Repository navigation
[pull] main from can1357:main - #117
Merged
Merged
Conversation
… canonical long-form spelling brush-core stores working_dir in long form on Windows: `set_working_dir` and `Shell::new` expand 8.3 short-name components via GetLongPathNameW (identity-only, no symlink resolution, no \\?\ prefix), so cd and the initial cwd no longer flip between ADMINI~1 and Administrator. pi-shell compares requested cwd by filesystem identity to avoid redundant churn when the host hands us a short-spelled path. For vault://, the sync bash path used fs.realpathSync (keeps 8.3 spelling) while the async read path used fs.promises.realpath (long form), so one vault file had two string identities. The sync path now uses the long-form native realpath so 8000 both resolve consistently. Fixes #7911
…etLongPathNameW - expand_to_long_path_impl: drop the to_str() detour (Windows OsStr is UTF-16 and may not round-trip through UTF-8, silently skipping expansion) and fix the sizing/fill TOCTOU by retrying when the long form grows between the two GetLongPathNameW calls. - shell_working_dir_matches: compare via expand_to_long_path (same short-name-only identity that working_dir is stored with) instead of Path::canonicalize, which resolves symlinks/junctions and made the comparison looser than the stored identity. - Regression tests feed a real 8.3 short spelling computed with GetShortPathNameW instead of a long-form passthrough, and add a test that a short-spelled requested cwd matches the stored long-form working dir.
- Registered numbered rows rendered by hashline edit results with their stored snapshot tags. - Covered hidden-line rejection and identical-content revert promotion. Fixes #13144
…n classifier refusals
…he named agent
- entries are a token count (90000) or percentage ("80%"); null clears a lower layer, removing the nested-object merge trap
- agents without an entry, including descendants of an overridden agent, use the root compaction thresholds on spawn and revival
- host-default task.agentCompactionThresholdOverrides for RPC/ACP embedders
- Made fd and find display POSIX separators across platforms. - Matched the builtin registry expectation to errno’s Unix gate. - Centralized forward-slash path formatting in pi-walker. Fixes #13164
…in docs and changelog
An automatic disable (definitive refresh failure, upstream token invalidation, broker disable) emitted credential_disabled with only the provider and cause, logged nothing, and showed nothing in the built-in UI, so a session silently moved to a sibling account. Carry the row id and account identity on the event, log every disable once at the pool, warn the live session for providers with a /login entry, and write that warning to stderr in print text mode.
normalize_path now leaves \\?\ and \\.\ paths unchanged, since Windows only honors those prefixes with backslashes; find only allocates a display override when the spelling actually changes.
Edit responses now also keep prior-read lines above the first change anchorable under the minted tag, skip response rows when the written file drifted from the preview, and only register provenance for updates.
- Reverted the unrelated PR-CI skip of the natives task-backed pre-abort test. - Scoped GetShortPathNameW to brush-core tests; dropped redundant cfg gates around expand_to_long_path, which is already a no-op off Windows. - Expanded only the requested cwd in shell_working_dir_matches; the stored working_dir is already long-form. - Rust tests now create a long-named directory with a real 8.3 alias instead of relying on TEMP, and cover cd as well as shell construction; they fail with expansion stubbed out. - Simplified the vault sync/async assertion and rewrote changelog entries with attribution.
…tin-paths fix(shell): normalize Windows builtin paths
…tency fix(windows): expand 8.3 short-name working dir and vault path to one canonical long-form spelling
…-threshold feat(coding-agent): configure compaction thresholds per task/eval agent
…-inputs docs(coding-agent): document which prompt inputs reach each session type
- Preserved pending completions when a session switches away, matching the dispose path, so /resume back to the owning session replays exits that happened meanwhile. - Replaced the drop-on-switch test with a resume replay assertion that fails immediately instead of timing out. - Moved the owner-string check after the routing assertions so a regression names the wrong recipient. - Added the changelog entry for #13199.
- Separator normalization now applies only on Windows; on Unix the display path keeps raw bytes, which -exec {} receives and non-UTF-8 names previously lost to U+FFFD.
- Added a regression test for a non-UTF-8 name through set_display_root.
- nextest runs each test in its own process, so the in-process counter is always 0, and macOS SystemTime ticks in microseconds; two tests starting in the same microsecond shared a directory and one's Drop deleted it under the other (seen as ENOENT in collect_entries_respects_pre_cancelled_token). - Added the process id to the directory name.
…oughput analysis - Added `--detailed` bench flag to run separate single-user, parallel, and prefill phases. - Parallel phase uses `--par` concurrency to measure aggregate throughput and scaling efficiency. - Prefill workload automatically caps synthetic input size relative to the model's context window.
