# W31 — The shell says where it is First wave of **Milestone #38** (*the control surface: what the board can do*). Three changes, all on the agent-facing shell surface rather than the Discord board, and all one subject: **an instance that tells an agent which repo it is.** The board waves come next (W32: #11 and #18; W33: #26; W34–W35: #10). Design ruled with the Operator on 2026-07-28 in a grill session; the milestone-level scope, the two Requests deliberately excluded, and the one platform constraint are in issue **#38**. ## Why these three first They are the wave with no Discord surface, so they are provable by `cargo test` plus one live drive of the installed binary — no Operator board read is owed. That matters because the board waves each end in the Operator's eyes, and shipping the surface-free work first keeps the milestone moving while those queue. They also close a wound taken during the grill itself. Three writes went to `alchemy-testbed#23` believing the instance was bound to spt-progress-tool: every ack read `… added to #23` and none named a repo. Repaired by hand (comments deleted, `state: CUT` reverted to `state: DONE`, close reason restored). #37 is that misfire minted. ## A. #37 — every write verb's ack names the bound repo `CONTEXT.md` keeps repo out of the argument vector by design ("repo is never a verb argument"), and that rule stays. The input side is defended by the instance alias ("its alias names that repo") — which was `null` on both of this owner's instances until the grill set it, and which `spt shell teardown` **frees**, so the defense is absent exactly when a node has just churned its instances. The ack is the durable half. - **T1 — impl.** Every write verb's answer echoes `owner/repo#N` where it echoes `#N` today: `create`, `comment`, `state`, `tag`, `flag`, `link`, `dispatch`, `milestone`, `watch`, `unwatch`, `sweep`, `release`. Read verbs already name the repo — unchanged. - **T2 — unit.** One test per verb family asserting the repo appears in the answer. The regression this guards is not a crash; it is an ack that reads correct against the wrong repo, so the assertion is on the repo string, never on "contains #N". - `REQ-ACK-NAMES-REPO` → `["doc", "impl", "unit"]`. ## B. #36 — `bind-repo` finishes the job Two halves, one answer. **Self-alias.** `spt shell rename --owner ` is reachable from the shell process: `--owner` is node-local namespace selection, not authenticated identity (measured — `spt shell list --owner liam` answers from this session with no challenge), and alchemy already shells out to the `spt` CLI for every answer it sends. The owner id is known at exactly the right moment: the pump learns it from the first owner frame (`main.rs:138`), and `bind-repo` **is** an owner frame. Note the shell-side `spt api` surface has no alias verb at all — this rides the owner-facing CLI deliberately, and that is why it works today with no spt-core dependency and nothing routed to doyle. **The followup.** The same answer names what is missing: no channel provisioned for the bound repo → the verb that provisions one, plus the nudge that offering it is the user's call, not the agent's. No Discord bot configured at all → say *that* instead. Never silence: a bind that succeeds and a board that never appears is the silent degradation this project refuses to ship. - **T3 — impl.** `bind_repo` renames the instance and appends `alias set to ` to its answer. A rename that fails is reported in the same answer and never fails the bind — the bind is the thing the owner asked for. - **T4 — impl.** The followup lines, gated on the registry read the shell already does for `daemon status` (the config dir's `daemon.toml`), not on a Discord call. - **T5 — unit.** Bound-and-registered → no tip. Bound-and-unregistered → the `daemon project add` tip. No daemon config → the setup line. Rename failure → bind still succeeds, and the answer says both things. - `REQ-BIND-SELF-ALIAS`, `REQ-BIND-FOLLOWUP` → `["doc", "impl", "unit"]`. ## C. `daemon project add` stops taking a repo Operator ruling, and `CONTEXT.md` conformance rather than a preference: `daemon project add owner/repo` has violated "repo is never a verb argument" since it shipped. - **T6 — impl.** `add` takes no positional and registers the invoking instance's bound repo; a supplied repo is **refused by name** under `REQ-VERB-ARITY-REFUSE` rather than ignored; re-adding an already-registered repo is idempotent and says so. - **T7 — impl.** `remove` keeps its optional explicit ref, defaulting to the bound repo. The asymmetry is deliberate and stated in the help: `add` is a claim a shell makes about itself, `remove` is cleanup that outlives its subject — an archived, renamed or deleted repo can no longer be bound, and a bound-repo-only `remove` would strand a dead entry the daemon polls every cycle with no verb able to reach it. - **T8 — unit.** The refusal names the token. `remove` with a foreign ref still works. - `REQ-DAEMON-PROJECT-BOUND-REPO` → `["doc", "impl", "unit"]`. ## Docs `CONTEXT.md` carries the doc-stage evidence for all four requirements in this wave: the ack rule and the alias rule in the **Alchemy** agent-surface section, the registry-verb arity in the **Hub Daemon** paragraph that already describes `daemon project`. The **Target** term landed at plan time (it is language, not behavior, and W34 needs it defined). `manifest.toml` changes with C — `daemon`'s opts contract loses the repo argument — so the manifest and `runtime.rs`'s arg table move together or `REQ-VOCAB-PARITY` fails, which is the check working. ## Gate `cargo test` green, `./traceable-reqs.exe check` exit 0, **and** the three behaviors driven against the installed binary rather than the build tree: a write ack naming the repo, a fresh bind reporting its own rename, and `daemon project add` refusing an argument. Verify against what shipped or it is not verified — v0.7.0 through v0.7.2 are the standing evidence for that rule. ## Not in this wave - **#23** — routed to spt-core, rests in Cut, lives on as `BigscreenVR/spt-bs-releases#2` under doyle's ruling with a persistent Watch carrying its outcome back. Seamless update is theirs to design; what changes for us when it lands is that the alias-freed-by-teardown argument in A decays. - **#25** — untouched, and it keeps its precedence over #26 (W33). - No verb edits a Request's or Milestone's **body** from the shell; #38's body was written with `gh` because `milestone create` is single-line and there is no `edit` verb. Not minted yet — raise it with the Operator before it becomes a habit.