Imported from tiangong-ai/agent-skills (bluesky-cascade-fetch/SKILL.md). Install upstream with npx skills add tiangong-ai/agent-skills --skill bluesky-cascade-fetch. Copyright stays with the author.

Bluesky Cascade Fetch

Use the CLI-owned bluesky.public-posts/fetch-cascades operation. This Skill supplies intent routing and result-use boundaries only; the CLI TypeScript 7 runtime owns source discovery, schemas, endpoint policy, pagination, retries, normalization, limits, partial results, and receipts.

Before running

1. Read references/tiangong-data-requirement.json.
2. Use the caller- or workspace-resolved stable CLI. The requirement declares compatible capability and operation contract majors; it does not select a package build.
3. Run data describe with that same CLI. Continue only when the capability ID and required contract majors match, and copy the exact current capability/operation versions from that response into the run request.

tiangong-ai data describe bluesky.public-posts --json

Use the current Discovery Metadata to confirm source coverage, freshness, restrictions, selection hints, provides, and doesNotProvide. Do not rely on provider facts remembered from an older Skill revision.

Choose the seed source

• Use search when the user supplied a topic or query.
• Use author-feed only for a named public actor view.
• Use feed only when the caller supplied a known public feed-generator AT-URI.
• Use list-feed only when the caller supplied a known public list AT-URI.
• Enable thread expansion only when visible reply topology is needed. Prefer seed-only retrieval when discovery breadth matters more than cascades.
• Preserve an explicit user time window and filters. Never widen or replace them silently.
• When diagnosing an empty historical search, repeat the same bounded request with source.applyServerTimeFilter set to false. Keep the client-side time window unchanged and classify any still-empty result as provider coverage, not proof that no discussion occurred.

Prepare the request

Build one tiangong.data.run-request.v1 envelope and validate it against the current input schema from data describe. Replace both version placeholders with the exact versions from the same data describe response.

{
  "schemaVersion": "tiangong.data.run-request.v1",
  "capabilityId": "bluesky.public-posts",
  "capabilityVersion": "<describe.manifest.capabilityVersion>",
  "operationId": "fetch-cascades",
  "operationVersion": "<describe.manifest.operations[0].operationVersion>",
  "input": {
    "source": {
      "mode": "search",
      "query": "climate policy",
      "sort": "latest",
      "language": "en",
      "tags": ["climate"],
      "applyServerTimeFilter": true
    },
    "startDateTime": "2026-03-10T00:00:00Z",
    "endDateTime": "2026-03-11T00:00:00Z",
    "pageSize": 50,
    "expandThreads": true,
    "maxThreads": 20,
    "threadDepth": 8,
    "threadParentHeight": 5
  }
}

Do not put an endpoint override, credential, output path, scheduler, or provider-specific field absent from the schema into the request. Persistence and recurring collection belong to the caller.

Run

tiangong-ai data run bluesky.public-posts fetch-cascades \
  --input /absolute/path/to/request.json --json

Preserve the complete tiangong.data.run-result.v1 envelope, including contract, warnings, errors, summary, and receipt, when handing results to another workflow.

Result boundaries

• Treat search ranking, feed selection, visibility, timestamps, counters, and thread nodes as mutable provider snapshots, not exhaustive or verified facts.
• Surface blocked/not-found nodes, failed threads, empty results, partial, and truncation. Never reinterpret them as complete absence or complete coverage.
• Retain hitsTotal, per-page invalid-record counts, cascade validation, and the fallback-host indicator; they are acquisition diagnostics, not evidence quality scores.
• Public posts can contain personal, sensitive, misleading, or unsafe content. Quote or retain only what the task requires and preserve provenance.
• Reply topology is a visible snapshot. Do not infer causation, influence, representativeness, identity, or sentiment labels from it alone.
• Use a separately governed repository/firehose workflow for exhaustive AT Protocol records and a content workflow for media or linked pages.
• Cross-source comparison, evidence admission, persistence, polling, and research conclusions belong to the caller or Auto Research.

Reference

• references/tiangong-data-requirement.json: stable capability requirement; it is not a package lock.