T-K0Y5-every-kind-every-surface-audit
State: planning/proposed · Impact: medium · Complexity: medium
The kind × surface story has grown incrementally — a kind added to the wizard is not always reachable in the preview, the related tab, or the CLI — and nothing drives every kind through every surface in one pass to find the gaps. This task runs after the Stage 3 layout tasks land, extends the publishing-model table with the new surfaces, and closes what it finds.
| Location | Role today |
|---|---|
solutions/agent-pants/docs/vault/design/publishing-model.md | A three-column table (Edited / Created by the wizard / Published) for the five kinds; no columns for tree, preview, related, build or CLI |
solutions/determined/tests/fixtures/agent-pants-freeform/ | The fixture vault: two marketplaces, plugins carrying skills, commands and agents, a shared skill, and reference documents |
solutions/determined/products/agent-pants/tests/e2e/ | Five specs (walkthrough, walkthrough-dialogs, authoring, install, capture), none organized as a kind × surface matrix |
solutions/agent-pants/packages/agent-pants/src/commands/ | publish, marketplace (build/check), list, adopt, status — the CLI’s own surface over the same five kinds |
solutions/agent-pants/docs/vault/reference/harness-comparison.md | Records which harnesses support commands and agents at all, so a deliberate refusal (Cursor commands, Amp agents) has a source to cite rather than being asserted |
Proposed
Section titled “Proposed”publishing-model.md’s kind table grows a column per new surface — Tree, Preview,
Related, Build, CLI — alongside the existing three. Each cell is populated by walking
that kind through that surface against the fixture vault. A gap this task can close
(a CLI list omission, a missing preview branch) is fixed here; a gap that needs a
Stage 3 task not yet closed is cross-linked rather than worked around. A deliberate
refusal (Cursor commands, Amp agents — no canonical surface for either) is recorded as
a refusal in the table, not left blank.
Approach
Section titled “Approach”- Extend
publishing-model.md’s kind table with Tree, Preview, Related, Build and CLI columns, keeping Edited/Wizard/Published as they stand. - For each of the five kinds, walk it through every surface against
tests/fixtures/agent-pants-freeform: open it in the tree, edit it, create one via the wizard, view its preview, view its related tab, publish or build it, and list/build/check it via the CLI. - Record each cell’s finding; where a surface refuses a kind on purpose (a marketplace
has no per-destination preview; a command has no Cursor renderer), the cell names
the refusal and cites
harness-comparison.mdrather than being left blank. - Fix a gap that is this task’s own to close; file a backlog entry
(
sdlc backlog create) and cross-link it in the table for a gap that belongs to a Stage 3 task already closed by the time this runs, or leave the cell pointing at an open one. - Add a Playwright spec that drives the five-kind × surface matrix against the fixture vault, so a future regression is a named test failure, not a stale table cell.
| Area | Note |
|---|---|
solutions/agent-pants/docs/vault/design/publishing-model.md | Extend the kind table with the new surface columns |
solutions/determined/products/agent-pants/tests/e2e/ | Add a spec covering the kind × surface matrix against the fixture vault |
solutions/agent-pants/packages/agent-pants/src/commands/ | Close whatever CLI-side gap the audit finds |
Acceptance criteria
Section titled “Acceptance criteria”- AC-1:
publishing-model.md’s kind table has a column for each of Tree, Preview, Related, Build and CLI, with no blank cell — every cell states support or names a refusal. - AC-2: Every one of the five kinds (
skill,agent,command,plugin,marketplace) is exercised on every surface in the new Playwright spec againsttests/fixtures/agent-pants-freeform. - AC-3: A gap this task fixes has a passing test that would have failed before the fix; a gap it defers is filed and cross-linked from the table, not silently dropped.
- AC-4:
moon run product-agent-pants:checkand the agent-pants Playwright projects pass.
Out of scope
Section titled “Out of scope”- Building a canonical surface for a harness that genuinely has none (Cursor commands, Amp agents) — these are recorded as deliberate refusals, not built.
- MCP servers, hooks conversion tables, and per-target override warnings — those are T-H56N-hooks-as-vault-entities, T-0AF7-hook-editor-and-harness-gap-table and T-IIZO-per-target-overrides’s work, read here but not re-audited.
Dependencies
Section titled “Dependencies”- T-GHE3-edit-form-in-center-pane, T-K1A2-preview-pane-per-entity-and-target, T-7UDE-related-files-tab and T-H3RG-publish-as-secondary-view must close first — this audit exercises the tree/editor/preview/related/publish surfaces those tasks build, and cannot find real gaps in surfaces that do not exist yet.
Discovery context
Section titled “Discovery context”From the user’s planning request of 2026-09-07: “every kind is served by tree, editor, wizard, preview, related, publish, build and CLI.” Sequenced last in Stage 3 per the milestone plan’s own note that this task “runs after the Stage 3 layout tasks.”
Depends on
Section titled “Depends on”T-GHE3-edit-form-in-center-pane, T-K1A2-preview-pane-per-entity-and-target, T-7UDE-related-files-tab, T-H3RG-publish-as-secondary-view