[xizhilanre]

Problem

speckit.git.feature runs as the mandatory before_specify hook, and its feature-branch scripts run git fetch --all --prune for non-dry-run invocations. When a remote is HTTPS and has no usable credential helper, git writes Username for 'https://…': to the tty — the output redirect does not hide it — and waits for input. /speckit.specify then appears to hang.

The dry-run path already guards git ls-remote with GIT_TERMINAL_PROMPT=0; the real fetch path did not, in any of the three twins.

Root cause

Each implementation called the fetch without the guard:

file	before
extensions/git/scripts/bash/create-new-feature-branch.sh	git fetch --all --prune >/dev/null 2>&1 || true
extensions/git/scripts/powershell/create-new-feature-branch.ps1	git fetch --all --prune 2>$null | Out-Null
extensions/git/scripts/python/create_new_feature_branch.py	subprocess.run(["git", "fetch", "--all", "--prune"], …) with no env

Change

Apply the same GIT_TERMINAL_PROMPT=0 guard the dry-run path uses, in all three twins, so the fetch fails fast and || true / catch / nothing-to-check continues:

• bash: GIT_TERMINAL_PROMPT=0 git fetch --all --prune >/dev/null 2>&1 || true
• PowerShell: set $env:GIT_TERMINAL_PROMPT = '0' around the fetch, restore to $null afterwards (the same idiom already used for git ls-remote in the same file)
• Python: env={**os.environ, "GIT_TERMINAL_PROMPT": "0"}

Verification

New tests/test_git_fetch_prompt_guard.py (3 tests, one per implementation). Each registers a git shim that records GIT_TERMINAL_PROMPT when fetch is invoked and passes every other call through to the real git, then drives the real script and asserts the recorded value is 0.

uv run python -m pytest tests/test_git_fetch_prompt_guard.py -q

with the fix:                        2 passed, 1 skipped
with extensions/git/scripts reverted: 2 failed, 1 skipped

The Python and PowerShell tests fail without the change with AssertionError: assert '' == '0' — the guard was absent, which is the regression this pins. The bash test is skipped on this Windows host (only the WSL bash launcher is installed, which tests/conftest.requires_bash rejects); it runs on the Linux CI.

Related suites, same command, before vs after (the 25 remaining failures/errors are pre-existing on this host — no native bash, no PS 5.1 harness setup — and are identical on the unmodified main tree):

uv run python -m pytest tests/test_git_fetch_prompt_guard.py tests/test_timestamp_branches.py tests/test_init_dir.py -q
before: 27 failed, 18 passed, 73 skipped
after:  25 failed, 20 passed, 73 skipped

The two-test delta is exactly the new regression tests flipping from fail to pass.

Windows PowerShell 5.1 evidence

The PowerShell twin was exercised under Windows PowerShell 5.1.22621.4391 (dot-sourced with a git function shim, in an ASCII-path temp checkout):

PS> powershell.exe -NoProfile -ExecutionPolicy Bypass -File h.ps1
BRANCH_NAME: 001-probe-ps51
FEATURE_NUM: 001
# recorded GIT_TERMINAL_PROMPT at fetch = '0'

Only PS 5.1 constructs are used (try/catch, $env: assignment).

Lint

uvx ruff@0.15.0 check src tests   # the CI command

passes for the changed test file; the extension scripts are outside that CI gate.

AI disclosure

Implemented with DeepSeek Harness (agent) using model deepseek-v4.1-flash, running autonomously. Extent: code generation for the three guard lines and the regression tests, plus running the test evidence above. The submitting account holder reviewed the diff and is the one responding to review.

Closes #5199

[Copilot]

🟡 Changes recommended

A critical release-delivery issue remains because the bundled script changes are not accompanied by the required extension manifest and catalog version bump, so existing installations will not receive the fix.

1 open issue

• Bump the Git extension version in the manifest and catalog · New

What changed in this PR

This pull request prevents Git feature-branch fetches from blocking on HTTPS credential prompts by disabling terminal prompting across Bash, PowerShell, and Python implementations. It also adds regression tests for the guard.

Changes:

• Sets GIT_TERMINAL_PROMPT=0 for non-dry-run fetches in all three script variants.
• Adds coverage verifying the environment variable during fetch operations.
• Requires a synchronized extension version bump in extension.yml and extensions/catalog.json.
• Requires the PowerShell test to fall back to Windows PowerShell 5.1 when pwsh is unavailable.

This pull request requires changes.

File	Description
tests/​test_git_fetch_prompt_guard.py	Adds regression tests for Python, Bash, and PowerShell, but currently skips the PowerShell path on supported Windows hosts that only provide Windows PowerShell 5.1.
extensions/​git/​scripts/​python/​create_new_feature_branch.py	Passes GIT_TERMINAL_PROMPT=0 to the fetch subprocess.
extensions/​git/​scripts/​powershell/​create-new-feature-branch.ps1	Temporarily disables terminal prompts around the PowerShell fetch.
extensions/​git/​scripts/​bash/​create-new-feature-branch.sh	Disables terminal prompts for the Bash fetch; the bundled extension version metadata must also be bumped for this content change to be released.

_{Review effort: Lite (auto)}

[Copilot]

🟡 Changes recommended

The manifest version bump leaves the bundled Git extension test asserting the old version, creating an active critical test failure that must be fixed.

2 open issues

• Update the git extension version assertion to 1.0.2 · New
• Bump the Git extension version in the manifest and catalog

_{Review effort: Lite (auto)}

[mnriem]

Please address Copilot feedback, resolve conflicts and address test & lint errors