Long run_bash output in omni-agent is truncated before actionable error lines, causing persistent agents to miss the important tail of build/log output.
Observed symptom:
agentd logs omni-coder --mode persistent --last ... shows tool_result output payload cut mid-stream for large commands like bild --loud ....Goal:
tail, rg, or sed against a saved file.Proposed behavior:
run_bash (and run-as wrapper), when combined stdout/stderr exceeds threshold:_/tmp/.Acceptance:
tail -n or rg on that file.typecheck.sh Omni/Agent.hs and bild --test Omni/Agent.hs pass.Follow-up commit ab9c64d0: moved truncation metadata + 'Full output saved to:' to the top of run_bash truncated output so short tool_result snippets still include the actionable file path.
Implemented in commit 16738f22.
What changed
Omni/Agent/Tools.hs(run_bash):omni-agent-run-bash-*.log).Full output saved to: <path>linetail/rg)runBashTool persists oversized output to temp file.Verification
lint Omni/Agent/Tools.hstypecheck.sh Omni/Agent.hsbild --test Omni/Agent.hsomni-coder:Full output saved to: ...Rollout
Omni/Agentd.hsso bundled runtime resolves to updatedagentstore path.intent-coder,tax-prep-2025,skills-cli,sd-coder,cfg-coder,omni-coder,paracoder)./nix/store/gg0nsavpndahrdqzvidb1wqs582jrlgx-agent/bin/agent