Retire the top-level plans/ directory and correct the document in it

[Masterplanner25]

plans/ held one file, written 2026-05-30 and never reviewed. It is
docs/design/workflow-framework/00-framework-plan.md now, beside the existing
docs/design/workflow-dsl/ — a subsystem design folder is the shape this repo
already uses, and a top-level directory for a single document is not.

The drift mattered more than the location

Its primary code reference had been wrong for three months. The "Main Code
Areas" section pointed at src/nodus_workflow — a path the NAME-COL-001 rename
removed on 2026-05-31, the day after this plan was written. Anyone following
the plan's own first reference found nothing.

Nine of its links resolved for nobody: machine-local absolute paths of the
form /abs/path/C:/dev/Coding%20Language/src/.... Worth noting that the link
checker added in #833 does not catch these — they are not .md targets — so
this is a second species of broken link, found by reading rather than by the
gate. They are inline code now; a relative link from
docs/design/workflow-framework/ into src/ is four levels of ../ that nobody
will keep correct.

The roadmap is live and was not rewritten

Spot-checked rather than assumed: §1 claims the run inventory uses
offset-derived cursors, and runner.py still has _cursor_offset and
_encode_cursor. The remaining-work items — cursor model, coordination posture,
admin operations, storage scaling, auth hardening — are all still open.

Rewriting a plan is a design session, not a docs sweep.

What it gains

A Last reviewed: header, per the convention #715 established, naming the two
items that moved underneath it:

• LIMITS-002: Default workflow runner uses non-durable LocalWorkflowStore #174 answered part of §2's open question — "is SQLite the final production
  backend or an intermediate one?" It is the default at 6.0.0, and
  migrate-store shipped in 5.10.0. What sits beyond it for multi-node
  coordination is still open.
• Test suite accumulates workflow-run files in the repo's .nodus/, and LocalWorkflowStore rescans all of them — the suite destabilizes its own later runs #380 advanced §4 without closing it — list_runs() is about 4× cheaper
  (the cost was a mkdir per record and an exists before every open, not
  parsing), and still linear in accumulated runs.

The header also says plainly that nothing else was re-verified line by line,
so the next reader knows what they are holding.

Checks

• nodus_gate --all — all ten phases
• test_doc_links_resolve — clean
• ruff check . — clean
• The move is recorded as a rename, so history follows the file

🤖 Generated with Claude Code

https://claude.ai/code/session_01TnzKwmx6XecNEqP2482MuA