fix(codex): store command output whole in result_summary

[joeysbase]

Problem

The Codex agent truncated command output with f"Output: {output[:100]}" when building CommandTelemetry.result_summary. Because result_tokens is derived from result_summary length (ceil(len/4)), this pinned ~77% of Bash results near a ~31-token floor and under-reported every Codex tool result, skewing the cost model.

This also violated the codebase-wide untruncated-result_summary contract (the Claude agent already stores full output; test_resolve_pending_command_does_not_truncate_long_result_summary asserts a 5000-char summary survives). Codex was the outlier.

Change

• codex_agent.py: store command output whole in result_summary (display trimming belongs in renderers/reports, not capture).
• CE043 lint rule (tests/lint/rules/ce043_no_command_output_truncation.py, wired into runner.py): forbids constant-slice truncation of captured command output (output[:N], .stdout[:N], …) inside agents/, so the contract can't silently regress. Scoped to agents/, with a # noqa: CE043 escape hatch.
• Regression test (test_command_output_recorded_whole_not_truncated) + full lint-rule unit tests.

Code-review fixes applied on top

• streaming/renderers.py: cap the Rich live-feed ToolEnd preview at _MAX_RESULT_LEN so a now-whole result_summary can't flood the console; the reported char count still shows the true full length (+ regression test).
• models/telemetry.py: narrow the result_tokens docstring — the untruncated contract covers captured command stdout/stderr, not the intentionally-brief one-line summaries of non-command tool items (_summarize_tool_item).
• experiments/default.yaml: max_turns raised to 100 (intentional, for larger task runs) with the comment clarified as a hard safety ceiling; task_timeout/turn_timeout remain the practical guards.

Verification

make verify green — ruff, pyright (0 errors), custom lint (incl. CE043), 4210 tests, coverage 91.72%.

🤖 Generated with Claude Code

[github-actions]

Claude finished @joeysbase's task in 1m 36s —— View job

---

Code Review in Progress

Todo List:

• Read .github/code_review.md for review standards
• Read CLAUDE.md for project conventions
• Get full diff with git diff origin/main...HEAD
• Review each changed file with full context
• Check cross-file consistency
• Analyze what's missing
• Provide comprehensive review feedback