refactor(host)!: retire the skill.invoke RPC for the gesture boundary

Invocation is an ordinary session.prompt again: the pre-step gesture
boundary makes it deterministic host-side for every front end, so the
dedicated RPC (handler, wire schema, error codes, client face, fixtures)
and ui-skill's claim machinery are net deletions. The menu keeps decision
21 exactly — a pick lands literal /name text — plus the user-only marker
from skill.list's modelInvocable flag.

.agents/notes/implemented/feature/2026-08-08-user-explicit-skill-invocation.md

@@ -1,4 +1,4 @@
-# Agent Note: User-explicit skill invocation over skill.invoke
+# Agent Note: User-explicit skill invocation at the pre-step gesture boundary
 
 Status: implemented
 
@@ -10,28 +10,27 @@ A `disable-model-invocation: true` skill is user-only by design: it never enters
 
 ## Decision
 
-User-explicit invocation is a deterministic host-side injection, uniform for every user-invocable skill:
+User-explicit invocation is a host-side pre-step injection, uniform for every user-invocable skill and every front end:
 
-- `skill.invoke { sessionId, name, text? }` (host apiproxy) enforces user-invocation policy at the operation boundary (`skill-not-found` / `skill-not-invocable`), renders the skill with the shared `renderSkillContent`, appends the optional trailing text after a blank line, and injects the whole as one user-role message carrying the new `skill-invocation` `MessageSource` kind (`{ name, args? }`) before starting a turn through the same route-served gate as `session.prompt`.
-- `renderSkillContent` moved from `dsh-tool-skill` to the `dsh-skill` seam: the `skill` tool result and the injection share one verbatim `<skill_content>` shape, and the catalog text gained the seam rule — an inline-injected skill must be followed, not re-loaded through the tool.
-- `skill.list` serves every user-invocable skill and carries `modelInvocable`, so the browser menu lists user-only skills with a marker (description prefix — the `hint` field is claim-state ghost text the menu never renders).
-- ui-skill claims a menu pick or an entered `/name [args]` into the invoke transaction (`matchEnter` strong-waits the catalog; unknown names stay plain prompts). The unreached legacy `<skill>name</skill>` reference codec is removed.
-- The transcript materializes the injection as a dedicated `skill-invocation` node from source metadata (never re-parsed from the body) and renders a right-aligned bubble: `/name` chip, trailing text, and the injected block collapsed behind a disclosure.
+- `dsh-tool-skill` registers a second `agent/pre-step` listener (beside its catalog listener, the same seam `workspace-instructions` and the runtime-context snapshot ride): it scans the step's claimed messages for whitespace-bounded `/name` tokens — anywhere in the text, the same word-boundary shape the transcript chip decoration uses — collects first-seen-deduplicated names, loads each through `ctx.skills.get`, checks `isUserInvocable` on the loaded definition (the single lookup that produces what is injected), renders it with the shared `renderSkillContent`, and appends the injections after every other injection of the step: background first (workspace rules, runtime policy, catalog), the material the model must act on last, closest to its answer. Registration order pins the placement — the gesture listener registers before the catalog listener, so the waterfall hands it the catalog-bearing list to extend.
+- Precision is closed-set matching, exactly like slash commands: `/goal` resolves against the command registry, `/name` against the workspace's user-invocable skill directory; a miss stays ordinary prose, so nothing is ever guessed. Only `source.kind === 'user'` messages are scanned — external text cannot forge a gesture. Paths (`/usr/bin`), fractions (`5/8`), and prefixed tokens (`foo/name`) all break the boundary.
+- The client stays decision 21: a menu pick lands the literal `/name ` and the prompt ships it verbatim; ui-skill implements no adjudication hooks and no reference codec. `skill.list` (now the domain's only RPC) serves every user-invocable skill with `modelInvocable` so menus mark user-only entries. A name shared with a host command resolves to the command — adjudication claims the line client-side before it becomes a prompt.
+- The injection is a `user`-role message carrying the `skill-invocation` source (`{ name, form: 'instructions' }`), so `user/message` logging, the context-injection transcript row (labelled with the skill name), and replay all come free; `renderSkillContent` lives in the `dsh-skill` seam, shared verbatim with the `skill` tool result, and the catalog's closing sentence tells the model to follow an injected block instead of re-loading it.
 
