8000 [pull] main from can1357:main by pull[bot] · Pull Request #109 · ericrihm/oh-my-pi · GitHub
[go: up one dir, main page]

Skip to content

[pull] main from can1357:main - #109

Merged
pull[bot] merged 185 commits into
ericrihm:mainfrom
can1357:main
Sep 22, 2026
Merged

pull[bot] merged 185 commits into
ericrihm:mainfrom
can1357:main

Conversation

@pull
@pull pull Bot commented Sep 22, 2026 •
Copy link
Copy Markdown

See Commits and Changes for more details.


Created by pull[bot] (v2.0.0-alpha.4)

Can you help keep this open source service alive? 💖 Please sponsor : )

roboomp and others added 30 commits September 3, 2026 02:05
Captured thinking_signature and thought_signature from streamed assistant deltas and replayed them on the originating assistant message.

Fixes #10630
Stored per-call extra_content/reasoning signatures and message-level thought signatures in one envelope so a later message signature no longer clobbers the per-call one on replay.

Fixes #10630
…lowup

# Conflicts:
#	packages/coding-agent/src/session/session-tools.ts
…sor snapshot

- session-tools: derive candidate visibility per preparation, render and
  compute the signature inside its AsyncLocalStorage frame, and publish
  prompt + snapshot together at commit; a discarded, stale or throwing
  preparation leaves the committed prompt and hint untouched
- sdk: advisor ToolSession delegates skillHintVisible to the primary
  session instead of capturing the startup value in the spread
- tests: abandoned-refresh rollback (discriminating), scoped candidate
  render, throwing descriptor preservation, advisor present/absent/
  present transitions, standalone fallback + explicit snapshot precedence
- changelog: Unreleased entry for the drift fix
… a function"

Workspace loads tolerate a version-sentinel mismatch on purpose, so a checkout
that pulled a new release kept loading its previous addon. Every symbol added
since that build was `undefined`, and the failure surfaced at the first call
inside whichever tool used it: after the write read-projection guard landed,
each `write` call died with `hashlineIsReadTruncationNotice is not a function`,
naming neither the addon nor the fix.

Without this fix, omp happily runs with stale native module.

Function exports now go through `missingNativeExport`, which is `undefined` on a
current addon — capability probes such as
`typeof native.macOSSpellCheckerAvailable === "function"` keep their meaning —
and a throwing stub on a stale one, naming the symbol, the addon path, the
loaded and expected releases, and `bun run build:native`. The release sentinel
stays a raw passthrough so the loader's own presence check still sees the truth.

Assisted-by: omp (deepseek/deepseek-flash)
Assisted-by: omp (deepseek/deepseek-flash)
can1357 and others added 27 commits September 22, 2026 14:54
…ted thinking ladder (@eggpeat)

# Conflicts:
#	packages/catalog/src/compat/rules.json
…nd true color (@danilouchoa)

# Conflicts:
#	packages/tui/src/kitty-graphics.ts
#	packages/tui/src/terminal-capabilities.ts
Promote entries the union merge filed under released versions back to [Unreleased], drop fork-side stray lines, dedupe headings, and add entries for merged fixes that lacked one.
…json

The #12380 conflict resolution reset models.json to main, dropping the single field #12771 regenerated.
- Deleted the bunfig.toml test preload configuration.
- Removed the preload-agent-dir-isolation.ts test helper.
…ge query options

- Removed support for sending bare image paths directly to models.
- Dropped the `?q=` image query option that allowed vision model answers as text.
- Updated SVG handling to require explicit `:img` flag for image rendering.
- Clarified that images are now always decoded inline without special handling.
History recall picked the caret anchor from the direction of travel: Up
opened an entry at line 0 / column 0, Down at the last line's end. For an
entry that fits on a single visual row both edges are the same row, so the
rule degenerated into a bare column flip — the same prompt showed the caret
at the start when Up recalled it and at the end when Down reached it, while
the first edit relocated the caret to the end anyway, leaving delete/yank
chords aimed at column 0 on one path and at the tail on the other.

Single-row entries now always open at the end. Multi-row entries keep the
directional anchors (#99) because those are what make one keypress step one
history entry instead of walking inside the recalled prompt; a newline-free
line that wraps past the editor width counts as multi-row for the same
reason. An empty recall is unchanged.

Verified in the running TUI: recalled one-row entries now land at
(0, length) for both arrows, a seeded multi-line entry still opens at
(0, 0) on Up and at its last row's end on Down with one keypress per entry.
fix(tui): open recalled prompt history at a consistent caret
@pull pull Bot locked and limited conversation to collaborators Sep 22, 2026
@pull pull Bot added the ⤵️ pull label Sep 22, 2026
@pull
pull Bot merged commit 0f9139f into ericrihm:main Sep 22, 2026
5 checks passed
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

0