obsidian-test-vault-smoke

Files (5)

• SKILL.md
• openai.yaml · agents/
• cli-runtime.md · references/
• pagelet-smoke.md · references/
• ui-ux-smoke.md · references/

SKILL.md (first lines)

# Obsidian Test Vault Smoke

## Core Rules

- Do not claim Obsidian validation without deploying the current build to `test/` and observing the affected behavior in the running app.
- Treat Jest, lint, type-check, build, and source scans as implementation validation, not app smoke.
- Require visible-window interaction for UI/UX PASS; CLI, DOM, and screenshots support the finding but do not replace real interaction.
- Use `obsidian vault=test <command>` for every Obsidian CLI call. `vault=test` must precede the command.
- Choose the lightest tier that covers the changed surface.
- Follow the mobile environment rule in `AGENTS.md` before choosing a smoke
  skill. Use this skill for general mobile validation with CLI mobile simulator;
  a tool failure does not create a real-device requirement.

## Scope Freeze And Evidence Reuse

Before an expensive gate or app interaction:

1. Classify the latest runtime delta as `code-only`, `runtime`, `visual/layout`,
   `interaction/state`, or `release/broad`.
2. Freeze the observable target. For screenshot-led UI work, state direction,
   order, spacing, viewport, and interaction expectations in testable terms.
   Ask one product question before deployment if the target is ambiguous.
3. Map changed surfaces to the minimum evidence. Mark unaffected surfaces
   `SKIP`; do not run full Pagelet, provider, or cross-surface flows for a
   geometry-only change.
4. Reuse automated checks under `AGENTS.md` **Validation Planning And Reuse**,
   verifying relevant source/tests/fixtures/config and build inputs/results.
   Reuse observed app smoke when its runtime/assets, target vault state, loaded
   build, and acceptance conditions are verifiably unchanged. A new turn alone
   does not invalidate it. Changed or unknown relevant state requires only the
   affected proof; docs/evidence-only edits do not invalidate runtime smoke
   unless they change its acceptance conditions.
5. Freeze runtime files before `make deploy`. If they change afterward, rerun
   only the affected validation gate and app proof.

Stop when every frozen acceptance check has current deployed evidence. Do not
add adjacent gestures, orientations, provider calls, or unrelated surfaces
solely to strengthen an already sufficient proof.

## Validation Gate

Satisfy the complete **Local Validation Gate** from repo-root `AGENTS.md`
before app smoke. Reuse checks covered by the selected enclosing gate with
verified inputs/results: the production build covers type-checking and full
Jest covers the focused suites it includes. Supplement remaining checks,
including the runtime `<style>` / `innerHTML` / `outerHTML` source scan.
For failures, follow `AGENTS.md` **Test Failure Diagnosis**; an unavailable
app/device remains a validation gap, never a PASS.

Treat the source-scan `rg` exit code `1` with no output as PASS. Inspect every match manually. `make deploy` and the hosted community scan do not replace this local source scan.

## Smoke Tiers

| Tier | Use when | Required checks |
| --- | --- | --- |
| `quick` | Narrow code-only or test-only change | Complete Local Validation Gate with focused suites; stop before app smoke only when no runtime/user path changed |
| `app-runtime` | Runtime, command, packaging, Pagelet shell, Chat/Preview/Stats mount, Memory readiness | Local Validation Gate, `make deploy`, plugin reload, affected CLI/DOM probes, fresh console/error capture |
| `full-ui` | Visible UI, CSS/layout/copy, Pagelet workflow, settings, keyboard/focus, mobile emulation | `app-runtime` plus real Obsidian interaction, screenshots, UX notes, and provider/write-path checks when applicable |
| `release-gate` | Release, broad refactor, shared infrastructure | Local Validation Gate, `make deploy`, broad runtime matrix, required UI surfaces, and release evidence reconciliation |

`make deploy` already runs lint, a production build, full Jest, and asset
deployment. If those checks already passed for the current changes, use
`make deploy-current` to verify and copy the current build without repeating
them. Reuse conditions and test groups live in `AGENTS.md` Local Deployment;
build identity alone does not prove tests passed.

## Workflow

1. Inspect `git status --short`, relevant diffs, and affected surfaces.
2. Freeze the observable target and classify the runtime delta.
   Before changing app modes or reloading for smoke, record the actual CLI
   debug and mobile-emulation states and the commands needed to restore them;
   see [CLI runtime smoke](references/cli-runtime.md#setup-and-targeting).
3. Select the deployment path below and complete or reuse its checks together
   with the remaining Local Validation Gate checks from `AGENTS.md`.
4. Select the smoke tier.
5. For app smoke, deploy with `make deploy` below, or replace only that command
   with `make deploy-current` when the reuse conditions above hold, then reload:

```bash
make deploy
obsidian vault=test vault info=path
obsidian vault=test plugin:reload id=personal-assistant
obsidian vault=test plugin id=personal-assistant
```

6. For visible changes, perform the cheapest visual target check immediately
   after reload. Stop on a direction, order, clipping, or viewport mismatch
   before deeper interactions.
7. Read only the references required by the changed surface:
   - For any `app-runtime`, `full-ui`, or `release-gate` run, read [CLI runtime smoke](references/cli-runtime.md).
   - For visible UI/UX work, read [UI/UX interaction smoke](references/ui-ux-smoke.md).
   - For Pagelet work, read both CLI runtime smoke and [Pagelet smoke](references/pagelet-smoke.md).
8. For historical fixtures, regression expectations, and prior evidence, consult the current [Pagelet smoke checklist](../../../docs/development/validation/pagelet-smoke-checklist.md). Treat its verification log as provenance, not current-run evidence.
9. Record concrete `PASS`, `FAIL`, `BLOCKED`, or `SKIP` outcomes.
10. Restore the recorded debug/mobile states after success, failure, or
    interruption; `off` is correct only when that was the initial state.
    Restore mobile first if it triggers a reload, then restore and verify debug.
    Report an unknown initial state or failed restoration without claiming
    cleanup PASS.

## Safety Boundaries

- Limit provider-backed checks to repo-local test-vault fixtures unless the user approves broader data. Report provider/model, note paths, and prompts sent.
- If a provider, browser, CLI, or GUI tool blocks the action, do not bypass the block; report `BLOCKED` and residual risk.
- Stop before deleting or rewriting test-vault data that may be user-authored.
- Stop before publishing, pushing, creating releases, or mutating non-test external systems unless the user authorized that action.
- Keep hosted Obsidian Community scans distinct from the local source scan. A hosted scan submits a ref to an external service and is never an automatic part of this skill.
- During `release-gate`, invoke `obsidian-community-check` only when the user explicitly requested the hosted scan or the active `stable-release` workflow explicitly authorized it. Otherwise report it as not run.

## Output

```markdown
Validation:
- Runtime delta: `<code-only/runtime/visual-layout/interaction-state/release-broad>`
- Frozen target: `<observable acceptance checks>`
- Reused evidence: `<same-state evidence or none>`
- PASS: `<check>` - `<observed result>`
- FAIL: `<path>` - `<regression or product gap>`

From edonyzpc/personal-assistant, AGPL-3.0.

More from edonyzpc/personal-assistant