Track repository-local Yarn wrapper in the RHDH Fullsend installation

[durandom]

Purpose

Keep an open tracking issue for the RHDH-specific Yarn customization introduced by rhdh-agentic#98:

.fullsend/rhdh/bin/yarn

This issue should remain open for as long as the repository-local wrapper or equivalent RHDH-specific modification is required.

Context

rhdh-fullsend#24 tracked the immediate Yarn/Corepack outage and was closed after PR #98 restored working Yarn commands. That incident is resolved, but its solution remains a local customization that needs ownership and eventual upstreaming or centralization.

The reference rhdh-agentic installation currently inherits ghcr.io/fullsend-ai/fullsend-code:latest and exposes the target repository's vendored Yarn release through a wrapper mounted as /sandbox/workspace/bin/yarn.

The source is namespaced under .fullsend/rhdh/ because Fullsend v0.30 overlays the standard .fullsend layered directories before resolving the harness.

Current modification

The wrapper directly executes:

${FULLSEND_TARGET_REPO_DIR}/.yarn/releases/yarn-4.12.0.cjs

This avoids:

• corepack enable attempting to write shims into read-only /usr
• downloading Yarn during every agent run
• relying on a custom RHDH sandbox image solely to provide Yarn

The harness additionally delivers Yarn proxy configuration and an npm policy adjustment for encoded scoped-package paths.

Tracking concerns

• The Yarn version is hardcoded and must be updated in lockstep with .yarnrc.yml and .yarn/releases/.
• The wrapper assumes every target repository vendors Yarn at this exact layout.
• The customization currently lives in rhdh-agentic, rather than being generated or owned centrally by rhdh-fullsend.
• rhdh-fullsend documentation still describes the custom RHDH image and /usr/local/bin/yarn, while the reference installation uses the upstream image and /sandbox/workspace/bin/yarn.
• Other RHDH repositories must reproduce this behavior to match the reference installation.

Desired outcome

Move this requirement to the most appropriate shared layer—Fullsend core/base image, the shared agents installation, or a centrally managed rhdh-fullsend blueprint—so enrolled repositories do not need a hand-maintained Yarn wrapper.

If upstreaming is not practical, centralize generation and version discovery so repositories do not hardcode a Yarn release filename.

Acceptance criteria

• Decide the long-term owner and layer for Yarn provisioning in RHDH Fullsend agents.
• Remove the hardcoded yarn-4.12.0.cjs coupling or validate/update it automatically from .yarnrc.yml.
• Align rhdh-fullsend documentation with the reference rhdh-agentic installation.
• Ensure the same solution can be promoted unchanged to other RHDH repositories.
• Remove .fullsend/rhdh/bin/yarn from enrolled repositories, or explicitly document it as a supported centrally managed customization.
• Keep this issue open while the RHDH-specific modification remains deployed.