Repository Issue Activity (beta)

#5545 Dependency minimums raising is failing

#5309 Extensions API

#5398 feat(opensandbox): add OpenSandbox sandbox integration

#5527 LSP-backed post-edit diagnostics

#5589 `edit_file` with an empty `old_string` can corrupt file contents

#5588 `edit_file` with an empty `old_string` can corrupt file contents

#5587 Add opt-in Phoenix tracing to dcode

#5585 Separate user-authored AGENTS.md instructions from writable agent memory

#5583 Separate user-authored AGENTS.md instructions from writable agent memory

#5575 Arrow keys and word jumps should treat media/paste placeholders as single units

#5574 Dragging a folder into the chat input turns the draft into a slash command

#5573 Undoing a deleted [image N] placeholder restores bare text — the attachment is silently dropped

#2982 Generative UI from subagents: events don't reach the client

#4662 `invalid_tool_calls` end the run silently and deferred repair is marked successful

#5112 BaseSandbox.grep path globs fail because the inline Python command is not shell-safe

#5560 Large-result eviction silently skipped when tool_call_id is long (LiteLLM Gemini ids carry ~1.5KB thought signatures) — filename over OS limit, write fails, original kept

#3930 `StateBackend` / `StoreBackend` `ls()` returns empty `LsResult` for missing paths (inconsistent with `FilesystemBackend`)

#4420 TypeError: 'LsResult' object is not iterable in CompositeBackend.als()

#4978 `glob()` recursion semantics differ across backends

#5203 Distinguish empty and whitespace-only files in `read_file` output

#5559 Large-result eviction derives unbounded paths and notices from `tool_call_id` — breaks on LiteLLM Gemini ids carrying ~1.5KB thought signatures

#2941 Custom filesystem tool message after call tool

#4490 Add Resume Review Agent example to examples/

#5136 A task subagent's state can't be read back: 'Subgraph tools not found'

#5551 Backport `ask_user` middleware, question schema, and validation into deepagents