Repository Issue Activity (beta)

#2029 [BUG][Multi-Agent Structures][Every structure flattens the shared conversation into one user block, defeating the Agent-side transcript fix]

#2053 [BUG][Multi-Agent Structures][Seven more structures still flatten the shared conversation into one user blob]

#2202 [BUG][PlannerGeneratorEvaluator] The entire shared state file is pasted into every agent call, O(steps² × retries) (#2053 follow-up)

#2201 [BUG][ma_blocks] aggregate() flattens the worker conversation and records whole transcripts (#2053 follow-up)

#2200 [BUG][LLMCouncil] The Chairman gets the whole council as one user blob (#2053 follow-up)

#2198 [Bug][Memory] persistent_memory injects the whole MEMORY.md on every start, so startup context grows by ~5k characters per session, forever

#2193 [Bug][DebateWithJudge] run() and batched_run() reject img/imgs, while the telemetry decorator declares it captures them

#2194 [First-time contributor] Looking for beginner-friendly issues to contribute

#1748 [Bug][Autonomous loop flattens context into one string instead of a message array][Every iteration re-sends the whole transcript as a single user message, killing prompt caching and breaking tool_use_id pairing]

#1845 [Cleanup] agent.py _run_autonomous_loop — 862 lines in one method, 15 levels deep, with the dispatch block written twice (~250 lines)

#1844 [Cleanup][AOP] MCP tool closures re-wrap methods defined 30 lines above (~445 lines), plus redundant re-registration and an unreachable branch

#1841 [Cleanup] multi_agent_debates.py — one class template pasted seven times (~470 lines)

#1824 [Cleanup] base_swarm.py: remove 31 empty-body stubs + 19 unreferenced methods (~300 lines) and migrate its one subclass

#1853 [Bug] Six smaller correctness defects found during the code-waste audit

#2040 [BUG][MixtureOfAgents][Aggregator receives all layers flattened with no layer markers, and a System roster rendered as user prose]

#2033 [BUG][HierarchicalSwarm][Streaming path re-sends the full conversation while the sync path sends only a delta]

#2159 [Bug][Agent] agent.usage reports zero for streaming runs — the usage chunk is never read off the stream

#1973 [BUG] Sub-agents are created with no tools, making delegation near-useless

#1924 [Bug][Agent][Tool failures are captured as Agent.llm_error and recovered by re-running the model, so one broken tool costs up to 3 extra completions]

#1983 [FEAT] Add a glob tool for pattern-based file discovery

#1791 [Bug][Autonomous loop][File tools accept ../ and absolute paths, so the agent can read, write and delete anywhere on the filesystem]

#2168 [Cleanup][Agent] Tool results are recorded twice: drop the 13 'Tool Executor' prose rows now that Conversation holds typed tool rows

#2165 [FEAT][Autonomous loop] Run independent tool calls in parallel, with same-file calls kept in order

#2164 [FEAT][Conversation] Represent tool calls and tool results in Conversation so the parallel Transcript can be deleted

#2163 [FEAT][Autonomous loop] Per-tool permission policy (allow / deny / ask) and a workspace sandbox, replacing the bash blocklist