Retire developer identity and workspace mechanisms for parallel Agent Farm workflows

Summary

Trellis' current developer identity and workspace mechanisms create substantial friction for parallel multi-task work. They bind execution state, logs, indexes, and Git-visible paths to a developer-oriented workspace model that does not match the direction of parallel agent execution or a future Agent Farm collaboration model.

This model is a product of an earlier era of human programmer collaboration. In current Trellis usage, it has repeatedly caused practical problems: concurrent tasks and worktrees contend over shared workspace/index files, task progress becomes difficult to resume or reconcile, and otherwise independent execution units are forced through implicit identity and shared mutable state.

The result is that task progression becomes unnecessarily fragile and operationally cumbersome precisely when multiple agents or tasks need to proceed independently.

Requested direction

Retire the global developer identity mechanism and fully retire the workspace mechanism from the active runtime and installation/update flows:

• Do not use a global developer identity as implicit authority for task creation, selection, or session routing.
• Do not create, update, merge, index, or consume shared developer workspace directories, indexes, or journals during normal workflows.
• Use explicit caller/task ownership and operation-local session/task bindings instead of ambient identity.
• Keep task records and task-scoped evidence as the durable source for task progress.
• Preserve existing historical identity/workspace data without consuming or mutating it.

What remains useful

The developer/workspace mechanisms did provide useful task indexing and log navigation. That need is real, but it should be solved with a model that is safe for concurrent execution units rather than by retaining the current shared workspace mechanism.

Discussion #562 describes the intended direction in detail: https://github.com/mindfold-ai/Trellis/discussions/562

In particular, task records should remain durable and historical context should be recoverable on demand. The concrete task-index, completion-record, and on-demand historical recall design can be addressed by future Issues and PRs rather than preserving the current identity/workspace implementation.

Acceptance direction

A future implementation should make parallel tasks and worktrees independent, avoid shared tracked workspace writes, preserve task-scoped evidence, and provide explicit on-demand historical lookup without requiring a global developer identity.

The castbox/Trellis implementation and validation work for this direction is available in PR #4 in the fork: https://github.com/castbox/Trellis/pull/4

This Issue is a request for upstream design/implementation alignment, not a request to rewrite historical records or Git history.