fix: preserve cwd for Codex handoffs

[ai-jz]

What does this PR do?

Preserves the CLI working directory when /handoff <platform> moves a session into the gateway, so Codex app-server starts in the original project instead of the gateway install directory.

The change keeps the scope narrow:

• capture the CLI cwd when /handoff marks the session pending
• persist it on the handoff row and destination gateway session entry
• pass it to the cached AIAgent for that gateway session
• use it only when spawning the Codex app-server subprocess

Why?

Without this, a handoff from a project directory can continue from Slack while Codex app-server starts from the Hermes install cwd, so Codex loads the wrong local context and answers from the wrong workspace.

Related: #21427 covers broader resume cwd restoration. This PR is narrower and handoff-specific.

How has this been tested?

• scripts/run_tests.sh tests/hermes_cli/test_session_handoff.py tests/gateway/test_session.py tests/agent/transports/test_codex_app_server_runtime.py -q
• git diff --check

Platform tested: macOS.

[teknium1]

Thanks for the focused handoff fix. The underlying failure remains on current main: /handoff calls request_handoff() without a cwd (hermes_cli/cli_commands_mixin.py:623), while Codex falls back to resolve_agent_cwd() when agent.session_cwd is absent (agent/codex_runtime.py:342-345).

Problems

• Current main already persists the CLI workspace as sessions.cwd (hermes_state.py:743; run_agent.py:600-609). Adding a separate handoff_cwd column would create competing cwd state.
• Setting only agent.session_cwd is incomplete. Gateway cwd resolution is also bound through gateway/session_context.py:209-211, but GatewayRunner._set_session_env() currently does not pass a cwd (gateway/run.py:15000-15011). That leaves context-file discovery and other runtime-cwd consumers on the gateway fallback.
• The patch's implementation targets have moved: /handoff is now in hermes_cli/cli_commands_mixin.py:518, and the Codex spawn is in agent/codex_runtime.py:342-345.

Suggested changes

• Rework the salvage around sessions.cwd, propagate it through the gateway session-context path, and add an end-to-end handoff-cwd regression rather than only a Codex helper test.

Automated hermes-sweeper review.

[GottZ]

This was generated by AI during triage.

Summary

Two PRs touch the CLI-to-gateway handoff path but address distinct causes: #27326 preserves the originating workspace for Codex after handoff, while #65505 carries an explicit Discord initiator into newly created handoff threads. Neither current diff is merge-ready because each introduces or reuses the wrong persistence path for its data.

Related pull requests

• #27326 related — (+138/-15) — keep open and rework: The diff captures the CLI cwd and propagates it to the Codex app-server, addressing the reported wrong-workspace failure, but it creates a competing handoff_cwd alongside the existing sessions.cwd and updates only agent.session_cwd, leaving gateway context-file and other runtime-cwd consumers on fallback resolution. This agrees with the keep_open review on #27326: salvage the fix by using sessions.cwd and propagating it through the current gateway cwd-resolution path.
• #65505 related — (+199/-10) — keep open and rework separately: The diff adds the missing Discord initiator argument and inserts that member into both direct and fallback-created threads, but transports the value through durable sessions.user_id; this can overwrite ownership metadata and clear it for existing no-ID handoff callers. This agrees with the keep_open review on #65505: use a handoff-specific initiator field and add preservation coverage for both explicit-ID and no-ID paths.

Suggested consolidation

Do not merge or deduplicate these PRs as-is: rework #27326 around the canonical sessions.cwd and complete gateway cwd propagation, and keep #65505 separate because it fixes Discord thread membership rather than cwd restoration. After their respective blocking review concerns are resolved, they can be evaluated independently; neither PR should be closed as a duplicate of the other.

Cross-PR triage: Reviewed 2 pull requests and 0 issues in this complex. Each diff was read against this issue; Assessment working set: 32 kB of PR diffs, 3 kB of issue/PR text, 3 kB of discussion (4 comments), 0 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.