feat(workflows): add catalog launch guard command workflow

[ShadReyes]

What changed

• Added catalog-launch-guard (revision 2026-07-31-v3-deterministic-artifacts), a read-only post-import catalog QA workflow: deterministic identity reconciliation against a full cursor walk of the v202604 catalog, a structural audit (lifecycle, prices/currencies, options, variants, SKUs, imagery, routes, bundles, subscriptions), a stratified flag review that enforces a ≤10% false-positive bar before GREEN, a saved RAG launch dashboard with a lifecycle breakdown, and an under-350-word CEO page brief.
• Deterministic verification: audit steps author and execute their own Node verification script (write_file + run_cli npm exec) over catalog captures persisted page-by-page to per-run artifact files — every count, list, and hash in the report is script output, never a model tally. The script skeleton is false-pass-proof (single ok:true/false JSON summary read back from a file, non-zero exit on corrupt/partial input, capped lists with truncated flags, integrity block with SHA-256s) and was validated to reproduce 36/36 planted defect counts exactly on a synthetic 1,200-SKU catalog in ~0.2s with ~131× less context than raw records.
• Run hardening from a live production-company test: per-run artifact directories (context.run_prefix or self-minted) so concurrent runs cannot collide; QA reviewers verify from persisted artifacts instead of re-walking the live catalog; mid-audit catalog drift downgrades the gate to AMBER with the exact delta instead of failing a correct run; the CEO brief travels as an artifact file (step outputs truncate at 12k chars); and the final gate reports a measured wall-clock run window (labeled as including QA/review time) instead of an invented speed multiplier.
• Registered the workflow in the public Mist skills manifest.

Sales pitch

"Launching your store used to mean two people spending a week hunting for broken products and hoping they caught everything. Mist now checks all of it — every price, image, SKU, and bundle — in minutes, shows you exactly what's actually broken with the evidence attached, and gives you a green light you can trust. Days of QA become one approval click."

Why it changed

Fluid Mist hackathon deliverable. Post-import catalog verification normally costs a launch team days of manual QA; this workflow turns it into one evidence-backed approval gate that any Fluid company can run. It distinguishes real launch blockers from legitimate merchandising structures, audits 100% of records rather than a sample, measures its own false-positive rate, and never mutates a commerce record — every fix remains a merchant-approved action.

How to test

1. Run jq empty manifest.json workflows/catalog-launch-guard.workflow.json; expect no errors.
2. Run python3 scripts/validate_catalog.py; expect 42 skills and 8 workflows.
3. Run python3 -m unittest discover -s scripts -p 'test_*.py'; expect 16 passing tests.
4. Parse the workflow through Mist Desktop's WorkflowDefinitionSchema; expect 5 steps with no stripped authored fields.
5. Search the workflow file for company-specific names, products, IDs, or currencies; expect no matches.

Known limitations

The workflow proposes fixes but never merges, deletes, archives, reprices, publishes, or changes product relationships; executing a proposed fix remains a separately approved action.

Linear

N/A — Fluid Mist hackathon deliverable.