Before anything else: read $HOME/.agentic-workflow/toolkit/skills/_preamble.md and follow it (skill index, provider capability map, bootstrap check, session context). Run its Session Close section when this skill finishes.

---

Verify App — Platform Dispatcher

Detects whether this is a web or iOS project and delegates to the appropriate verification skill. Contains no verification logic — all execution lives in the sub-skills.

Tip: If you already know the platform, invoke directly: /verify-web or /verify-ios

Platform Detection & Dispatch

Follow _shared/platform-detection.md — canonical detection globs (with vendored-path excludes), the resolution table, and the dispatch contract. Resolve it from the stable toolkit path:

SHARED_DIR="$HOME/.agentic-workflow/toolkit/skills/_shared"
ls "$SHARED_DIR/platform-detection.md"

If resolution or ls fails, stop and report — do not guess the platform. Otherwise read the file and apply it: iOS → verify-ios, web → verify-web, both/neither → Ask the user. Echo dispatch: <sub-skill> args=<args> before dispatching, then dispatch literally: Invoke skill <sub-skill> with args "<original args verbatim>".

Argument Spec (pass-through)

All arguments are forwarded to the sub-skill unchanged — never drop or rewrite one:

Argument	Meaning
--journey <path>	Path to a verification-plan.md (template: _shared/verification-plan-template.md) — the sub-skill executes its journeys
--lenses <csv>	Narrow the lens set (catalog: _shared/verification-lenses.md)
--baseline	Require visual diffing against screens.json baselines — the visual lens fails if none exist
--yes	Non-interactive: the sub-skill auto-approves its plan
--base-url <url>	Override app-URL detection (web only; still forwarded verbatim)
--visual	Include the visual lens
auto / criteria	Diff-inference mode / explicit verification criteria

If --journey is given but the path does not exist, stop and report — do not dispatch with a blank or dead required arg.

Sub-skill Return Contract

The sub-skill must produce an evidence pack (_shared/evidence-pack.md) and report back {verdict, evidence_path}:

• verdict ∈ PASS | WARN | FAIL
• evidence_path = absolute path to the verification/<run-id>/ dir containing pack.json

After the sub-skill returns, confirm the pack exists:

SHARED_DIR="$HOME/.agentic-workflow/toolkit/skills/_shared"
source "$SHARED_DIR/repo-slug.sh"
ls -1dt "$AW_DIR/verification/"*/ | head -1
ls "$(ls -1dt "$AW_DIR/verification/"*/ | head -1)pack.json"

If the sub-skill returned no {verdict, evidence_path}, or the newest run dir has no pack.json, fail loudly:

"verify sub-skill returned no evidence pack — verification is not proven. Treat this run as FAIL."

Relay the sub-skill's verdict and evidence_path verbatim in your final report. A FAIL verdict blocks PR-open / mark-ready (gate rule in _shared/evidence-pack.md).

Auto-chain: design-verify

After a completed run, check whether screens.json baselines cover the screens the sub-skill verified:

SHARED_DIR="$HOME/.agentic-workflow/toolkit/skills/_shared"
source "$SHARED_DIR/repo-slug.sh"
ls "$AW_DIR/design/screens.json" 2>/dev/null || echo "no screens.json — skip design-verify chain"

If screens.json exists and has baselines entries for one or more of the verified screens, invoke Invoke skill design-verify with args "<verified screens>" automatically and report its diff verdict alongside the verification verdict. If no baselines cover the verified screens, skip the chain silently.

Next steps

• /design-verify — re-run design diffing on demand (auto-chained above when baselines cover verified screens)
• /bugReport — if functional gaps surfaced, capture them as a bug report