[JakkulaVeerababu]

Description

Fixes #5199

Before it numbers a new feature branch, the git extension runs git fetch --all --prune with its output redirected. Git writes credential prompts to the terminal, not to stdout or stderr. So when a remote uses HTTPS and has no usable credential helper, the fetch stops at Username for 'https://github.com': and waits. speckit.git.feature is the mandatory before_specify hook, so in an agent session /speckit.specify appears to hang. The dry-run path already runs ls-remote with GIT_TERMINAL_PROMPT=0. The fetch path did not.

This PR sets GIT_TERMINAL_PROMPT=0 for that fetch in all three scripts:

• Bash: GIT_TERMINAL_PROMPT=0 git fetch --all --prune >/dev/null 2>&1 || true
• Python: env={**os.environ, "GIT_TERMINAL_PROMPT": "0"} on the fetch subprocess.run
• PowerShell: sets the variable around the fetch in try/finally, then restores the caller's value (or removes the variable if it was unset), so nothing leaks into the calling session.

A failed fetch is still ignored, as before. Numbering then uses local branches and specs/.

Out of scope: SSH passphrase prompts. GIT_TERMINAL_PROMPT does not cover them, and forcing ssh -o BatchMode=yes would override a user's own GIT_SSH_COMMAND or core.sshCommand. That is the open question in the bug assessment, so I left it for a separate decision.

Testing

The new tests use a probe remote. It is an ssh:// URL whose core.sshCommand records the GIT_TERMINAL_PROMPT value it inherits from the fetch (or ls-remote), then fails. No network is used.

• TestCreateFeatureBash::test_remote_lookup_disables_terminal_prompt[fetch|dry-run]
• TestCreateFeaturePowerShell::test_remote_lookup_disables_terminal_prompt[fetch|dry-run]
• TestCreateFeaturePowerShell::test_fetch_restores_caller_terminal_prompt[set|unset]
• TestCreateFeatureBranchParity::test_fetch_disables_terminal_prompt (Bash and Python twins)

Results on Ubuntu 26.04 (WSL2), Python 3.13, bash 5.3, pwsh 7.6.8, git 2.53:

• main with only the new tests: 5 failed, 2 passed. The probe saw unset on the fetch path. The two dry-run cases already passed.
• This branch: 7 passed. tests/extensions/git: 181 passed, 2 skipped.
• Full suite: 13927 passed, 166 skipped, 11 failed. The same 11 tests (composed-template cases in test_*_python_parity.py) also fail on unmodified main in this environment. They do not touch the git extension.
• shellcheck --severity=error on the changed Bash script: clean.

Windows PowerShell 5.1 (5.1.26100, Desktop edition) with Git for Windows 2.55, same probe remote, non-dry-run:

Run	GIT_TERMINAL_PROMPT seen by the fetch	Caller's value after the script
main	unset	unset
this PR	0	unset
this PR, caller had 1	0	1

• Tested locally with uv run specify --help
• Ran existing tests (uv sync --extra test, then .venv/bin/python -m pytest, as AGENTS.md recommends)
• Tested with a sample project (if applicable)

Manual test results

Agent: Claude Code (Claude Opus 5.5), autonomous | OS/Shell: Ubuntu 26.04 on WSL2 / bash 5.3, pwsh 7.6.8, python 3.13

Each project was scaffolded from the branch under test with specify init <dir> --integration claude --script {sh,ps,py} --extension git. Its only remote was https://github.com/github/spec-kit-issue-5199-does-not-exist.git (answers 401), with no credential helper. The agent ran the script line that speckit.git.feature instructs it to run, under a pseudo-terminal whose stdin stays open and silent, like an agent terminal that nobody types into. The script lines were run directly, not from an interactive agent session.

Command tested	main	this PR
/speckit.git.feature (Bash script)	Stuck at Username for 'https://github.com':, killed after 20 s, no branch	Exit 0 in 1 s, 001-probe created
/speckit.git.feature (PowerShell script)	Same hang	Exit 0 in 3 s, 001-probe created
/speckit.git.feature (Python script)	Same hang	Exit 0 in under 1 s, 001-probe created

Test selection reasoning

Changed file	Affects	Test	Why
extensions/git/scripts/bash/create-new-feature-branch.sh	/speckit.git.feature, and /speckit.specify through the before_specify hook	T1	The command runs this script
extensions/git/scripts/powershell/create-new-feature-branch.ps1	Same	T1	PowerShell twin
extensions/git/scripts/python/create_new_feature_branch.py	Same	T1	Python twin
tests/extensions/git/*.py	None	–	Tests only

Required tests

• T1: /speckit.git.feature, once per script flavour. The rest of /speckit.specify is unchanged; its only changed step is this hook.

AI Disclosure

• I did not use AI assistance for this contribution
• I did use AI assistance (fill in the disclosure below)

AI disclosure: Claude Code (model: Claude Opus 5.5, maximum reasoning effort) ran autonomously on behalf of the account owner, JakkulaVeerababu. It did the investigation, wrote the code change in all three scripts and the tests, ran the checks above, and wrote this description. The commit has an Assisted-by: Claude Code (model: Claude Opus 5.5, autonomous) trailer. The account owner had not reviewed the change line by line when this PR was opened. That is why it is a draft: they will review it and then mark it ready.

🤖 Generated with Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.