refactor: pure moves — test tails, the clap surface, and eight submodules

[schickling-assistant]

What

Pure moves only: 14,145 lines relocated, plus two one-line fixes to tests that could not fail. Implements T1-GATE, T1-TAILS, T1-CLI and T1-SPLIT of reports/deepclean-st2.md §2.C.

	what	lines
T1-GATE	#[cfg(test)] on src/run.rs's mod tests — the only ungated test module of 50 in src, so its tail was compiled by cargo build --release. Plus #[test] restored on hooks::tests::pi_launch_classification_is_exact, which had silently never run (its omp twin beside it is attributed; cargo check reported it as dead code).	+2
T1-TAILS	four test tails → sibling tests.rs	12,906
T1-CLI	main.rs's clap surface → src/cli.rs	1,239
T1-SPLIT	eight submodules out of five parents	4,226

The splits, each a contiguous-span move with the minimum markers:

• codex_app_server/{protocol,context,process_group}.rs — 692/264/120 lines, 8/5/5 markers
• resync/{watch_set,read}.rs — 463/170 lines, 14/2 markers
• run/process.rs — 264 lines, 7 markers
• catalog_transaction/capability.rs — 310 lines, 5 markers. Not fs.rs: the parent does use std::fs::{self, ..}, so mod fs; is E0255.
• agent_author/{markers,stream,resource,desired_state,declared_field}.rs — 1,949 lines across 21 spans, 6 markers

What deliberately did not move

• The three do-not-split verdicts stand: run.rs's reconcile core (2,045 LOC / ~30 markers on exactly the state INVARIANTS guards), main.rs's verb impls (a common.rs would be main.rs renamed), codex's state-dir + io core (313 LOC / 22 markers, atomic_json alone has 6 callsites).
• agent_author's shared declaration writer stays in the parent for the same reason, as do its two #[cfg(test)] *_for_test scaffolds — which is what buys pub(super) instead of editing test code.
• pub const CODEX_CONTEXT_VERIFIED_VERSION stays in codex_app_server.rs: moving a pub item into a private child deletes it from st2::codex_app_server's API with no compiler complaint.

Every parent keeps re-export glue at each item's own visibility, so no file outside a parent/child pair was edited — all 21 external crate::catalog_transaction::* / crate::run::* paths, st2::agent_author::*, st2::resync::watch_set_for* and every sibling tests.rs's use super::* resolve unchanged.

Two consequences of moving a file, both real

1. include_str! is relative to its file. Five fixture paths in the moved codex tail needed ../tests/fixtures/ → ../../tests/fixtures/.
2. INVARIANTS.md cites proofs by file. tests/invariants.rs::qualified_proof_references_resolve reads every backticked <path>.rs::<test> span and asserts the test is declared there — and it went red, which is exactly what the newly-gated target is for (ci: run every hermetic test target in the st2 gate #499). 38 citations were repointed: 21 src/ding/mod.rs::, 7 src/resync.rs::, 5 src/run.rs::, 5 src/codex_app_server.rs::. Row text, guarantees and proof sets are unchanged; all 257 qualified proofs resolve.

Checks

• Gate nix build .#st2 --no-link -L: EXIT=0 (pty st2.clean-3.gate3, log /tmp/st2-clean3-gate3.log).
• Test counts, against refactor(delivery,identity,vrs): delete a label branch, give DELTA-006 a producer, fold two duplications #500's gate run: --lib 743 (was 742 — the +1 is the restored #[test]), run 60, hooks 11, codex_hooks 5, driver_expansion 13, discovery 75, validate 64, message 6, agent_publish 15 + 9 ignored, agent_address 12, agent_desired_state 8, catalog_graph 10, invariants 2, status_agents 12, vrs_ledger 3. No target lost, gained or renamed a test.
• Extra gate for T1-CLI: st2 completions {bash,zsh,fish} | sha256sum is byte-identical between refactor(delivery,identity,vrs): delete a label branch, give DELTA-006 a producer, fold two duplications #500's binary and this one (3d8af97b4ff296ab / 066a2a70d3ae1d36 / 9bc9d828e7155e20).
• Byte-identity proof of the moves, run before committing: re-indenting each new tests.rs (adding four spaces to exactly the lines that were dedented) reproduces the removed block byte-for-byte, for all four files; cli.rs equals main.rs:16-1254 after removing 28 pub(crate) markers, with zero residual differences.

Independent review

MoveReviewer (clean context, read-only, no build access) re-derived the tail extraction itself for all four files, ran its own quote-parity scan to locate all 83/76/40/9 in-string continuation lines and confirmed no string literal's content changed, exhaustively searched for include!/#[path]/file!()/module_path!()/concat!/env! hazards, and verified programmatically that each of the 28 cli.rs differences is exactly "pub(crate) " inserted after the leading whitespace with no clap attribute altered.

Verdict: "no semantic change". Two non-blocking notes, both recorded rather than fixed:

1. Four intermediate commits do not compile on their own (src/run/tests.rs arrives one commit after mod tests;; the codex include_str! repoint arrives three commits after the tail move). Squash-merged, so main never sees them; worth knowing if anyone bisects this range.
2. cli.rs:719's fn selection signature grows from 102 to 113 columns. cargo fmt --check is deliberately not gated (flake.nix), and the line was already over width on main.

Posted on behalf of @schickling

field	value
agent_identity	dev3.direct.omp.43sz6ujq
session	dev3.43sz6ujq
agent_persona	generalist
agent_supervisor	unavailable
agent_tool	OMP
agent_tool_version	18.1.7
agent_runtime	OMP 18.1.7
tooling_profile	dotfiles@39a19af