams already have choices:
+- a custom pod
+- a workflow engine
+- a runtime with its own agent resource
+
+The Clawdlinux operator uses runtime adapters around the workload boundary.
+The governance controls stay at the pod and network layers.
+
+That keeps the hard control in one place:
+isolation, egress, cost paths, and audit primitives apply to labeled workloads.
+
+The adapter chooses how a workload starts.
+It should not reimplement the security model.
+
+This is the boring architecture choice.
+Boring is useful when a platform team has to maintain it.
+
+Docs: https://github.com/Clawdlinux/agentic-operator-core/tree/main/pkg/runtime
+
+Visual: one diagram with three runtime inputs converging on the same policy and evidence boundary.
+
+## Post 5: A Helm chart where operators search
+
+Platform engineers do not need another landing page first.
+They need a chart they can inspect and pull.
+
+The Clawdlinux operator chart is published as a public OCI artifact.
+
+```bash
+helm pull oci://ghcr.io/clawdlinux/agentic-operator-core/agentic-operator \
+  --version 0.1.0
+```
+
+The chart metadata describes the real scope:
+Kubernetes agent governance, gVisor isolation, egress policy, cost controls, and audit verification.
+
+Artifact Hub listing:
+https://artifacthub.io/packages/helm/clawdlinux-operator/agentic-operator
+
+The first install still needs a real cluster review.
+Run it with the CNI, RuntimeClass, storage, and image assumptions your environment supports.
+
+Repo: https://github.com/Clawdlinux/agentic-operator-core
+
+Visual: Artifact Hub package screenshot beside the `helm pull` command.
+
+## Post 6: What this does not claim
+
+Clawdlinux is not a claim that Kubernetes makes agents safe by default.
+
+It is not a claim that generated NetworkPolicy objects are enforced by every CNI.
+It is not a claim that every run already produces a complete signed attestation.
+It is not a claim of FedRAMP, HIPAA, or air-gap certification.
+
+It is a set of Kubernetes governance and evidence primitives:
+- runtime-neutral workload adapters
+- gVisor RuntimeClass mutation for labeled pods
+- generated network policy objects
+- cost and model-routing paths
+- offline JSONL and hash-chain verification tools
+
+The practical test is small.
+Pick one workload.
+Declare its allowed destinations.
+Run the cluster checks.
+Export the record.
+Verify it offline.
+
+If that workflow does not work in your environment, the right result is a bug report or a missing control.
+Not a marketing claim.
+
+Repo: https://github.com/Clawdlinux/agentic-operator-core
+Visual: one screenshot of the actual verification command and its output.
+
+## Distribution checklist
+
+- [ ] Add the public OCI chart to Artifact Hub. Registration is complete; wait for processing.
+- [ ] Verify the Artifact Hub package page after processing.
+- [ ] Request Google Search Console reindex for `https://clawdlinux.org/` and `/products/operator`.
+- [ ] Publish one post at a time. Do not schedule all six without feedback.
+- [ ] Use the raw hero image for Post 1 only.
+- [ ] Use real terminal captures for Posts 3 and 5.
+- [ ] Do not publish an anonymous post under a fake person or fake customer.
+- [ ] Keep the Clawdlinux brand visible. Keep the author identity optional.
+- [ ] Track installs, replies, and qualified conversations. Do not use impressions as validation.