-Peer-product survey (Pi, OpenCode, Claude Code, Kimi Code, Codex, DeepSeek-Reasonix — local checkouts) was unanimous: user-explicit triggering is programmatic injection as a user-role message with zero model participation on every product, prompt-guided tool loading exists only on the model-autonomous track, and the disable-model-invocation equivalents gate only the model-side surfaces. Kimi's origin-metadata rendering and the Claude Code/Kimi no-reload prompt rule translate directly onto `MessageSource` and the catalog sentence.
+Peer-product survey (Pi, OpenCode, Claude Code, Kimi Code, Codex, DeepSeek-Reasonix — local checkouts) was unanimous that user-explicit triggering is programmatic injection with zero model participation; the final shape is closest to Codex's core-side `$name` mention scanning, which likewise frees every front end from implementing recognition.
 
 ## Alternatives considered
 
-- **`agent.inject()` context injection** — no peer precedent; the gesture is a user turn, not an environment notice, and context-row presentation, compaction, and attribution all mismatch. Rejected.
+- **`skill.invoke` RPC (host injects, client claims)** — implemented first, in two iterations: a single mixed message (user text folded into the body), then a gesture prompt plus injection delivered through inbox primitives. Rejected after real-session testing: the mixed message polluted the injection with user prose; the two-message form depended on wake-ordering subtleties (`followup` claims the whole next-turn queue synchronously inside the first waking call, stranding any later message in the next turn — reproduced live), and the dedicated RPC duplicated a path `session.prompt` already provides while leaving TUI/ACP to reimplement recognition. The pre-step seam removes the RPC, the claim machinery, and the ordering hazard outright.
+- **`agent.inject()` from the RPC handler** — the inject queue (`next-step`, wake-free) is claimed ahead of the next-turn prompt, putting the injection above the gesture in the log; and pairing it with a waking `followup` reintroduces the same ordering coupling. The pre-step listener injects inside the step assembly, where ordering is explicit.
 - **A host `/skill <name>` command** (command registry, plan-mode precedent) — two-token UX, no name completion, and user-only skills stay undiscoverable in the menu; the per-cwd skill catalog also fits the static command registry poorly. Rejected.
 - **Client-side expansion** (fetch body, splice into the prompt) — authorization becomes bypassable client courtesy, the log loses the invocation semantics, and Codex deleted its equivalent mechanism (custom prompts) in favor of core injection. Rejected.
-- **Host prompt-pipeline scanning for `/name`** (Codex `$name` core mentions) — duplicates the adjudication layer and risks swallowing literal slashes in prose; the claim path already covers the need. Rejected.
-- **Per-injection preamble line** (Kimi's `User activated the skill …`) — dropped in favor of a one-time catalog sentence: same context, paid once, and the injected block stays byte-identical with the tool result.
+- **Structured reference payload on the prompt wire** (Codex's `UserInput::Skill` analogue: the client ships `{skills: [...]}` beside the text and the boundary prefers it over scanning) — considered and deferred: the existing slash-command system is itself line-text on the wire, and closed-set directory matching already removes the guesswork; recorded as a ledger item should gesture precision ever need client intent.
+- **Per-injection preamble line** (Kimi's `User activated the skill …`) — dropped in favor of the one-time catalog sentence: same context, paid once, and the injected block stays byte-identical with the tool result.
 
 ## Consequences
 
-- Decision 21's plain-text reference path is superseded at submission: the draft still carries plain text and lexicon-derived chip visuals, but submit claims into a deterministic injection instead of shipping the literal and hoping. The model-autonomous track (catalog + `skill` tool) is unchanged.
-- Every user-invocable skill invocation now costs its full rendered body unconditionally — the price of determinism the peer survey showed everyone pays.
+- Decision 21's plain-text reference is now the whole client story: the draft carries plain text, chip visuals derive from the lexicon, and the sent text is judged by the host boundary — a hand-typed gesture, a menu pick, and a TUI prompt are indistinguishable and equally deterministic.
+- Every user-invocable skill invocation costs its full rendered body unconditionally — the price of determinism the peer survey showed everyone pays. Mentioning a known skill name mid-sentence loads it; that is the Codex mention semantic, accepted deliberately.
 - The `skill-invocation` source rides `user/message`, so Model-visible ⟺ logged holds with no new event type, and replay/UI read metadata rather than text markers.
-- TUI and ACP can adopt `skill.invoke` later for the same semantics; until then the TUI's client-side expansion remains its own path.
 - Accepted residual of dropping the per-injection preamble: the no-reload framing rides only the catalog, and a workspace whose skills are all user-only never publishes a first catalog — an injection can arrive with no framing at all, and the model may redundantly try the `skill` tool once (the replacement catalog's empty arm carries the sentence; the never-published case does not). Publishing a catalog for framing alone was judged worse than that one recoverable error.