Prepare for tinycloud 0.3.22: bulk MEDIA imports, --enrich-metadata, Dropbox --recursive (skill floor → 0.3.22)

[kdr]

tinycloud 0.3.22 (Cloudglue SDK 0.7.28, spec v0.7.21) turns bulk imports from a metadata-collection loader into a general one. Features 46→49; verbs stay 17.

What the skill now teaches

• import_type decides the cost. An import's kind follows the target collection's type, inferred at create and fixed for its life. A metadata collection still imports source_metadata only (free). Any other type (media-descriptions, entities, rich-transcripts, face-analysis) now imports the media itself, processed like collections add — billed per file, counted against the account's file usage limits, capped at 10000 files/run, resumable with an append rerun after a credit stop. Feature library.collections.imports.media.v1.
  The skill gets this as a cost table plus an explicit instruction to ask the user before starting a media import and to cap it with --max-files — the summary and data.import.import_type are how an agent tells which kind it just created.
• --enrich-metadata / --no-enrich-metadata (feature library.collections.imports.enrich.v1) — backfills fields the connector listing omits: Gong parties + Call Spotlight (re-embedded, so the content becomes searchable) and Dropbox media_info duration/dimensions. Metadata-import only; it and --thumbnails are a clean upstream 400 on a media import.
• --recursive (feature library.connectors.recursive.v1) on connectors files and as a recursive filter-set key on imports create — walks a whole Dropbox subtree under --path instead of its direct children, so a tree is one recursive pass rather than one pass per folder. On the wire it's the string "true"/"false".

Floor raise

tinycloud-skill.json min_version / supported_range / skill_version → 0.3.22, required_features +library.collections.imports.media.v1 +library.collections.imports.enrich.v1, with preflight.sh MIN_VERSION and REQUIRED_FEATURES kept byte-identical (CI diffs them).

library.connectors.recursive.v1 is deliberately not in required_features — connector-capability ids are excluded there by convention, the same way library.connectors.iconik.v1 and .refresh.v1 were handled at the 0.3.15 floor raise.

Also in here

.claude-plugin/plugin.json and marketplace.json → 0.3.22. These had drifted: both were still on 0.3.19 because the 0.3.20 and 0.3.21 releases skipped the bump, and no CI job checks them. Worth deciding whether to add a check — happy to do it in a follow-up rather than widen this PR.

Verification

Gate	Result
npm test	42 pass / 0 fail
TINYCLOUD_TEST_TARBALL=… npm test (real 0.3.22 tarball)	42 pass / 0 fail
shellcheck install.sh scripts/smoke-test.sh …/preflight.sh	clean
EXPECTED_VERSION=0.3.22 bash scripts/smoke-test.sh	all smoke tests passed
CI plugin/skill metadata — all 3 checks run locally	JSON valid · preflight↔manifest parity (37 features in sync, >=0.3.22 <0.4.0) · frontmatter
preflight.sh vs the 0.3.22 binary	exit 13 — compatible, credentials missing
All 37 required features present on 0.3.22	missing: none

The binary-side behavior these docs describe was verified live against an account with 9 active connectors: a media import ingested a real Dropbox file (files_imported: 1), and --enrich-metadata backfilled media_info on 5 records that the raw connector listing returns as null.

[cursor]

PR Summary

Low Risk
Docs, skill metadata, and version pins only—no launcher or binary logic. Residual risk is agents following the new billing guidance incorrectly, not a code-path change.

Overview
Raises the skill/package floor to 0.3.22 so agents teach tinycloud’s new bulk-import cost model (SDK 0.7.28). package.json, plugin/marketplace metadata, tinycloud-skill.json, and preflight.sh now require >=0.3.22 plus library.collections.imports.media.v1 and library.collections.imports.enrich.v1.

Skill docs (SKILL.md, verbs.md, CLAUDE.md) stop treating library imports as metadata-only. import_type follows the target collection: metadata stays free source-metadata indexing; any other type ingests media, billed per file, with a 10k-file cap, resume-via-append, and an explicit “ask the user first” instruction.

Also documents --enrich-metadata (Gong/Dropbox backfill, metadata-import only) and Dropbox --recursive on connectors files / import filters (string "true"/"false"). Plugin JSON jumps from 0.3.19 to 0.3.22 to catch skipped bumps.

^{Reviewed by Cursor Bugbot for commit adca09f. Bugbot is set up for automated code reviews on this repo. Configure here.}