Tools like file_write, file_edit, and shell now reset the agent's step
counter when called successfully. This allows the agent to continue
working without hitting the soft step-budget guardrail as long as it's
making active progress (writing files, running commands) rather than
looping on research/reading.
Changes:
- Add ResetsCounter bool to ToolInfo and MetaFile structs
- Parse resetsCounter from .meta JSON files
- Reset step counter in agent loop when ResetsCounter tool succeeds
- Add resetsCounter: true to file_write, file_edit, shell
- Reduce maxSteps from 50 to 25 (resetsCounter makes this safe)
- Rename AgentCfg.SaveSession/FlushSave to Save/Flush for brevity
- Remove restoreLegacySession — all sessions now use multi-agent format
- Pass Save/Flush callbacks to agents in CreateAgent and restore paths
- Add AuditLog fallback warning in NewAgent
The OutputFormat function was set once at runtime build time and stored
in ag.runtime.OutputFormat. When tools were dynamically loaded via
tool_load, the tool definitions were refreshed but the OutputFormat
lookup still used the stale runtime version that didn't know about
newly loaded tools.
Build the outputFormats map from the fresh tool list on each turn,
just like we already do for the tool definitions. This ensures
outputFormat (e.g., 'diff' for file_edit) is available immediately
after loading a tool.
ToolServer is vital infrastructure — it must never be nil. If the
initial dial fails, retry via ProcessKeeper (which respawns
ollie-remote). If still unavailable, fail agent creation with a
clear error asking the user to try again.
Reverted the nil-check band-aid in turn.go — ToolServer is now
guaranteed non-nil for any running agent.
Prevents nil pointer dereference when a newly created agent's
tool server connection failed to establish (keeper.Dial returned
nil, or session was restored with a dead toolsConn).
Previously the agent loop used a frozen Tools slice from BuildRuntime.
When tool_load promoted a tool, it was added to the registry but the
runtime's Tools slice was never updated — the LLM never saw the new
tool's schema until a session rebuild.
Fix: on each turn, call ToolServer.ListTools() to get the current set
of built-in + dynamically loaded tools, then convert to backend.Tool.
This makes tool_load behave identically to a built-in: the schema is
injected into the tool definitions on the very next turn.
Levi Neely
Ollie Agent