T-I7N0-agent-manager-screenshots-docs-ledger

State: planning/proposed · Impact: medium · Complexity: medium

Goal

A milestone this UI-heavy ships across many PRs, and screenshots, the in-app manual and library ledgers drift the moment a shipped surface outpaces its documentation. This task is the standing companion that keeps them true stage by stage, rather than a catch-up pass batched at the end.

Today

Location	Role today
solutions/determined/products/agent-pants/tests/e2e/capture.e2e.ts	Existing Playwright spec producing the experience’s screenshot sets
solutions/determined/experiences/agent-pants/docs/screenshots/	Per-surface contact sheets (design/, skill-page/, dialogs/, plugin-page/, marketplace-page/, each an index.html)
solutions/determined/experiences/agent-pants/docs/getting-started.md	In-app manual chapter; per the workflow-and-ux assessment’s P1-5, claims the app “publishes nothing”
solutions/determined/experiences/agent-pants/docs/current-limits.md	Correctly documents what Install/Build/Import do today
solutions/determined/experiences/agent-pants/docs/authoring.md	Authoring chapter; P1-5 flags it as also understating publishing
solutions/determined/experiences/agent-pants/docs/publishing.md	Publishing chapter
docs/LIBRARIES.md	Repo-wide library roster; no row yet for code-editor or the harness capability matrix
solutions/agent-pants/docs/vault/design/publishing-model.md	Design doc; carries the surface-by-kind table T-K0Y5-every-kind-every-surface-audit extends
solutions/agent-pants/docs/vault/design/application-over-vault.md	Design doc for the application-over-vault redesign
.github/workflows/agent-pants-ci.yml	Path filters cover solutions/agent-pants/** only; solutions/determined/experiences/agent-pants/** and solutions/determined/products/agent-pants/** are absent
.github/workflows/ci.yml	Runs moon run product-agent-pants:check (typecheck) but no step runs the agent-pants Playwright projects

Proposed

capture.e2e.ts gains a screenshot case for every new surface the milestone’s other tasks ship, each landing an S-0012 contact sheet. The four in-app manual chapters are rewritten in place per stage so they never claim less than the app does. docs/LIBRARIES.md carries rows for the code-editor package (tier 0) and the harness capability matrix. publishing-model.md and application-over-vault.md stay current with what shipped. A CI path filter covers solutions/determined/experiences/agent-pants/** and solutions/determined/products/agent-pants/**, and the agent-pants Playwright projects run in CI, not only locally.

Approach

1. After each stage’s functional tasks land, add or extend a capture.e2e.ts case for every new surface, producing light and dark screenshots into docs/screenshots/<surface>/.
2. Regenerate or hand-update each surface’s index.html contact sheet per S-0012-human-reviewable-ui-changes.
3. Rewrite getting-started.md, current-limits.md, authoring.md and publishing.md in place against the shipped state after each stage.
4. Add the code-editor tier-0 row and a capability-matrix row to docs/LIBRARIES.md once T-MVW5-code-editor-package and T-HLEW-harness-capability-matrix ship.
5. Update publishing-model.md and application-over-vault.md in place as their tracked surfaces ship.
6. Add a path-filtered CI workflow (or extend agent-pants-ci.yml’s filter) covering solutions/determined/experiences/agent-pants/** and solutions/determined/products/agent-pants/**, with a step running the agent-pants Playwright projects.
7. Verify per stage: moon run :check :test and the agent-pants Playwright projects pass in CI.

Areas

Area	Note
solutions/determined/products/agent-pants/tests/e2e/capture.e2e.ts	New screenshot cases per stage
solutions/determined/experiences/agent-pants/docs/screenshots/	New and updated per-surface contact sheets
solutions/determined/experiences/agent-pants/docs/getting-started.md	Keep current with shipped publishing behavior
solutions/determined/experiences/agent-pants/docs/current-limits.md	Keep current per stage
solutions/determined/experiences/agent-pants/docs/authoring.md	Keep current per stage
solutions/determined/experiences/agent-pants/docs/publishing.md	Keep current per stage
docs/LIBRARIES.md	Add the code-editor tier-0 row and a capability-matrix row
solutions/agent-pants/docs/vault/design/publishing-model.md	Keep current with shipped surfaces
solutions/agent-pants/docs/vault/design/application-over-vault.md	Keep current with shipped surfaces
.github/workflows/agent-pants-ci.yml	Extend the path filter, or add a sibling workflow for the determined experience/product

Acceptance criteria

• AC-1: Every new UI surface shipped by another Stage 0-7 task has a capture.e2e.ts case and an S-0012 contact sheet entry, light and dark.
• AC-2: The four manual chapters make no claim that understates a shipped feature, spot checked against the assessment’s P1-5 wording each time a publishing-related task ships.
• AC-3: docs/LIBRARIES.md has a code-editor row and a capability-matrix row once their tasks close.
• AC-4: A GitHub Actions path filter covers solutions/determined/experiences/agent-pants/** and solutions/determined/products/agent-pants/**, and a CI run shows the agent-pants Playwright projects executing.
• AC-5: moon run :check :test passes for every commit this task lands.

Out of scope

• The screenshot content and documentation decisions themselves belong to each shipping task; this task is the ledger and the CI/doc plumbing, not the design review.
• Entries under solutions/agent-pants/docs/vault/decisions/; this task touches design docs, not decisions.

Dependencies

• Runs alongside every Stage 0-7 task in M-BMFD; no single task blocks its start, and each iteration follows that stage’s functional tasks landing first.

Discovery context

From the user’s Agent Manager planning request (2026-09-07) and S-0012 (screenshots for every touched UI surface, light and dark). The CI and manual gaps are documented in solutions/agent-pants/docs/vault/design/workflow-and-ux-assessment.md as P1-5 (manual understates publishing), P1-6 (the Playwright suites never run in CI) and P1-7 (no CI gate covers the determined experience or product), tracked there under the gate ci-covers-determined-agent-pants.

---

← Back to Tasks