[forrestjgq]

Summary

Space on an attempt row in the raven trajectory browser previously
showed two truncated per-turn preview lines. This PR replaces it with an
interactive full-conversation viewer. It re-lands the content of PR #382,
which was auto-closed by the 2026-09-12 history linearization without
being merged; the branch is rebased onto the linearized main, with one
extra commit expressing the CJK display-width test fixtures as unicode
escapes to satisfy the source-language gate that main gained since.

• Data layer raven/trajectory/conversation.py: rebuilds an attempt's
  span snapshot into a causally ordered stream of labeled records (User
  input, LLM input/output, Tool input/output, skill, memory, and
  subagent events). Event-time semantics (inputs at span start, outputs
  at span end) plus nesting depth from parentSpanId keep the causal
  order. Full content comes from artifact files (paths restricted to
  the trace store, streamed 512 KiB cap per file); span previews are
  only a visible fallback. ERROR spans and expected-but-unreadable
  payloads always yield a placeholder record. LLM inputs omit only the
  item-by-item verified common message prefix against the previous call
  (omissions and rewrites are announced); tool_calls of unexecuted
  tools keep their full arguments.
• Rendering: kind-colored labels aligned to one column, hanging-indent
  bodies with word-aware CJK-safe wrapping, degraded/error/meta note
  lines, single-pass turn separators (interleaved traces are never
  regrouped), and a stacked layout on narrow terminals that never drops
  characters.
• Interactive viewer (prompt_toolkit full-screen, entered for every
  preview on a terminal): scrolling, a muted bottom bar with an
  'h for help' hint, filter/collapsed state and the scroll range (laid
  out by display width, never overflowing any terminal), a scrollable
  help page reachable in full on small screens, a global collapse (s:
  bodies over five display lines fold to five plus an ellipsis; error
  and degradation notes never fold), and a label filter (%: real input
  line, case-insensitive, empty clears; filtered-out turn groups lose
  their separators). Resizes re-wrap and re-judge folding live; the
  content caret is hidden (only the filter input shows one). q/Esc
  returns to the list; Ctrl+C cancels the whole browser from any viewer
  state. A rebuild failure falls back to the legacy per-turn preview;
  without a terminal the preview degrades to direct printing.
• Every dynamic field passes the sanitization gates (control
  characters, rich markup, and ANSI never reach the terminal). Adds the
  Conversation Record term to CONTEXT.md.

Type

• Fix
• Feature
• Docs
• CI / tooling
• Refactor
• Other

Verification

• uv run pytest tests/test_trajectory_conversation.py tests/test_cli_preview_viewer.py tests/test_cli_trajectory_browse.py tests/test_trajectory_store.py -q -> 298 passed (viewer assertions
  run against really rendered screens)

• uv run pytest -q (full, on the rebased branch) -> 23463 passed,
  125 skipped; 33 failed + 20 errors reproduce with identical node ids
  on a clean origin/main worktree - environment baseline, not
  introduced here (passed-count delta of 96 equals the tests this
  branch adds)

• make check-source-language, make check-large-files,
  make lint-types, make lint-python, make lint-imports -> all pass

• Manual: rendered 6 real local attempts at widths 72/120/30 with zero
  over-width lines (largest attempt 7060 lines in 0.35s); drove the
  real TUI in a pty at normal and 44x12 sizes (viewer, help paging,
  collapse, label filtering, Ctrl+C cancel)

• Relevant tests pass locally

• Relevant lint / type checks pass locally

• User-facing docs or screenshots are updated when needed

Risk

• Read/display side only; no persisted format changes. The untrusted
  input surface (artifact content, typed filter input) is contained by
  the store-prefix path check, the 512 KiB streamed cap, and per-field
  sanitization.

• User-visible change: the Space preview is now a full-screen viewer;
  the legacy rendering remains as the automatic fallback when the
  rebuild fails, and non-terminal environments print directly.

• Rollback: revert this PR, or restore the previous _preview_screen
  call site to get the old preview back.

• Security impact considered

• Backward compatibility considered

• Rollback path is clear for risky changes

Related Issues

N/A

[gloryfromca]

Blocking: current LLM input artifacts must be resolved before the full preview can merge.

I found one blocker in the live trace compatibility path. I covered the full diff, AGENTS.md and the canonical trajectory terms in CONTEXT-MAP.md/CONTEXT.md, the preview callers, the current trace writer and artifact resolver, relevant history, backward compatibility, and whether the tests weakened or bypassed behavior; I found no separate architecture issue.

Verification: uv run pytest tests/test_trajectory_conversation.py tests/test_cli_preview_viewer.py tests/test_cli_trajectory_browse.py -q passed (221 tests); uv run pytest tests/test_tracing_api.py tests/test_trajectory_bundle.py -q passed (89 tests); the source-language and large-file check scripts passed. A direct reproduction using TraceStore.address_items and audit.artifact.v2 showed contains full message: False, with the record displaying a $msg hash instead.

