agent: keep enqueue time on exported coop messages

[hhh2210]

The bug

sent_messages stores to and content only. _extract_conversation then copies a missing timestamp. _message_timestamp_key documents ISO input, but float(ts) rejects those strings and returns 0. Missing timestamps sort first, and the stable sort groups by sender, so a reply can be written before the message it answers.

send_and_wait returns after the peer replies. A timestamp taken at that point records the reply, not the send. Redis already stores an ISO timestamp at rpush.

The fix

On a successful send, keep that enqueue timestamp on the connector and copy it onto sent_messages for both plain sends and --wait. The sort key parses ISO-8601, including a trailing Z, and rejects non-finite numeric values such as NaN and infinities in both numeric and string form. Missing, malformed, and non-finite values sort as 0; a historical record with a null timestamp stays null.

Verification

Local validation on Python 3.13.13:

• python -m pytest tests/runner/test_coop.py tests/agents/mini_swe_agent_v2/test_sent_timestamp.py — 20 passed. Covers enqueue-time propagation, blocking sends, ISO/UTC offsets, float/string non-finite values, overflow-to-infinity strings, and stable mixed ordering. The non-finite regressions fail before the follow-up fix.
• python -m pytest tests/ -v --tb=short — 442 passed, 84 skipped. Docker/Modal/GCP integration tests retain their default skips. For this run only, Git tag/commit signing was disabled via GIT_CONFIG_COUNT overrides so personal signing settings do not alter temporary test repositories; no Git config file was changed. LITELLM_LOCAL_MODEL_COST_MAP=true was set to avoid fetching a price map in unit tests.
• ruff check src/cooperbench/, ruff format --check src/cooperbench/, and mypy src/cooperbench/ — all passed (76 source files).
• Ruff check/format for both changed test files and git diff --check — passed.

The full-suite run emits two pre-existing unknown-marker warnings for slow and timeout. No remote CI result is available yet; local checks do not establish the Linux/Python matrix.

Left out

The no-git prompt that tells an agent to fetch origin/<peer> is a separate PR.

Checklist

• Code follows project style
• Tests pass locally
• Documentation updated (if needed)

Made with Cursor

[Copilot]

Copilot review overview

🟡 Changes recommended

Non-finite numeric timestamps must be rejected before sorting.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 1

Open (1)

• Reject non-finite timestamp values before sorting · New

What changed in this PR

Updates cooperative-message exports to preserve enqueue timestamps and sort mixed timestamp formats chronologically.

Changes:

• Records enqueue timestamps for normal and blocking sends.
• Parses ISO-8601 timestamps, including trailing Z.
• Adds regression tests for timestamp propagation and ordering.

File	Description
tests/​runner/​test_coop.py	Tests timestamp ordering behavior.
tests/​agents/​mini_swe_agent_v2/​test_sent_timestamp.py	Tests timestamp propagation.
src/​cooperbench/​runner/​coop.py	Normalizes message timestamps for sorting.
src/​cooperbench/​agents/​mini_swe_agent_v2/​connectors/​messaging.py	Stores enqueue timestamps.
src/​cooperbench/​agents/​mini_swe_agent_v2/​agents/​default.py	Copies timestamps into sent-message records.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Copilot review overview

🟡 Changes recommended

Extreme numeric and ISO timestamps can still raise exceptions and abort conversation export.

Get a fresh assessment by requesting another Copilot review.

Review effort: Balanced
Findings: 1

Open (1)

• Handle overflow when converting large JSON timestamps · New

Resolved since last review (1)

• Reject non-finite timestamp values before sorting