* fix: don't drop pending tool-call-marker prefix on tool.started/run.done
The `filterBridgeToolCallMarkupDelta` filter holds back any text that
ends in a partial prefix of `[Calling tool:` (i.e. `[`, `[C`, `[Ca`,
..., `[Calling tool`) so it can decide whether the buffered chars are
the start of a tool-call markup block to be hidden, or just regular
text to be released by the next delta.
The bug: that "release on next delta" assumption breaks at TWO points:
1. **On `tool.started`**: the next chunk for this assistant message is
the tool call itself, NOT a follow-up text delta. Buffered chars
sit there forever and nothing flushes them — they vanish silently
from the user-visible stream.
2. **On run completion**: the code did
`state.bridgePendingToolCallMarkup = undefined` directly, dropping
any pending chars without forwarding them.
Both cases produce the user-visible symptom of "abrupt cuts in text
right before/after tool calls (terminal, read_file, write_file...)" —
1 to 13 characters disappear at exactly the boundary where the model
was emitting natural prose that happened to end with `[`.
The fix introduces `flushPendingToolCallMarkup(state)` and calls it:
- In the `tool.started` branch BEFORE recording the tool call, so the
buffered chars are appended to the open assistant message and emitted
as a normal `message.delta` to the client.
- At run-done BEFORE clearing the buffer, same flush path.
This is a pure recovery patch — no change to the marker detection
logic itself. If the buffer turns out to actually be a real
`[Calling tool: ...]` marker that just hasn't completed yet, that
case is still caught by the existing `markerIdx >= 0` branch in the
filter on the next delta. The only behavioral change is that the
"orphan" cases (text that ends with `[` but never becomes a marker)
are no longer dropped.
* fix bridge marker flush persistence
---------
Co-authored-by: Paulo Cavallari <paulocavallari@users.noreply.github.com>
* Allow bridge sessions to run concurrently
* Stabilize bridge concurrency test
* Set bridge approval timeout to 120 seconds
* harden bridge approval concurrency
---------
Co-authored-by: Codex <codex@openai.com>
* fix: recursive skill scan for nested sub-category directories
The Web UI skill scanner (scanSkillsDir) only checked one level deep:
skills/<category>/<subdir>/SKILL.md. Sub-category containers like
mlops/evaluation/ (which has DESCRIPTION.md + subdirs but no SKILL.md)
were skipped entirely, hiding all 12 nested skills under mlops.
Changes:
- scanSkillsDir: extract collectSkills() recursive helper that depth-first
searches for SKILL.md at every level under a category directory.
Directories without SKILL.md but with subdirectories are recursively
descended into.
- listFiles handler: replace hardcoded join(category, skill) path with
recursive findSkillDir() search, so nested skill file browsing works
(e.g., mlops/evaluation/lm-evaluation-harness).
Fixes mlops category showing 1 skill instead of 13. All 20 other
categories verified with zero regression.
* fix: pA handler also needs recursive search for nested skill file content
The readFile_ (pA) handler was constructing direct paths like
skills/category/skill/... which fails for nested sub-category
skills (mlops/evaluation/lm-evaluation-harness). Added fallback
recursive search when direct path returns 404.
Also fixed listFiles (sA) handler to use recursive search for
the same reason - previous fix to dist was not in source TS.
Verified:
- lm-evaluation-harness SKILL.md content: 200 ✅
- vllm SKILL.md: 200 ✅
- huggingface-hub (non-nested): 200 ✅
- reference file in nested skill: 200 ✅
* fix: pA handler also needs recursive search for nested skill file content
The readFile_ (pA) handler was constructing direct paths like
skills/category/skill/... which fails for nested sub-category
skills (mlops/evaluation/lm-evaluation-harness). Added fallback
recursive search when direct path returns 404.
Also fixed listFiles (sA) handler to use recursive search for
the same reason - previous fix to dist was not in source TS.
Verified:
- lm-evaluation-harness SKILL.md content: 200 ✅
- vllm SKILL.md: 200 ✅
- huggingface-hub (non-nested): 200 ✅
- reference file in nested skill: 200 ✅
* harden recursive skill lookup
---------
Co-authored-by: gs <gs@localhost>
Co-authored-by: gutanulaif <gutanulaifa@gmail.com>
Add modular group-chat mention routing helpers for the reserved @all token, route it to every non-sender agent, and strip routing tokens before model input.
Expose @all in mention autocomplete, highlight it in group messages, reserve literal all agent names, and cover boundary/partial-match regressions with tests.
* feat: support profile-aware group chat bridge flows
* feat: route cron jobs through hermes cli
* Fix group chat routing and isolate bridge tests
* Add Grok image-to-video media skill
* Default Grok videos to media directory
* Fix bridge profile fallback and cron repeat clearing
* Refine bridge chat and gateway platform handling
* Filter bridge tool-call text deltas
* Preserve structured bridge chat history
* Prepare beta release build artifacts
* Fix Windows run profile resolution
* Fix Windows path compatibility checks
* Fix profile-scoped model page display
* Hide Windows subprocess windows for jobs and updates
* Hide Windows file backend subprocess windows
* Avoid Windows gateway restart lock conflicts
* Treat Windows gateway lock as running on startup
* Force release Windows gateway lock on restart
* Tighten Windows gateway lock cleanup
* Update chat e2e source expectation
* Bump package version to 0.5.30
---------
Co-authored-by: Codex <codex@openai.com>
ekko