mcpx: Phase 3 of #1834 — reviewable sampling in mcpx elicit + matrix/register truth-up + nix#256 evidence

Phase 3 of #1834 — reviewable sampling in `mcpx elicit` + truth-up + nix#256 evidence

Design: #1834 (mcpx sampling trust boundary). Design doc: files/designs/code-mcpx-sampling-trust-boundary.md in the keep-the-four-repos-healthy goal workspace. Origin: dezren39/darwin-nix#256. Gated on Phase 2 of #1834 (the rendering shows the validated shapes, and the truth-up needs both earlier phases' as-built state).

Problem

A person cannot review a sampling request, because the prompt is not shown. Sampling questions are broker rows whose Message is the request's system prompt (or "<server> asks for a model completion") and whose Schema field carries the full CreateMessageRequest — but renderQuestion (pkgs/mcpx/internal/cli/elicit.go) renders Schema as a JSON Schema: describeSchema looks for properties, a sampling request has none, so the messages render as nothing and the answer example offered is '{}'. elicit.Route (internal/elicit/elicit.go ~L502-503) also sends every Sample question to the agent audience unconditionally. Gaps: sampling-app-review-ui, sampling-app-view-edit-prompts, sampling-app-review-responses, sampling-client-user-approval-controls in pkgs/mcpx/internal/conformance/coverage_clientfeatures.go.

Scope — the existing surface learns sampling (no new UI)

renderQuestion for a Sample question renders it as sampling: requesting server, system prompt, and the message list (role + content per message, text verbatim) — not via describeSchema. mcpx elicit show gives the full request JSON.
The answer example for a Sample question is a real CreateMessageResult skeleton (role/model/content/stopReason), not '{}'.
CLI help/docs for Sample questions state the review story plainly: a pending question can be inspected and answered/overridden by a person through list/show/answer; answered results are recorded on the broker row and visible via mcpx elicit result <id>. Do NOT claim a pre-delivery staging state exists — building one is an open question on mcpx: design — sampling trust boundary (reviewable, validated, honestly error-coded) #1834, not part of this phase.
Existing approval controls (audience routing, policy, decline) stay as they are; no new gate is invented.

Truth-up

pkgs/mcpx/docs/spec/conformance-matrix.md rows for all seven scoped #256 gaps updated to as-built — retired, or explicitly partial per row where the as-built state only partially meets the row (the review rows in particular: claim only what the rendering/routing changes actually deliver).
Compare-register sampling rows touched by this design (SMP-05, SMP-06, SMP-10 in pkgs/mcpx/docs/compare/register/sampling.md) updated to as-built.
Completion evidence comment on mcpx: sampling relay -- human review, validation of results, error mapping (8 conformance rows) dezren39/darwin-nix#256 against each of its four acceptance bullets with the landing PRs; close nix#256 if and only if every bullet is met, naming any that is not.

Out of scope: validation or error-code changes (Phases 1–2), a pre-delivery staging hold, sampling rate limits (nix#257), elicitation rendering (that is #1675 Phase 3's ground — reuse its patterns, don't redo it).

Acceptance criteria

mcpx elicit list/show on a pending sampling question displays the server, system prompt and actual messages; the offered answer skeleton is a valid CreateMessageResult shape.
Matrix rows for the seven scoped gaps and register rows SMP-05/SMP-06/SMP-10 match the code as built (partials stated as partials).
mcpx: sampling relay -- human review, validation of results, error mapping (8 conformance rows) dezren39/darwin-nix#256 carries completion evidence against each acceptance bullet and is closed if and only if all are met.

CODE-766

Phase 3 of design #1834, filed by the design architect (2026-10-28), status/blocked gated on Phase 2 (#1836). Triage packet enqueued to repo-health-code-pilot as task 2720 (verified: queue=repo-health-code-pilot, status=queued, last_error=null).

🤖 Agent note · handed off

Agent: queue worker (repo-worker-code), lane B shift
Session / ID: queue task 2853 (triage of mcpx: sampling relay -- human review, validation of results, error mapping (8 conformance rows) #1955)
Found via: queue claim (triage of the relocated mcpx: sampling relay -- human review, validation of results, error mapping (8 conformance rows) #1955)
Task: mcpx: sampling relay -- human review, validation of results, error mapping (8 conformance rows) #1955 (the mcpx: sampling relay -- human review, validation of results, error mapping (8 conformance rows) dezren39/darwin-nix#256 relocation) has been closed as a duplicate of the mcpx: design — sampling trust boundary (reviewable, validated, honestly error-coded) #1834 set. This issue (mcpx: Phase 3 of #1834 — reviewable sampling in mcpx elicit + matrix/register truth-up + nix#256 evidence #1837) is the named home for its human-review acceptance item, and its truth-up already carries the nix#256 evidence; mcpx: Phase 1 of #1834 — typed sampling errors (-1 on rejection, -32602 tool-shape, expiry stays -32603) #1835 covers the error-mapping rows and mcpx: Phase 2 of #1834 — sampling message + result validation at the broker (pairing, era shape, no silent fill-in) #1836 the validation rows. No scope moved or shrunk — the pointer is recorded on mcpx: sampling relay -- human review, validation of results, error mapping (8 conformance rows) #1955 and mcpx: design — sampling trust boundary (reviewable, validated, honestly error-coded) #1834.
Expectation: Nothing new asked of this issue; this note is the dedupe pointer from the mcpx: sampling relay -- human review, validation of results, error mapping (8 conformance rows) #1955 side.
Links: mcpx: sampling relay -- human review, validation of results, error mapping (8 conformance rows) #1955 (closed, duplicate), mcpx: design — sampling trust boundary (reviewable, validated, honestly error-coded) #1834, mcpx: Phase 1 of #1834 — typed sampling errors (-1 on rejection, -32602 tool-shape, expiry stays -32603) #1835, mcpx: Phase 2 of #1834 — sampling message + result validation at the broker (pairing, era shape, no silent fill-in) #1836

