Changelog

All notable changes to the RHDP Skills Marketplace.

---

v2.13.4 — 2026-04-20

Fixed

• agnosticv + showroom plugins: versions were stuck at 2.12.4 while health and ftl had moved to 2.13.3 — all four plugins now consistently report v2.13.4
• FTL sub-tool skills removed: content-reader, solve-writer, validate-writer, env-connector were exposed as both slash commands and agents — they are now agents only. Only /ftl:rhdp-lab-validator is the user entry point
• env-connector: oc cp was pushing validation.yml but validate-writer produces validate.yml — would have caused runner pod to never receive the validate playbook
• rhdp-lab-validator: name: ftl:lab-validator corrected to name: ftl:rhdp-lab-validator to match directory and invocation path

---

v2.13.3 — 2026-04-16

FTL Skills — Sub-agents now callable via Task tool

Added ftl/agents/ directory with 4 proper agent definitions. Previously the 4 sub-skills (content-reader, solve-writer, validate-writer, env-connector) were only registered as skills, causing the orchestrator to fail when spawning them via the Task tool.

• ftl/agents/content-reader.md — AsciiDoc parser agent, subagent_type: ftl:content-reader
• ftl/agents/solve-writer.md — solve.yml generator agent, subagent_type: ftl:solve-writer
• ftl/agents/validate-writer.md — validate.yml generator agent, subagent_type: ftl:validate-writer
• ftl/agents/env-connector.md — live test runner agent, subagent_type: ftl:env-connector
• Orchestrator updated — rhdp-lab-validator/SKILL.md now explicitly uses Task tool with correct subagent_type for each step

---

v2.13.2 — 2026-04-16

FTL Skills — Orchestrator Quality Fixes (8 issues)

Automated skill-reviewer audit found and fixed 8 issues in the FTL orchestrator and sub-agent skills:

• Archived shadow agents/ directory — old contradictory versions of all 4 sub-agents
• “Existing files?” gate added to orchestrator Step 0 — uses existing solve/validate as baseline
• rhpds.ftl.validation_check FQCN enforced — bare validation_check fails without collection routing
• Self-healing wired into orchestrator — Playwright failures route to vision recovery, not solve-writer
• SOLVE_ACTIONS structured format — task-N: {action, check, async} for direct validate-writer use
• Numbering and labeling fixes in env-connector (Step 8) and content-reader (Section C)
• Merged split frontmatter in all 4 sub-agent SKILL.md files

---

v2.13.1 — 2026-04-16

FTL Skills — Self-Healing Vision Pattern

When UI versions change, Playwright selectors break. Solution: intent-based automation + live vision recovery.

• ftl:content-reader — Vision analysis of image:: screenshots from .adoc assets. Stores intent descriptions, not CSS selectors.
• ftl:solve-writer — Playwright scripts use INTENT constants. Saves before/after/debug screenshots.
• ftl:env-connector — Screenshot evidence collection. Self-healing: failure → vision → new selector → retry. ui-versions.json per test run.

---

v2.13.0 — 2026-04-16

FTL Skills — 4 Agent SKILL.md Files

Added separate, standalone SKILL.md files for each of the 4 agents orchestrated by ftl:rhdp-lab-validator:

• ftl:content-reader — AsciiDoc reader. Extracts role="execute" blocks, vision analysis, GUI step decision tree.
• ftl:solve-writer — Writes solve.yml with all automation patterns.
• ftl:validate-writer — Writes validate.yml using rhpds.ftl.validation_check.
• ftl:env-connector — Live test runner: push → restart → test cycle → report.

---

v2.12.4 — 2026-04-15

FTL Skills

ftl:lab-validator — full rewrite

Rewritten from scratch. No longer scaffolds AgV or Showroom — focuses on writing and testing solve.yml / validate.yml playbooks. Pre-flight message updated with all required files and AgV config. Key field learnings from LB2860/LB2010/LB2865 incorporated: k8s_exec string rule, kubeconfig requirement, namespace probing, idempotency, async patterns, dev.yaml warning, specific error messages.

---

v2.12.3 — 2026-04-15

AgnosticV Skills

agnosticv:catalog-builder + agnosticv:validator — dev_mode common.yaml check removed

ocp4_workload_showroom_antora_enable_dev_mode: "false" is no longer generated in common.yaml and no longer checked by the validator. The only enforcement is that dev.yaml has it set to "true". cloud-vms-base is unaffected — vm_workload_showroom has no dev_mode toggle.

---

v2.12.2 — 2026-04-15

AgnosticV Skills

agnosticv:validator — Check 25b VM E2E severity to WARNING

showroom_ansible_runner_image / _tag both missing on cloud-vms-base CIs was a SUGGESTION. Now WARNING, consistent with the OCP/tenant Check 25a change in v2.12.1.

