Beagle SCM

SUBS-013: full git submodule add parity from be put

LOW (feature gap — design needed). Escalated from SUBS-009. There is no single-command flow that introduces a brand-new submodule the way git submodule add <url> <path> does: stage a new 160000 gitlink for the mounted sub and synthesize the .gitmodules section, so a subsequent POST records the new sub. SUBS-009 landed only the actionable refusal; the add itself is blocked on two design decisions.

Open decisions (must be agreed before forking)

• URL source for .gitmodules url =. A mounted-but-untracked sub's

  row-0 .be anchor is file:/…/.be//?/<title>#<sha> — a local shard pointer, not a portable upstream. Candidates:

  1. require the user to have written the .gitmodules url = first (the

     current working flow; be put only stages the gitlink);

  2. recover the upstream from the sub-shard's reflog (the URL it was

     fetched from), falling back to the declared .gitmodules;

  3. a new explicit arg carrying the URL.
• Pin source / grammar. be put <subpath> has no pin slot, and

  be put <subpath>#<sha> is already the move-form (PUTDSTBAD). Either a new grammar (e.g. a dedicated sub-add verb-shape) or "pin = the sub-wt's current tip read from its .be anchor" (parity with the existing gitlink-bump path).

Existing partial flow (works today)

User writes the .gitmodules section, runs sniff sub-mount ./<sub>#<pin>, then be put <sub> stages the gitlink; POST records gitlink + .gitmodules (covered by test/post/15-sub-add-then-post). SUBS-013 is about collapsing that into a single git submodule add-style command.

Fix sketch

Once URL + pin sources are chosen: detect a mounted-but-untracked sub dir under an explicit be put <path>, read the pin, synth the .gitmodules section via SUBSu8bSynth, and stage both the 160000 gitlink and the .gitmodules blob.

Tests

test/put/11-sub-add-refusal (landed with SUBS-009) pins the current refusal; the auto-add case follows once the grammar is decided.