🤖 Agent note · picked up

Agent: queue worker (repo-worker-code), Lane D shift
Session / ID: queue task 2720
Found via: queue claim
Task: Triage mcpx: Phase 3 of #1834 — reviewable sampling in mcpx elicit + matrix/register truth-up + nix#256 evidence #1837 (mcpx Phase 3 of design mcpx: design — sampling trust boundary (reviewable, validated, honestly error-coded) #1834) — confirm as-built scope of reviewable sampling in mcpx elicit + matrix/register truth-up + nix#256 evidence, settle the non-text content-block rendering shape, and enqueue the implementation follow-up. No implementation in this task.
Expectation: effort m (triage) — verdict comment on mcpx: Phase 3 of #1834 — reviewable sampling in mcpx elicit + matrix/register truth-up + nix#256 evidence #1837 + implementation packet enqueued, or a named blocker if Phase 2's as-built state is not yet available
Links: issue mcpx: Phase 3 of #1834 — reviewable sampling in mcpx elicit + matrix/register truth-up + nix#256 evidence #1837 · design mcpx: design — sampling trust boundary (reviewable, validated, honestly error-coded) #1834 · Phase 1 mcpx: Phase 1 of #1834 — typed sampling errors (-1 on rejection, -32602 tool-shape, expiry stays -32603) #1835 · Phase 2 mcpx: Phase 2 of #1834 — sampling message + result validation at the broker (pairing, era shape, no silent fill-in) #1836 · mcpx: sampling relay -- human review, validation of results, error mapping (8 conformance rows) dezren39/darwin-nix#256

🤖 Agent note · triage verdict — BLOCKED on Phase 2 landing

Agent: queue worker (repo-worker-code), Lane D shift
Session / ID: queue task 2720
Task: Triage mcpx: Phase 3 of #1834 — reviewable sampling in mcpx elicit + matrix/register truth-up + nix#256 evidence #1837 (mcpx Phase 3 of mcpx: design — sampling trust boundary (reviewable, validated, honestly error-coded) #1834 — reviewable sampling in mcpx elicit + matrix/register truth-up + nix#256 evidence)
Links: mcpx: design — sampling trust boundary (reviewable, validated, honestly error-coded) #1834 · Phase 1 mcpx: Phase 1 of #1834 — typed sampling errors (-1 on rejection, -32602 tool-shape, expiry stays -32603) #1835 · Phase 2 mcpx: Phase 2 of #1834 — sampling message + result validation at the broker (pairing, era shape, no silent fill-in) #1836 · mcpx: sampling relay -- human review, validation of results, error mapping (8 conformance rows) dezren39/darwin-nix#256

Verdict: scope confirmed as written, but Phase 3 is not yet triage-completable — its gate (Phase 2 as-built state) does not exist. This task is blocked on #1836's implementation landing; status/blocked remains, truthfully.

What triage verified fresh (2026-10-29 ~03:58Z):

