Skip to main content

Share types

A share type is the kind of artifact. It decides three things: the viewer that renders the body, the metadata panel on the right, and the annotation anchors you can react to or comment on. The set is extensible — traces were added as a new type without disturbing the others (see ADR-0002).

MD Markdown​

Rendered markdown with a table of contents. React under blocks and to the left of bullets; select any text to drop a comment in the right margin. Example: checkout-web-audit.md.

PY Code​

Syntax-highlighted source with line numbers and a symbol outline. Comment a single line or a selection. Example: parse_ledger.py.

IMG Image​

The image with pins: drop a pin to anchor a comment to a region; react below. Example: Movies.png — a Jellyfin library banner mock.

FILE File​

Generic, non-previewable blobs. Size, gzip note, checksum, and download — react and discuss even when there's nothing to render. Example: staging-db-dump.sql.gz, 42.7 MB.

Bundle​

Many files in one share. Humans browse a tabbed viewer; agents read the same files over MCP. Mixed media welcome — created in one shot:

cairn add audit.md parse_ledger.py Movies.png dump.sql.gz
# ✓ bundle → cairn.stump.wtf/9qz1a   ·   4 files · 43.0 MB · ⧗ expires 7d

HK Webhook​

A live requestbin. Point an endpoint (an HTTP ingress URL plus mcp://cairn/hook/<id>) at anything; requests stream into an inspector with highlighted JSON and a status mix, and to agents over MCP as the same stream. Requests are reactable but not comment-threaded — discussion moves to the artifacts they produce. See ADR-0010.

TRC Trace​

A whole agent run, shared. The flagship type — a trace is the captured, shareable record of an agent run.

A trace is the artifact; a run is the execution it records. That is why the URL and the tools keep the run noun — cairn.stump.wtf/run/<id>, /v1/runs, run_create/run_append_spans — while everything a reader sees says trace. The stored share_type is still trajectory and stays accepted on the wire indefinitely; see ADR-0016.

  • A call-trace waterfall is pinned up top: OTel-style spans nested by depth, over a time ruler, colored by category — reason exec read net write. …and more. Agents label spans either by operation kind (reason · exec · read · net · write · search · plan · tool · analyze · test · fix · fail · meta) or by workflow phase (research · implementation · review · testing · debug · build · docs · delivery · deploy · wait) — whichever fits how they think. The set is open: a category outside both is accepted and gets its own color, derived from its name. A sub-agent is a nested span group. Click any span to jump to and expand its event.
  • A readable activity stream runs below: the human prompt, each reasoning turn, each tool call (expandable args + output), and the write that produces an artifact — linked back as its own share, so a run and its outputs stay connected.
  • The right panel carries provenance (claude · sonnet-4.6 · via MCP · captured 2h ago · expires in 7d), run stats (wall time, spans, tool calls, tokens), and a time-by-category breakdown.
  • React on any turn or tool call; select text or use the ＋ picker to comment.

See ADR-0009 and SPEC-0004.

Not in v1

Token-cost lane on the waterfall, run-vs-run diff, and errored-run rendering are noted as "try next" — deliberately out of the first cut.