BE-039: pager put/delete crossing a sub mount — authorityRepo anchors on arg0's FULL path, repo becomes the SUB, raw rest args resolve in the wrong tree

Now: DONE

A pager :put test/bro/glob/* (glob-expanded per BE-036 — expansion verified correct) staged arg0 into the test/ SUB as bro/glob/glob.js (row displayed UNPREFIXED) and skipped the rest arg test/bro/glob/run.sh as "does not exist". Chain: the initial view has no address (BRO-017) so the composer context falls back to be.navCwd() = //BE-036; shapeArg0 merges the authority into ARG0 ONLY (rest stays raw wt-relative, the Nav/URI-011 law); then authorityRepo (core/loop.js:168) resolves be.wtdir("//BE-036/test/bro/glob/glob.js") — the nav WITH arg0's whole file path — and anchors repo = be.find(dir), which climbs UP from the FILE and lands on the NEAREST repo: the mounted test/ sub. Arg0 is recut sub-relative; the run's repo is silently the SUB; every raw rest arg then resolves against the sub root. CLI is unaffected (no authority → authorityRepo returns null → cwd repo; put's own stageInSub handles mount-crossing correctly, multi-arg verified green). DELETE hazard: a same-named path existing in the sub makes a raw rest arg silently target the WRONG file — destructive mis-target. Method Issues.

Input

Context

User repro 2026-07-10, pager at the BE-036 wt root, spell :put test/bro/glob/* → banner ?, row put bro/glob/glob.js (unprefixed, staged in the sub wtlog), skip test/bro/glob/run.sh does not exist — skipped (file exists at the top level; left unk).
Hermetic repro (this dir): run.sh + drive.js over the test/sub/lib/subcase.sh fixture — put("//cli/vendor/sub/x/k.txt","vendor/sub/x/l.txt") reentry → put x/k.txt staged in the sub + vendor/sub/x/l.txt does not exist — skipped. Exact match of the user output shape.
The seam, core/loop.js:168 authorityRepo: navStr = uri._make(undefined, u.authority, u.path, …) keeps arg0's PATH; dir = be.wtdir(navStr) → a file path; repo = be.find(dir) → nearest enclosing repo = the SUB when the path crosses a mount; rel = dir.slice(repo.wt.length + 1) recuts arg0 sub-relative.
Ingredients, each its own ticket: BRO-017 (initial view loses its address → navCwd fallback), the BE-037 hive navCwd minting //NAME for a SRC_ROOT worktree, BE-036 glob multiplying args (exposes the multi-arg case). The composer law itself is correct: rest args MUST stay raw (BE-030, memory of the URI-011 bindRest retirement).
Verified-good controls (same fixture): CLI argv multi-arg cross-mount put ✓, argv+--tlv ✓, loop.cli reentry without authority ✓, reentry after a sub-file view ✓ — only the authority-on-arg0 form breaks.

Goals

Verbs own resolution: put/delete resolve EVERY arg — arg0 and rest alike — via resolve(context, base, rel) (BE-030); the context reaches the verb as the CONTEXT slot (the be/nav scope through the reentry), never baked into an arg's spelling.
Pager :put <sub-crossing> <rest…> behaves identically to the CLI form: all args staged, rows displayed with the full top-relative prefix, correct parent-side tallies; mount descent remains the VERB's job (stageInSub, proven correct).
A regression test for the context + mount-crossing + multi-arg combination (put AND delete).

Constraints

ALL verb args stay raw as typed/expanded — the composer's ctx-merge onto arg0 (shapeArg0/mergeUri) is retired for verb word-calls; slot-edit spells (?ref/#frag re-anchoring the current view) are navigation, not verb args, and keep their merge.
Keep BE-030/BE-033 nav semantics (//name = worktree, NAVNONE on miss) and the BE-011/BE-027 confinement chokepoints.
Repro-first (Issues); fixture repro in this dir must become a test/sub/ case.

WIP

Design decisions

RULED (gritzko 2026-07-10): "the put itself must deal with context+rel_path resolution; pager must not resolve arguments." NOT an authorityRepo re-anchor patch: the pager passes the context as context and the args verbatim; put/delete resolve each arg via resolve(context, base, rel). Folds into the BE-030 unification; authorityRepo's find-from-full-path climb disappears with the merged-arg0 spelling that fed it.

TODOs

Port run.sh/drive.js into a test/sub/ case: test/sub/pagercross (put + delete arms, real composer + driveSpell reentry). RED on the old composer even shows the live destructive delete (deleted 1 file(s) — the sub's own x/k.txt); GREEN after (put stages both full-prefix; delete = safe CLI-identical DELDIRTY refusal).
Context through the reentry: spell.compose returns {verb, arg0, rest, context} (verb word-calls: args RAW, no arg0 merge; slot-edits keep theirs); driveSpell(spell, context) → loop.cli(argv, {reentry, context}); new core/loop.js contextRepo(ctxUri) scopes be.repo/authority from the CONTEXT (authority+ctx-path only, no arg path — find lands on the tree top), non-fatal miss, CLI path byte-unchanged.
put/delete behave CLI-identically from the pager (verbs' own stageInSub/wtpath resolution now sees the right repo + raw args); full resolve(context,base,rel) per-arg unification stays under BE-030.
AMEND RULING NEEDED: test/bro/pager/driver.js 7 cc-* checks assert the RETIRED arg0-merge shape (cc-put-multi wants arg0 //THERE/a.txt) — the only net new red; worker left them per the no-test-edits rule. Amend to the new {raw arg0, context} shape on gritzko's order.
Delete cross-mount DESCENT (CLI too) is a pre-existing gap (no stageInSub twin — named cross-mount delete refuses DELDIRTY); own ticket if wanted.
Audit other be.find(<deep path>) anchor sites for the same climb-past-mount class.

Blockers and bummers

Trunk moved mid-work (BE-038 todo verb landed); the BE-039 wt is behind 1 — jab get FF before land (loop.js overlap risk looks low).
Fixture navCwd mis-names the cell (harness firewall .be climbs topWt out of $WORK) — the test feeds the clean-hive context shape directly; navCwd cell-naming is BRO-017/BE-037 territory.

Outcome

(open — fix staged in wt /home/gritzko/src/BE-039, uncommitted: core/loop.js, shared/spell.js, views/bro/bro.js, views/bro/pager.js + sub-staged test/sub/pagercross/*; suite 161/163 vs baseline 162/163, the delta being the 7 retired-shape cc-* checks awaiting the amend ruling; orchestrator re-verified repro RED/GREEN + counts 2026-07-10)