[gloryfromca]

Blocking: v2 message-blob reads must honor the preview size bound.

The original $msg rendering defect is fixed and its thread is resolved. I found one new blocker in the resolver path. I rechecked the full PR diff and the 80fe873 delta, AGENTS.md, CONTEXT-MAP.md/CONTEXT.md terminology, the preview caller, the live v2 writer/resolver and image-bearing request path, relevant history, backward compatibility, and whether tests were weakened; no other actionable issue survived refutation.

Verification: uv run pytest tests/test_trajectory_conversation.py tests/test_cli_preview_viewer.py tests/test_cli_trajectory_browse.py tests/test_tracing_api.py tests/test_trajectory_bundle.py -q passed (313 tests). The source-language and large-file scripts passed. A real TraceStore reproduction with one 700 KiB image-bearing message produced a 716,900-byte blob, a 716,878-character preview record, degraded=None, and the complete payload in memory despite _ARTIFACT_LIMIT being 524,288 bytes.

[gloryfromca]

No blockers; this can merge as far as I am concerned.

The bounded v2 blob fix resolves the remaining blocker, and both threads I opened are now resolved. I rechecked the full PR and the f49c291 delta, AGENTS.md and trajectory terminology, the preview caller, current v1/v2 writers and resolver behavior, image-bearing and damaged-store compatibility, relevant history, architecture boundaries, and whether tests were weakened; I found no new actionable issue.

Verification: uv run pytest tests/test_trajectory_conversation.py tests/test_cli_preview_viewer.py tests/test_cli_trajectory_browse.py tests/test_tracing_api.py tests/test_trajectory_bundle.py -q passed (316 tests). The source-language and large-file scripts passed. The original 700 KiB reproduction now performs one bounded 524,289-byte read, emits an 87-character placeholder with the degradation note, and does not materialize the payload.

[0xKT]

Not a blocker -- two things, neither of which changes a rendered pixel today.

1. _resolve_v2 is a second decider of the v2 per-field rule, and it already disagrees with the
owner (raven/trajectory/conversation.py:415)

raven/tracing/artifact_v2.py:6-7 says of that module: "The canonical serialization, the envelope
key, the reference validation and the per-field resolution rules all live here and nowhere else" --
and gives the reason, that a JavaScript mirror exists and a cross-language round-trip test compares
the two.

_resolve_v2 delegates the parts its own docstring names -- is_v2, ref_sha1 -- but re-decides
the rest: it strips artifactFormat itself and resolves messages itself, and it does not resolve
TEXT_FIELDS (systemPrompt, prompt), which the owner does at artifact_v2.py:145-148.

Measured on one shell, both resolvers pointed at the same blobs:

OWNER artifact_v2.resolve_payload   systemPrompt = 'You are Raven.'
NEW   conversation._resolve_v2      systemPrompt = {'$msg': 'e7b04557...'}
messages equal                      True      <- control: both read the same blobs
whole payload equal                 False
fields that differ                  ['prompt', 'systemPrompt']

The messages line is the control: it is identical, so the divergence is the TEXT_FIELDS policy
and not my harness. (My first attempt put the blobs under the owner's default layout rather than
state/logs/audit-artifacts, which made every message read as missing -- worth saying because that
version would have looked like a much larger defect and was my setup, not your code.)

Nothing renders differently today: this module never reads systemPrompt or prompt -- grep over
conversation.py finds neither outside the resolver -- so I am filing this as what it is, a second
live decider of one rule rather than a visible bug.

The reason for the fork is legitimate and worth keeping: resolve_payload's inner load() reads
the blob with an unbounded read_text (artifact_v2.py:133), which is exactly the bound this PR set
out to impose. That argues for giving the owner a bounded-read hook -- a read= callable, say --
and calling it from here, rather than keeping a partial copy that starts out already differing on
two fields. As it stands the tree has three implementations of one wire rule (this one, the owner,
and server.js:1088) and a round-trip test that compares only two of them.

Smaller, same family: _oversize_placeholder (conversation.py:386) is a second placeholder shape
beside artifact_v2.placeholder, which this module uses correctly two lines away at :405.

2. Every preview open renders the whole conversation twice (raven/cli/trajectory_browse.py:535)

if not _conversation_lines(records, width):      # rendered, tested for emptiness, discarded
    ...
if _viewer_available():
    def _make_lines(...): return _conversation_lines(records, view_width, ...)
    if pviewer.view_lines(_make_lines, ...)      # rendered again

The first call exists to answer "is there anything to show", and pays for a full render to do it.
For a long attempt that is the whole cost twice on every open. Either keep the first result and
hand it to the viewer as the initial width's lines, or answer the emptiness question from
records.

For the record, what I checked hardest and found sound: the 512 KiB bound is applied where the
docstring says it is -- _load_message_blob routes every blob through _read_artifact
(conversation.py:402-403) rather than reading it directly, so a content-addressed inline image gets
the same bound as any other artifact; and the four test files this PR touches, plus the owner
module's own suite, are 248 passed on this head.