8000 [pull] main from can1357:main by pull[bot] · Pull Request #114 · ericrihm/oh-my-pi · GitHub
[go: up one dir, main page]

Skip to content

[pull] main from can1357:main - #114

Merged
pull[bot] merged 22 commits into
ericrihm:mainfrom
can1357:main
Sep 23, 2026
Merged

pull[bot] merged 22 commits into
ericrihm:mainfrom
can1357:main

Conversation

@pull
@pull pull Bot commented Sep 23, 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 : )

andrebrait and others added 22 commits September 23, 2026 10:54
… copy

`git commit` spawns `git maintenance run --auto --quiet --detach`, which
writes and removes `.git/objects/maintenance.lock`. beforeEach copies the
fixture repo with fs.cp, and a lock that vanishes between readdir and lstat
fails the copy with ENOENT, aborting the case. Disable auto maintenance and
gc on the fixture, matching the fsmonitor hardening already in this file.
The parity case slept 20ms and read the overlay count once, so any run where
the injection landed later than that delay reported `during: 0` and failed.
highlightElement holds the overlay for `duration` in the host runtime before
removing it, so the case now waits for the node with tab.waitForSelector and
widens the hold to 1s, leaving a failure that names the selector instead of an
opaque count diff. `await pending` still proves the teardown removes it.
…ts write

The tree-kill oracle compared the kill's latency against a backgrounded
descendant's 1.5s write delay: a loaded runner that killed the tree slightly
later still failed a working tree-kill ("orphaned descendant wrote the marker
after the timeout"). The descendant now records its pid and the case asserts
that process is gone, the oracle the reparented and session-escape cases in
this file already use, on the same ESCAPE_TIMEOUT_MS budget they use for a
pid-file wait that runs inside the timed command (#10259). The marker is gone
with the latency comparison it existed for.
…discovery

- Added multi-account discovery for Codex cyber access programs (Daybreak).
- Implemented OAuth account affinity for models requiring specific cyber entitlements.
- Added automatic request replay logic to fallback to standard mode when a specific cyber program is rejected by the backend.
- Updated usage dashboard to display Daybreak-enabled accounts.
…stead of racing its write

The SIGTERM-ignoring case still compared a 150 ms timeout plus 2 s poll
against a marker the descendant wrote after 1.5 s, so it measured runner
speed rather than the kill. The descendant now records its pid, keeps
ignoring TERM through exec, sleeps far past the budget, and the case
asserts it is dead via the same pid oracle and ESCAPE_TIMEOUT_MS budget
as the sibling escape cases.
…ure copies

The commitToBranch fixture was not the only one that commits and then
copies its .git with fs.cp/cpSync. The applyNestedPatches parent and
nested repos, the autoresearch template repo, and the gh PR fixture
template can hit the same ENOENT when git maintenance run --auto removes
its lock mid-copy. Disable maintenance.auto and gc.auto before their
first commit, as the fixed site does.
fix(coding-agent): strip inline data URI images from reader-mode fetch output
…irst messages

- Added support for treating vague or greeting messages as no title in system prompts.
- Introduced #deferredTitle state to track declined title generation and assistant reply.
- Implemented #startAutoTitle method to run a single auto title generation and handle decline.
- Added #advanceDeferredTitle to retry title generation after an assistant reply when the model initially declined.
- Provided isTitleContextReply helper to detect assistant replies eligible for retitling.
- Added test confirming automatic title retry behavior when first message is ambiguous.
…C cleanup of recaps

- GC command now also deletes rows from `session_recaps` when removing a session.
- Idle recap turns are now recorded to `session_recaps` table via SessionManager.recordRecap.
- Session index schema expanded with `session_recaps` table and supporting index.
- SessionManager gained recordRecap method that journals recaps for persisted sessions.
- GC deletion now safely handles missing history tables by early return.
- Introduced an in-process Swift bridge for the Apple Foundation Models API (macOS 27+), enabling on-device text and reasoning model execution.
- Added support for tool calling, vision content, and reasoning effort within the on-device model provider.
- Configured automated discovery for on-device Apple models on eligible Apple silicon hardware.
- Added strict JSON schema lowering for tool definition compatibility with the Apple Foundation Models provider dialect.
- Implemented a build-time toolchain probe in Bazel and Cargo to handle conditional Swift compilation and weak-linking of the FoundationModels framework.
- Added bridge availability check for device eligibility, OS version, and Apple Intelligence status.
…encodings

- Registered `toks` command in cli-commands.ts and added corresponding help entry.
- Implemented command to resolve a file path or literal text, then compute byte, character, and line statistics.
- Added JSON output mode and formatted table showing token counts for each offline encoding with ratios and comparison to the O200k base encoding.
- Renamed edit headers: `*** Edit File:`, `*** Find`, and `*** Replace` supersede the former `SM:` prefixed headers.
- Added `*** Insert Before` and `*** Insert After` for adding lines without modifying or replacing existing code.
- Updated internal grammars, prompts, and parsers to enforce new header schema and support granular insertion operations.
…usage estimation

- Added import of renderToolExamples for rendering examples.
- Extended ContextTool type to include examples property.
- Appended rendered examples block to token fragments during estimation.
- Added TypeSafe Jev 1.13 tokenizer to offline token counting via measured whole-word vocabulary and base-subset BPE reconstruction.
- Excluded request frame from Jev token counts to maintain consistency with existing families.
- Implemented count-only Jev encoding path, mirroring Claude's non-invertible reconstruction.
- Added Jev to `toks` CLI command and `natives` package bindings.
…he commit fixture copy (@andrebrait)

test(task): stop git background maintenance racing the commit fixture copy
…d of sampling once (@andrebrait)

test(browser): wait for the highlight overlay instead of sampling once
…stead of racing its write (@andrebrait)

test(config-value): assert the descendant is dead instead of racing its write
- Replaced the monolithic `hub` tool with a dedicated `wait` tool for blocking operations and specific protocol handlers (`proc://` for job/service control, `agent://` for peer messaging).
- Implemented `ProcProtocolHandler` and integrated `write` and `read` tool extensions to support fine-grained coordination of background processes, services, and agent-to-agent communication.
- Refactored internal tooling and UI renderers across `coding-agent`, `collab-web`, and `tui` packages to utilize these specialized protocols, removing the legacy `hub` implementation and its associated test suites.
- Added a skill description catalog system to manage and compress model-aware tool documentation, improving system prompt efficiency and on-demand tool capability discovery.
- Updated grammar to treat a trailing End Patch as a wrapper closer rather than literal insertion text.
- Adjusted parser to stop processing at a closing sentinel followed only by blank or fence lines.
- Added tests verifying that the trailing End Patch closes a wrapped insert and preserves existing content.
@pull pull Bot locked and limited conversation to collaborators Sep 23, 2026
@pull pull Bot added the ⤵️ pull label Sep 23, 2026
@pull
pull Bot merged commit 3d3ec7e into ericrihm:main Sep 23, 2026
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.

3 participants

0