DD-338 Phase D.5 — marketplace refresh runbook + automation wrapper

[piersdd]

Summary

Final substantive Phase D sub-phase: a thin ESM Node wrapper around the existing scripts/build-catalog.js plus an operational runbook covering the D.2 → D.3 → D.5 → daemon refresh chain.

• scripts/dd-338-refresh-catalog.mjs (200 LOC) — orchestrates snapshot → regenerate → diff → summary. Move-renames dist/catalog.json to dist/catalog.json.prev, runs build-catalog.js as a child process with stdio inherited, then diffs across five granularity axes (scope_filtering, field_projection, deterministic_ordering, audit_surface, domain_scope) plus entry-level tier/readiness and presence-level tool add/remove. Output to dist/dd-338-refresh-summary.json (gitignored) + brief stdout summary.
• docs/dd-338-marketplace-refresh.md — five-step runbook (regenerate → review → commit+push → daemon refresh → verify) with structured-summary shape example + failure-modes table.
• scripts/dd-338-refresh-catalog.test.js — 10 new test cases covering baseline run, identical catalogs, single tool-axis flip, entry tier flip, entry add/remove, tool-presence add/remove, plus three formatSummary rendering cases.

Behaviour

• First run (no dist/catalog.json present): summary marks baseline:true, every entry counted as an addition.
• Steady-state (no source changes): summary reports No catalog changes detected. — verified end-to-end against current main (6f29143): 59 plugins + 11 packs = 70 entries indexed on baseline, empty-diff on re-run.
• Build failure: prev snapshot is preserved at dist/catalog.json.prev, exit code propagates from build-catalog.js.

Test plan

• node --check scripts/dd-338-refresh-catalog.mjs — syntax clean
• node --test scripts/dd-338-refresh-catalog.test.js — 10/10 pass
• Wrapper end-to-end on current main — baseline first-run + empty-diff second-run both observed
• Existing test suite — 188/188 pass after build-forge-context.js regenerates the ignored scripts/generated/ artefact (pre-existing dependency, not introduced here)
• Architect dry-run against a real D.3 --apply output to confirm the diff matches the proposals (deferred to first live D.3 cycle)

Scope-out

• R2 / Cloudflare Worker publish — owned by stallari-registry-infra deploy pipeline
• Daemon-side cache invalidation — stallari-cli registry refresh documented in runbook §4
• Actually applying tier flips — that's D.3's tier_issuance.py --apply

Closes [[DD-338]] Phase D.5. Final substantive Phase D sub-phase before the ≤10% refuse-envelope gate verification (D.4's territory).

🤖 Generated with Claude Code