---

v2.12.1 — 2026-04-15

AgnosticV Skills

agnosticv:validator — Check 27: showroom in cluster CI (new)

ocp4_workload_showroom and vm_workload_showroom must not appear in any CI with “cluster” in the display name, or that has __meta__.components. These are per-user workloads — cluster CIs provision shared infrastructure only. → ERROR

agnosticv:validator — Check 26: litellm placement extended

Previously only detected cluster CIs via __meta__.components. Now also checks display name for “cluster” — consistent with the pattern used in Check 25 and Check 27.

agnosticv:validator — Check 25: E2E severity to WARNING

“E2E not configured” demoted from SUGGESTION → WARNING. Partial E2E config (enabled but image or FTL workload missing) demoted from ERROR → WARNING. Cluster CIs are excluded entirely (no change).

agnosticv:catalog-builder — pool references without /prod

Generated item: values in __meta__.components no longer include the /prod suffix (agd-v2/ocp-cluster-cnv-pools not agd-v2/ocp-cluster-cnv-pools/prod). The ordering system selects the appropriate pool — hardcoding /prod bypasses dev-mode ordering.

---

v2.12.0 — 2026-04-15

AgnosticV Skills

agnosticv:validator — Private validator delegation (issue #12)

The public validator now has a Step 0 that detects the commitv skill in the private AgV repo before running any checks. If found, it loads and uses it for internal RHDP-specific validation. If not found, it falls back to built-in checks and tells the user. Keeps sensitive internal check logic private without losing the public entry point.

agnosticv:catalog-builder — Expanded reference catalog discovery (issue #14)

Step 2 discovery now searches beyond agd_v2/ and openshift_cnv/. Added: ai-quickstarts/, enterprise/, summit-2026/, sandboxes-gpte/, zt_rhel/, rhdp/. Results filtered by config: field to exclude agDv1 catalogs automatically.

agnosticv:catalog-builder — Tenant CI console_embed fix (issue #15)

The auto-set block in sandbox-tenant-ci-questions.md listed ocp4_workload_ocp_console_embed alongside Showroom as things that get added for Q-T4. This contradicted the correct “Do NOT add” note and caused the builder to put console_embed in tenant CIs. Removed from the auto-set comment — console_embed belongs in the Cluster CI only.

Documentation

best-practices.md — Remove AI attribution suppression (issue #17)

Removed “No AI attribution” from 4 locations: the global CLAUDE.md example, the AgnosticV CLAUDE.md example, a prompt workflow example, and the git rules table. When writing with Claude as a co-author, attribution should be included.

---

v2.10.10 — 2026-04-13

AgnosticV Skills

/agnosticv:validator — Password security improvements

Addresses password issues flagged in Nate Stencell’s review of Summit 2026 catalog items.

What changed:

• Validator now deep-scans the full YAML tree (dicts + lists at any depth). Previously only top-level keys were checked — nested passwords inside AAP credential inputs, workload configs, and list items were invisible.
• Added b64encode, sha256, sha1 to bad patterns — catches the guid|md5|int(base=16)|b64encode chain.
• Applies to all agnosticv directories (catalog/, summit-2026/, agd_v2/, tests/).

Now caught (examples):

• vault_password: MzIzNTE0OTEw — hardcoded base64 AAP vault credential (nested)
• "" — GUID hash-based password
• "" — md5 + b64encode chain
• "band-on-the-run" — literal static password
• "redhat" — literal cosign password

---

v2.11.3 — 2026-04-01

FTL Plugin

New skill: /ftl:rhdp-lab-validator

Zero Touch grading for RHDP Showroom labs. Generates grader and solver playbooks.

• Covers 4 lab types: OCP tenant, OCP dedicated+bastion, RHEL VM+bastion, AAP
• Wraps existing bash scripts — no rewrite needed
• Generates multi-task ✅/❌ validation (mandatory for all modules)
• curl test commands after each module

---

v2.10.9 — 2026-04-08

AgnosticV Skills

/agnosticv:validator — False positive fixes (Nate Stencell, issue #11)

• Removed purpose: field check — no longer required
• Removed /prod pool suffix check — ordering system determines pool environment

Docs

• docs/setup/claude-code.md: clarified /plugin install scope selection
• Security guidelines merged from Guillaume Coré (PR #13)

---

v2.10.8 — 2026-03-26

Showroom Skills

/showroom:create-lab and /showroom:create-demo — scaffold fix

• Step 3.1 now lists all three scaffold questions: Q0 (catalog type), Q1 (tabs), Q2 (theme)
• Previously Q2 (ui-bundle theme) was skipped — wrong theme URL written to site.yml
• Showroom collection version updated to v1.6.0 in reference files

---

View full CHANGELOG on GitHub