The Phase 3 problem is confirmed unchanged on current main. pkgs/mcpx/internal/cli/elicit.go: renderQuestion (~L204) still feeds the question's Schema to describeSchema (~L233), which looks up properties (~L242) — a sampling request carries a full CreateMessageRequest there, not a JSON Schema, so its messages render as nothing; exampleFor (~L273) is the path that offers '{}' for such a question. pkgs/mcpx/internal/elicit/elicit.go Route still returns ToAgent unconditionally for Mode == Sample. No superseding PR or competing branch was found for this work.
The Phase 2 gate is not met. mcpx: Phase 1 of #1834 — typed sampling errors (-1 on rejection, -32602 tool-shape, expiry stays -32603) #1835 (Phase 1) and mcpx: Phase 2 of #1834 — sampling message + result validation at the broker (pairing, era shape, no silent fill-in) #1836 (Phase 2) are both OPEN; their triage tasks were only claimed today (2026-10-29T03:53Z, mcpx: Phase 2 of #1834 — sampling message + result validation at the broker (pairing, era shape, no silent fill-in) #1836 = task 2719, Lane C) and neither phase's implementation has landed. There is therefore no Phase 2 as-built state to confirm the rendering scope or the truth-up rows against, and no landing PRs to cite as completion evidence. Enqueueing the Phase 3 implementation packet now would mean guessing the as-built shapes it must match — so no follow-up packet is enqueued in this pass, deliberately, and the issue's acceptance criteria are left fully intact (no silent shrink). Packet scope also excludes unblocking this phase before Phase 2 lands.
nix#256 state correction: mcpx: sampling relay -- human review, validation of results, error mapping (8 conformance rows) dezren39/darwin-nix#256 is CLOSED, but its closure (2026-10-28T22:44Z) was a relocation ("Moved to mcpx: sampling relay -- human review, validation of results, error mapping (8 conformance rows) #1955"), not an acceptance-met closure — its four acceptance bullets remain unchecked, and mcpx: sampling relay -- human review, validation of results, error mapping (8 conformance rows) #1955 was itself closed as a duplicate of the mcpx: design — sampling trust boundary (reviewable, validated, honestly error-coded) #1834 set with this issue as the named home for the human-review item. The completion-evidence comment on nix#256 is therefore still owed at Phase 3 completion, and its closed state must not be read as "bullets met."
Settled by this triage — non-text content blocks in the rendered message list: text blocks render verbatim; each non-text block renders as a one-line typed placeholder in message order giving the block type and mime/source kind (e.g. [image content: image/png — payload not inlined]), with the payload/base64 never inlined and the block never silently dropped. The reviewer sees the full request shape, the terminal stays readable, and no binary or credential-bearing payload spills into the CLI surface. This matches the design doc's content-shape terms in files/designs/code-mcpx-sampling-trust-boundary.md.

Resume condition: when #1836's implementation PR has merged, re-run this triage — confirm the Phase 3 scope against Phase 2's as-built state, then enqueue the Phase 3 implementation packet with the acceptance criteria above intact (rendering + CreateMessageResult skeleton + honest review-story docs + seven-gap matrix truth-up + SMP-05/SMP-06/SMP-10 + nix#256 evidence comment).

Gate check from the #1834 triage (task 7749): Phase 3's blocker is satisfied.

mcpx: Phase 2 of #1834 — sampling message + result validation at the broker (pairing, era shape, no silent fill-in) #1836 / PR fix(mcpx): validate sampling messages and results at the broker (#1836) #2443 merged 2026-10-30 and mcpx: Phase 2 of #1834 — sampling message + result validation at the broker (pairing, era shape, no silent fill-in) #1836 was closed 2026-11-03; the validator is on main (pkgs/mcpx/internal/elicit/sampling.go:158 ValidateSampleRequest, :213 ValidateSampleResult; wired at pkgs/mcpx/internal/daemon/hooks.go:329,394).
Phase 3's problem is still present on main d3476bce7: pkgs/mcpx/internal/cli/elicit.go:268 renders through describeSchema (no sampling rendering, '{}' answer skeleton) and internal/elicit/elicit.go:831-834 routes Sample to the agent unconditionally. So this issue's scope is unchanged and now unblocked.
Queue task 2720 (triage of this issue) is blocked with the resume condition "Resume when mcpx: Phase 2 of #1834 — sampling message + result validation at the broker (pairing, era shape, no silent fill-in) #1836 implementation merges" — that condition is met; it needs unblocking/requeueing so the Phase 3 implementation packet can be enqueued.
No label change made here (the owning task/lane keeps this issue's state); the mcpx: design — sampling trust boundary (reviewable, validated, honestly error-coded) #1834 epic stays open as tracker until this lands.

agent-note:v1
state: reviewed (coverage loop-back)
Coverage pass 2026-11-03, loop-back review: this issue was re-reviewed (drift / overlap / already-handled / stale / blockers / ordering) so every open issue in the four repos is reviewed and in the system. Verdict: still live; in the system as task 2720 (triage, blocked, priority 30) in repo-health-code-pilot. Labels audited and synced. If it is already handled at current main, its triage closes it with evidence; if oversized, the triage files the split; blockers belong on this issue.
Found via: coverage loop-back 2026-11-03

Navigation Menu

Repository navigation

mcpx: Phase 3 of #1834 — reviewable sampling in mcpx elicit + matrix/register truth-up + nix#256 evidence #1837

Phase 3 of #1834 — reviewable sampling in `mcpx elicit` + truth-up + nix#256 evidence

Problem

Scope — the existing surface learns sampling (no new UI)

Truth-up

Acceptance criteria

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

mcpx: Phase 3 of #1834 — reviewable sampling in mcpx elicit + matrix/register truth-up + nix#256 evidence #1837

Description

Phase 3 of #1834 — reviewable sampling in mcpx elicit + truth-up + nix#256 evidence

Problem

Scope — the existing surface learns sampling (no new UI)

Truth-up

Acceptance criteria

Activity

developing-today commented on Oct 28, 2026

dezren39 commented on Oct 28, 2026

dezren39 commented on Oct 28, 2026

dezren39 commented on Oct 29, 2026

dezren39 commented on Oct 29, 2026

dezren39 commented on Nov 3, 2026

developing-today commented on Nov 3, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Phase 3 of #1834 — reviewable sampling in `mcpx elicit` + truth-up + nix#256 evidence