[loopx-agent]

Retires the obsolete tmux multi-agent and auto-research demos as one reversible cleanup. Source checkouts and installed CLI builds no longer advertise or register auto-research or multi-agent; rejected launch attempts return argparse errors before writing state.

Goal And Delivered Outcome

• Basis: accepted maintenance request to remove both old demos, their dedicated entrypoints, helpers, docs and tests. No separate written specification; this PR states the requested scope.
• Removes demo/multi_agent/, demo/auto_research/, the old CLI adapter and shared tmux launcher. The launcher has no surviving production caller.
• Removes dedicated smoke/test/documentation consumers, the obsolete canary profile and help entries. Mixed governance/namespace/vision checks retain their active assertions. Current indexes and teaching examples use surviving paths; historical RFC references are pinned to a pre-retirement revision in both languages.
• Preserves demo/workspace, examples/collaboration-delivery, examples/managed-research-team, generic collaboration/research owners, historical incident evidence and runtime data.
• Target: main; integrated and verified merge base c6c328d50b456c3dbb88de7021414c1921b69174. The original cleanup commit and subsequent merge of main are DCO-signed. The complete 111-path cleanup diff is byte-identical before and after integration; upstream fixes are already part of the target branch.

Author Declaration

• Written by: model_agent — OpenAI Codex (GPT-6).
• Implemented against: this PR's maintenance scope and repository policy at the stated base.

Criterion	Disposition	Implementation / evidence
Retire both old prototypes and exclusive helpers	implemented	Explicit tracked-file/caller inventory; 89 dedicated files removed
Remove active entrypoints and stale guidance	implemented	CLI registration/dispatch/help, canary, indexes and course sections
Preserve generic owners and modern examples	implemented	Unchanged surviving example trees; retained-path validation below
Reject retired commands without side effects	implemented	Both CLI routes tested against an isolated registry and file snapshot
Maintain historical context and RFC language pairs	implemented	Pinned historical links, retained incident, bilingual diff gate

Scope And Continuation

Maintainer huangruiteng merged this PR at 2026-11-06T15:50:23Z as 546c08c0d3f22358ef09b13e69f7edaa806121b8 (parent 1e6add14ec2c6799be7fbfb8b7de128c8938dff3). This author continuation performed no approval or merge. The remote-main source removes the retired paths and preserves the three retained example trees unchanged. Its cleanup diff is byte-identical to the reviewed proposal. The full merged tree also contains upstream #8178, so it was tested separately rather than assumed identical to the author integration head. Required failures below remain unresolved, and no formal PR review is recorded. No installed adoption, authenticated apply, installation, release or production deletion is claimed. Reverting the cleanup commit restores the removed prototype.

The CLI remains its existing Python transport owner; no new decision owner, vocabulary or parallel implementation is introduced. Same-domain simplification removes dynamic demo imports and blanket exception fallbacks. This update incorporates the existing upstream F811 repair rather than introducing another fix.

Validation

• Tested author integration head: 73071ecf4d12994fbef04431133c7229892274d7; independently tested merged main: 546c08c0d3f22358ef09b13e69f7edaa806121b8.
• Run state: finished. Input classes: synthetic, public_fixture.

Check kind	Result	Evidence / limitation
unit / integration / real_backend	failed	Author head and merged main were run separately: 230 passed, 3 failed each across CLI, source CLI, workspace story and Codex history. Passing coverage includes isolated SQLite replay and retired-command rejection/persistence checks; neither whole run is a pass
regression diagnosis	failed	The same three selected tests independently fail on immutable main c6c328d and head 73071ec in checkout-local environments: both outer-controller entry variants return mode=full instead of should-run; first-use JSON lacks its expected disclosure. Required failures remain blocking; no baseline exemption
static	failed on merged main	Complete kernel Ruff command passes on author head 73071ec. Actual main 546c08c fails F841 at tests/control_plane/test_goal_create_completion.py:184, from upstream #8178; this file is byte-identical to the squash parent. The old F811 findings are resolved. No exemption
regression_parity	passed	Complete cleanup diff unchanged across main integration. Earlier immutable pre-retirement base/proposal checks: retired command help exits 0 → 2; retained demo and deepresearch help stays byte-identical; no demo execution or model calls
integration	passed (prior revision)	249 retained CLI/workspace, modern research and delegation tests plus namespace, generic role-successor, canary and Goal vision smokes on the previously documented revision. These are retained evidence, not a fresh full-suite run on this head
static	passed	Exact-head bilingual RFC gate; public scan of all 22 surviving changed files; committed diff whitespace and DCO checks
integration	failed	Author head and independently tested merged main both reject untouched controller-executor-unified-contract-v0 lifecycle/dated-log headers and stale RFC status indexes. No gate waived
quality qualification	blocked	Exact-scope CQR cqr_ff8196d9eea3a24e2f5b records two blocking risks and the failed required validators. Native verification returns invalid_receipt; strict qualification remains held
premerge	blocked	All 19 selected checks and six direct checks passed with zero command failures; the overall gate is quality_invalid_receipt because the required failures remain recorded in CQR
manual	not_run	Installed App, live tmux/model workers and production state; outside this removal proposal's validation boundary

Previous CI on ce42014 failed in Python Tests and Release Artifacts. The old kernel lint failure is now fixed locally by upstream integration. Release Artifacts previously failed its installed-browser classified-total assertion (1 versus —); its cause and outcome on the new head remain unqualified. New-head CI readback around the merge contains cancelled and queued jobs; no passing CI conclusion is claimed. No formal PR review is recorded. The observed merge does not change the failed local validators or strict qualification result.

Frontend / Visual Evidence

• UI impact: none in the cleanup diff. No dashboard/site/desktop layout or documentation chrome changed; screenshot/viewport review is not applicable. CLI discovery and rejection behavior are covered by entrypoint checks above.
• Source data: synthetic, public_fixture. No private Goal state, local paths, raw logs, credentials, captures or benchmark trajectories are published.

Type Of Change

• Breaking cleanup of retired demo-only CLI commands; documentation and test updates.
• The cleanup and main-integration author commits include DCO sign-offs; paired RFC changes pass the bilingual gate.