fix: resolve duplicate TERMINAL_CWD in _set_thread_env call

[wtj-0527]

Problem

When terminal.cwd is configured in config.yaml, get_profile_runtime_env() includes TERMINAL_CWD in the returned dict. Then streaming.py explicitly passes TERMINAL_CWD=str(s.workspace), causing:

TypeError: api.config._set_thread_env() got multiple values for keyword argument 'TERMINAL_CWD'

Root Cause

_profile_runtime_env dict may contain TERMINAL_CWD from profile config (via _TERMINAL_ENV_MAPPINGS['cwd'] = 'TERMINAL_CWD'), which conflicts with the explicit TERMINAL_CWD parameter passed to _set_thread_env().

Solution

Remove TERMINAL_CWD from _profile_runtime_env before calling _set_thread_env to avoid the duplicate keyword argument error. This ensures the session's workspace path is always used.

Testing

• Verified no TypeError in WebUI logs after fix
• Service starts and runs correctly
• Streaming API works without errors

[nesquena-hermes]

Thanks for the targeted fix, @KingBoyAndGirl!

What this does: Removes TERMINAL_CWD from _profile_runtime_env (via pop()) before calling _set_thread_env(), preventing the TypeError that fires when terminal.cwd is configured in a profile's config.yaml.

Note on approach vs. sibling PRs: There are now three PRs addressing the same root cause in issue #1270:

• This PR (fix: resolve duplicate TERMINAL_CWD in _set_thread_env call #1271) — pop() the conflicting key before the call (surgical, minimal change)
• PR fix: merge WebUI agent thread env before setting #1266 — merge the profile env and per-run env into a single dict before the call (handles all keys generically, not just TERMINAL_CWD)
• PR [codex] fix profile terminal env keyword collision #1265 (referenced in issue comments) — introduces a _build_agent_run_env() helper to encapsulate env construction

The pop() approach here is the most minimal and lowest-risk fix. The downside noted in the PR body is that the same fix would need to be repeated for each conflicting key (HERMES_EXEC_ASK, HERMES_SESSION_KEY, HERMES_HOME as called out in issue #1270). PR #1266's merge approach handles that more generically.

Recommendation for maintainer: Pick one of the three PRs to merge and close the other two to avoid a double-apply. All three are correct fixes; the choice is a style/scope preference.

Leaving for maintainer review and merge decision.

[wtj-0527]

Thanks for the summary, @nesquena-hermes! 🙏

I've noted that there are three PRs addressing the same issue (#1270):

• fix: resolve duplicate TERMINAL_CWD in _set_thread_env call #1271 (this PR) - pop() approach, minimal change
• fix: merge WebUI agent thread env before setting #1266 - merge env approach, handles all keys generically
• [codex] fix profile terminal env keyword collision #1265 - _build_agent_run_env() helper approach

My stance: I'm happy with whatever the maintainer decides! Each approach is correct; it's a style/scope preference.

If #1266 or #1265 is chosen, just let me know and I'll close this PR. The goal is to get the fix merged, not to push a specific implementation. 😊

Maintainers: please pick whichever approach you prefer!

[nesquena-hermes]

Thanks for flagging the three-PR situation, @KingBoyAndGirl — that context is helpful for the maintainer.

Quick summary for whoever merges:

• fix: resolve duplicate TERMINAL_CWD in _set_thread_env call #1271 (this PR) — pop() approach: removes TERMINAL_CWD from _profile_runtime_env before the _set_thread_env call. Minimal, surgical.
• fix: merge WebUI agent thread env before setting #1266 — merge approach: uses {**base, **overrides} to prevent duplicate keys generically across all env vars, not just TERMINAL_CWD.
• [codex] fix profile terminal env keyword collision #1265 — helper approach: extracts a _build_agent_run_env() helper to centralize env construction.

The merge approach (#1266) is the most defensive against future duplicate-key bugs. The pop() approach here is safe and correct for the current case. The helper approach is the most structural refactor.

Once one of these is merged, the other two should be closed. The maintainer's call on which shape they prefer.

[nesquena-hermes]

Hold — superseded by #1266

Thanks for the fix, @KingBoyAndGirl! The .pop('TERMINAL_CWD', None) approach is correct for unblocking the immediate TypeError. However, PR #1266 (from @hi-friday) solves the same problem at a higher level: it introduces _build_agent_thread_env() which merges all env keys into a single dict before _set_thread_env, so duplicates are impossible by construction — not just pruned case-by-case.

Since #1266 fully subsumes the fix in this PR and includes tests, we'll route through #1266 instead. Closing this one to keep the diff set clean.

We'll close this PR when #1266 lands. Thanks again for the sharp diagnosis!

[nesquena-hermes]

PR #1266 (which supersedes this) shipped in v0.50.240. Closing.