…gs management - Added `cfg://` protocol for reading, overriding, and persisting agent settings. - Enabled agent to propose settings changes via `write cfg://<setting>`, requiring user approval in the host UI. - Supported persisted config updates via `write cfg://<setting>/save` for settings overrides. - Added `omp` configuration capability allowing the agent to manage its own behavior settings.
- Settings.get() substituted the schema default only for `undefined`, so a YAML key written without a value (`retry:` / ` fallbackChains:`) resolved to `null` and crashed resolveRoleChain during tool construction at session startup. - Added a regression test loading a config.yml whose record and enum keys carry no value. Fixes #13183
- Introduced a declarative configuration registry with typed handles, supporting memoized derivation, coalesced updates, and live effect binding for runtime settings synchronization. - Implemented a spec-driven `InternalUrlRouter` to unify URL protocol handling, resource localization, and write-tier approval across all tool domains. - Replaced scattered environment variable and string-based config lookups with strongly-typed registry handles, enabling cleaner session-scoped configuration overrides. - Added support for `cfg://` URLs, allowing tools and agents to inspect and modify configuration state with user approval. - Added `attachment://` and `conflict://` internal URL protocols to streamline session image resolution and interactive conflict resolution.
- Emitted one prompt_result per accepted prompt at the agent's yield, correlated by id, with status, structured provider error (local dump paths stripped), and sessionSettled. - Added session_settled frame, agent_end.yielded, and get_state isSettled/hasPendingAsyncWork to separate agent yield from session quiescence after background work. - Added --no-ui for --mode rpc so extensions run headless without extension_ui_request dialogs. - Added open_session (resume-or-create in a directory), set_event_filter, and messageId on message frames. - Updated the TypeScript and Python RPC clients, docs, and changelog.
- Sent http-referer https://omp.sh/ and x-title omp on every Vercel AI Gateway route (Anthropic and OpenAI-compatible) via applyInferenceHeaders; caller or config headers still win. - Added APP_URL to pi-utils and reused it with APP_NAME for the OpenRouter attribution headers.
- Added `read` views for ELF/PE/Mach-O executables and .i64/.idb files: overview, pseudocode, asm, imports, exports, strings, xrefs, with line ranges; the read surface never names the backend. - Added the discoverable `ida` tool for lifecycle, rename/comment/set_type/make_function, and persistent per-DB Python exec with pseudocode/xref/caller helpers. - Ran one long-lived Python worker per DB in a process-wide registry shared by all agents, with content-addressed store IDBs, file locking, SIGINT-based timeouts, and save on close/exit. - Added ida.enabled, ida.python, and ida.installDir registry settings; IDA features are exposed only when a local install shipping idalib is found.
- Added support for universal (fat) Mach-O binaries in IDA tool, allowing selection of specific architecture slices via the `:@<arch>` syntax. - Implemented automatic slice detection for universal binaries, defaulting to the host CPU architecture and enabling manual override. - Materializes specific Mach-O slices into temporary store IDBs, ensuring IDA analyzes only the selected thin binary architecture. - Updated IDA and read tool documentation to include universal binary slice syntax and host-based selection mechanics. - Fixed a panic in provenance tracking when editing the first line of a file, ensuring empty edit ranges do not attempt to access invalid provenance history.
Each transport closes its properties with additionalProperties: false, which does not see serverBase through allOf, so enabled, timeout, requestIdFormat, auth, and oauth were rejected. Allowlist every shared field per transport and document alias precedence and /mcp reload for the instructions opt-out.
# Conflicts: # packages/coding-agent/src/sdk.ts
- Added `threads_by_pid` to the `proc_snapshot` module to extract thread-level metadata on Linux, macOS, and Windows. - Implemented `ps -M` flag to display individual threads, following BSD-style formatting for scheduler state, priority, and CPU time. - Standardized thread CPU time output to the `M:SS.cc` format used by system utilities. - Added cross-platform thread-info record `ThreadInfo` to abstract kernel-specific thread reporting structures. - Verified multi-threaded execution output in `pi-shell` integration tests.
- Migrated settings registry to a high-performance slot-based memoization system, replacing string-based map lookups with array-indexed property access. - Refactored change notifications to use per-setting listener buckets, preventing redundant recomputations and improving bulk-reload performance. - Optimized `Derived` value caching to update entries in-place, reducing object allocation and churn on recomputation. - Streamlined environment variable parsing by caching raw inputs, ensuring re-parsing occurs only upon actual variable value changes.
- Implemented `gated` helper that wraps a command with a shell script waiting for a gate file before exec. - Updated tool initialization to create a gate file and invoke Bun.spawn with the gated command, writing the gate file to unblock the PTY.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
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 : )