Repository navigation
fix(neopi): render usage reports without literal Markdown fences #174
Description
Note
GPT-6 Astra on behalf of PsychedelicShayna
Observed failure
The native conversation displays literal opening and closing triple-backtick Markdown fence characters around /usage output. The provider/model lists, quota text and colored bars are visible; no literal ANSI escape sequences were observed across the complete overlapping output coverage. Fence rendering is a separate formatting defect, not an ANSI regression or a persistence claim.
Evidence
Observed in draft #167's private clean build 426bde5bcfead481715c85a0dadcdd37a0b9c198, with native dcc5fed8eb unchanged, on 2026-10-29 around 17:02 UTC. Terra HIGH submitted /usage once and captured its full output in twelve overlapping 800×1000 frames. The parent personally inspected all twelve, including the opening fence before Usage and the closing fence after the final quota reset. The images contain private account labels and are retained privately, not uploaded.
This does not concern the unrelated passive-scroll finding #168, settings copy correction #171 or whether native commands are journal-persistent. No additional command matrix or reproduction loop was run.
Expected behavior
Render the native command report's meaningful formatting without exposing its enclosing Markdown fence syntax, while retaining ANSI colors, readable quota bars and complete output. Do not suppress or truncate content as a workaround.
Activity
fix(neopi): render native report fences as formatting
PsychedelicShayna commentedon Oct 29, 2026
Note
GPT-6 Astra on behalf of PsychedelicShayna
Repair verified privately on draft #167
Commit
7ebfe63623c0f372fa569d5a62203e3e679496aeis signed, verified and pushed to #167's existing topic branch. It removes only the complete bare outer fence envelope emitted by the pinned native report, then uses the unchanged ANSI parser. Partial, inline and mixed-document fences remain literal; raw command text, persistence and reconnect behavior are unchanged.The focused regression failed before the repair (2 failed, 12 passed), then all 14
neopiCommandOutput.test.tscases passed. Targeted formatting, lint and client-runtime typechecking passed. The private desktop artifact built and launched with the unchanged native pindcc5fed8eb9a9b4dcac3a15eec106c27b540720c. Those checks were not rerun for the graphical continuation.After correcting our private launcher's explicit disabled-instance seed through Settings → NeoCore, Terra HIGH created a fresh project-free thread and submitted
/usageonce. Main personally inspected all 16 retained frames, including the ten overlapping report frames from its first line to its final line. Both outer fences are absent, no literal ANSI escapes are visible, and indentation plus meaningful colored quota bars are retained. The one-submission count comes from Terra's action record; still images alone cannot count clicks.Independent Claude Opus 5.5 HIGH personally inspected the same 16 frames and their complete overlap chain: #174-only GUI-evidence PASS, no proven #174 defect. This is an incremental graphical addendum alongside the earlier SOURCE-ONLY PASS; the source-only report remains unchanged. Private account information in the screenshots is not published.
The three earlier attempts remain unverified for this criterion: cancelled fixture-ID typing, click-only selection without a fixture, and fresh-thread creation blocked by the deliberately disabled sandbox instance. No cancelled action was retried in the successful continuation. There is no claim about mobile, reconnect, journal persistence, streaming/scroll-follow,
/usage resetor other fenced emitters.The repair remains in draft #167; this is not a merge or live-install announcement. This issue stays open pending integration and promotion. CORE-1 remains separately tracked and unfixed in #168, despite acceptance of the native core with that issue open.
GPT-6 Astra via NeoPi; graphical verifier GPT-6.1 Terra HIGH; independent source and graphical-evidence reviewer Claude Opus 5.5 HIGH.