fix(cli): forward --worktree to run_oneshot so -z -w works (#67458)

[webtecnica]

Closes #67458.

Problem

The --worktree / -w flag was silently ignored in one-shot mode (-z). Running hermes -z "make a commit" -w would commit directly to the current branch instead of using an isolated git worktree.

Root cause

run_oneshot() in hermes_cli/oneshot.py didn't accept a worktree parameter, and neither call site in hermes_cli/main.py forwarded the parsed worktree flag.

Fix

• Added worktree: bool = False parameter to run_oneshot()
• Wired worktree setup (imports from cli.py: _git_repo_root, _prune_stale_worktrees, _setup_worktree) and cleanup (_cleanup_worktree) — mirroring the interactive-mode path
• Set HERMES_CWD and TERMINAL_CWD env vars so the agent runs inside the worktree
• Clean up the worktree in the finally block (preserving it when it has unpushed commits — same policy as interactive mode)
• Forward worktree=getattr(args, "worktree", False) from both run_oneshot() call sites in main.py

Testing

• Verified the fix compiles and the diff is minimal (only 42 lines added across 2 files)
• The worktree functions reused from cli.py are already covered by tests/cli/test_worktree.py

[alt-glitch]

This was generated by AI during triage.

Related to #67471 and #67476 for #67458. This is a competing isolated-worktree implementation with different CWD/cleanup handling; #67471 rejects the unsupported combination and #67476 supplies a separately tested implementation. Maintainer choice requested.

[teknium1]

Thanks for covering both one-shot dispatch paths; the current-main omission is real (hermes_cli/main.py:12970-12976 and 15127-15133).

Problems

• hermes_cli/oneshot.py:251 invokes _setup_worktree() before stdout redirection, but that helper prints creation details (cli.py:1630-1632). The added cleanup at line 303 also prints to stdout (cli.py:1776-1777, 1811). This breaks the documented -z contract that stdout is the final response only (hermes_cli/main.py:15121-15122).
• Lines 257-258 replace HERMES_CWD and TERMINAL_CWD, but the finally block does not restore them. After cleanup removes the worktree, an in-process caller can retain a dangling TERMINAL_CWD, which agent/runtime_cwd.py:67-73 consumes.

Suggested changes

• Redirect setup and cleanup notices to stderr and restore both environment variables in the final cleanup path.
• Add forwarding and lifecycle regressions for both dispatches, stdout purity, setup failure, and environment restoration.

Automated hermes-sweeper review.

[teknium1]

_setup_worktree() prints creation messages to stdout (cli.py:1630-1632), but -z promises stdout contains only the final response. Route this helper's output to stderr before calling it.

[teknium1]

Save the prior CWD environment values and restore them in the final cleanup path. As written, a direct caller retains TERMINAL_CWD after _cleanup_worktree() may delete this path.

[teknium1]

_cleanup_worktree() prints both normal cleanup and unpushed-commit notices to stdout (cli.py:1776-1777, 1811). Redirect this call to real_stderr so it cannot contaminate one-shot output.

[GottZ]

This was generated by AI during triage.

Summary

Three PRs address #67458 by forwarding -w/--worktree through both one-shot dispatch paths and adding worktree setup and cleanup. #67471 changes the process cwd, #67476 retargets and restores TERMINAL_CWD while preserving stdout purity, and #67484 leaves the reviewed stdout and environment-restoration defects in its visible diff.

Related pull requests

• fix(cli): run -w/--worktree in one-shot mode instead of silently ignoring it (#67458) #67471 duplicate — (+854/-73) — n/a: Forwards worktree through both dispatch paths, creates the disposable worktree, changes the process cwd for the run, restores it, and adds lifecycle and dispatch tests. Despite the COMMENTED keep_open review on fix(cli): run -w/--worktree in one-shot mode instead of silently ignoring it (#67458) #67471, its revised diff now substantially overlaps fix(cli): honor -w/--worktree in one-shot mode instead of silently ignoring it #67476, which additionally retargets the runtime cwd signal used by agent tools.
• fix(cli): honor -w/--worktree in one-shot mode instead of silently ignoring it #67476 best fix — (+853/-0) — n/a: The recorded best-fix selection forwards both dispatch paths, fails closed on setup errors, retargets and restores TERMINAL_CWD, routes setup and cleanup notices to stderr, and adds lifecycle and dispatch regressions. Its visible diff addresses the COMMENTED keep_open review's stated issues concerning stdout pollution, pre-import state capture, and dispatch assertions.
• fix(cli): forward --worktree to run_oneshot so -z -w works (#67458) #67484 partial — (+42/-0) — n/a: Forwards both dispatch paths and creates a worktree, but its visible diff lets setup and cleanup helpers write to stdout, does not restore overwritten HERMES_CWD or TERMINAL_CWD, and adds no regression tests. Despite the COMMENTED keep_open review on fix(cli): forward --worktree to run_oneshot so -z -w works (#67458) #67484, those identified defects remain in the shown diff while fix(cli): honor -w/--worktree in one-shot mode instead of silently ignoring it #67476 covers the same fix path.

Duplicates

#67471 and #67476 substantially implement the same honor--w solution; #67484 overlaps that solution but retains the documented stdout and environment-restoration defects.

Suggested consolidation

Keep #67476 open with a salvage path preserving its two dispatch forwards, fail-closed setup, clean stdout, TERMINAL_CWD restoration, cleanup, and targeted regressions. Close #67471 as a duplicate of #67476 despite its keep_open review because its revised diff now implements the same product choice with less explicit runtime-cwd handling; close #67484 as a duplicate of #67476 despite its keep_open review because its visible diff retains that review's stdout and dangling-environment issues.

Complex graph

flowchart LR
    classDef open fill:#dbeafe,stroke:#1d4ed8,color:#1e3a8a
    classDef merged fill:#dcfce7,stroke:#15803d,color:#14532d
    classDef closed fill:#e5e7eb,stroke:#6b7280,color:#1f2937
    classDef unverified fill:#f3f4f6,stroke:#9ca3af,color:#374151
    classDef best stroke-width:3px,stroke:#b45309
    classDef target stroke-width:3px,stroke:#4338ca
    I67458(["issue #67458 (open)"])
    subgraph Dup67471 ["PRs duplicating each other"]
        P67471["PR #67471 (open)"]
        P67476["PR #67476 (open)"]
        P67484["PR #67484 (open)"]
    end
    P67484 -.->|partial| I67458
    class I67458 open
    class P67471 open
    class P67476 open
    class P67484 open
    class P67476 best
    class P67484 target
    click I67458 "https://github.com/NousResearch/hermes-agent/issues/67458"
    click P67471 "https://github.com/NousResearch/hermes-agent/pull/67471"
    click P67476 "https://github.com/NousResearch/hermes-agent/pull/67476"
    click P67484 "https://github.com/NousResearch/hermes-agent/pull/67484"

Loading

Graph: solid arrow = fixes / best fix, dashed arrow = partial or unverified (see edge label); boxed group = PRs duplicating each other; amber border = best fix; indigo border = target; gray node = closed (state tag in the node label).

Cross-PR triage: Reviewed 3 pull requests and 1 issue in this complex. Each diff was read against this issue; Assessment working set: 72 kB of PR diffs, 11 kB of issue/PR text, 10 kB of discussion (14 comments), 5 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.