Activity

1. mnriem

   added

   triage-must-haveVerdict: high-value, important work for Spec Kit — do firstVerdict: high-value, important work for Spec Kit — do first

   bug-assess

   on Oct 30, 2026
2. github-actions commented on Oct 30, 2026

   

   github-actionsBot

   on Oct 30, 2026 – with GitHub Actions

   Contributor

   More actions

   **Bug assessment — git-fetch-prompt-block: Valid · severity medium

   ---

   Bug Assessment: git extension fetch can block on hidden credential prompt

   • Slug: git-fetch-prompt-block
   • Created: 2026-10-30T17:30:03Z
   • Source: issue [Bug]: git extension: non-dry-run git fetch can block /speckit.specify on a hidden credential prompt #5199
   • Verdict: valid
   • Severity: medium

   Report (summarized)

   Issue #5199 reports that /speckit.specify can appear to hang when the git extension executes a non-dry-run git fetch --all --prune against an HTTPS remote that requires credentials but has no usable helper. The report includes a minimal reproduction showing that redirecting stdout/stderr does not suppress terminal credential prompts, while GIT_TERMINAL_PROMPT=0 makes the command fail fast.

   Repository evidence matches the report: dry-run remote probing uses GIT_TERMINAL_PROMPT=0 for ls-remote, while non-dry-run fetch paths do not set this variable in Bash, PowerShell, or Python.

   Symptom

   When speckit.git.feature runs before speckit.specify, the non-dry-run fetch can block on an interactive username/password prompt from git, causing /speckit.specify to stall in agent/non-interactive sessions. Expected behavior is fast failure (exit code) with continued script flow, not an interactive wait.

   Reproduction

   1. Initialize a scratch repository and configure an HTTPS remote that requires authentication (or is otherwise credential-challenging).
   2. Run git -c credential.helper= fetch --all --prune >/dev/null 2>&1.
   3. Observe the terminal prompt (for example Username for 'https://github.com':) and blocking behavior.
   4. Run GIT_TERMINAL_PROMPT=0 git -c credential.helper= fetch --all --prune >/dev/null 2>&1 and observe immediate non-interactive failure.
   5. Trigger /speckit.specify with the git extension enabled so the before_specify hook executes speckit.git.feature.

   Suspected Code Paths

   • extensions/git/scripts/bash/create-new-feature-branch.sh:172 — dry-run remote check correctly uses GIT_TERMINAL_PROMPT=0 for git ls-remote.
   • extensions/git/scripts/bash/create-new-feature-branch.sh:194 — non-dry-run branch-number path calls git fetch --all --prune without GIT_TERMINAL_PROMPT=0.
   • extensions/git/scripts/bash/create-new-feature-branch.sh:527 — non-dry-run main flow calls check_existing_branches(..., false, ...), taking the interactive fetch path.
   • extensions/git/scripts/powershell/create-new-feature-branch.ps1:135 — dry-run remote check sets $env:GIT_TERMINAL_PROMPT='0' for git ls-remote.
   • extensions/git/scripts/powershell/create-new-feature-branch.ps1:166 — non-dry-run path runs git fetch --all --prune without non-interactive guard.
   • extensions/git/scripts/powershell/create-new-feature-branch.ps1:483 — non-dry-run main flow calls Get-NextBranchNumber without -SkipFetch.
   • extensions/git/scripts/python/create_new_feature_branch.py:252 — dry-run remote check uses env_extra={"GIT_TERMINAL_PROMPT": "0"} for ls-remote.
   • extensions/git/scripts/python/create_new_feature_branch.py:269-270 — non-dry-run path runs subprocess.run(["git", "fetch", "--all", "--prune"], ...) without env override.
   • extensions/git/scripts/python/create_new_feature_branch.py:526 — non-dry-run main flow calls check_existing_branches(...), entering the unguarded fetch path.
   • extensions/git/extension.yml:47-49 — before_specify hook for speckit.git.feature is optional: false, so this path is routinely exercised when the extension is enabled.

   Root Cause Hypothesis

   The script family intentionally hardens dry-run remote inspection (ls-remote) against interactive prompts, but the corresponding non-dry-run fetch path was not given the same non-interactive environment. Because git credential prompts are tty-directed and not reliably suppressed by redirecting stdout/stderr, these fetch calls can block execution despite output redirection. Confidence: high.

   Proposed Remediation

   Preferred: apply a consistent non-interactive guard to all non-dry-run fetch calls used for branch-number discovery. In Bash, prefix fetch with GIT_TERMINAL_PROMPT=0; in PowerShell, scope $env:GIT_TERMINAL_PROMPT='0' around the fetch and restore prior value; in Python, pass env={**os.environ, "GIT_TERMINAL_PROMPT": "0"} (or existing helper equivalent) to subprocess.run. Keep existing behavior of tolerating fetch failure (|| true / equivalent) so offline and restricted environments continue gracefully.

   Alternatives (optional):

   • Skip fetch entirely and rely only on local refs + specs in non-dry-run mode. Trade-off: avoids prompt risk but can allocate branch numbers behind remote state more often.
   • Also add SSH non-interactive guard (e.g., batch mode) where appropriate. Trade-off: broader behavior change and higher compatibility risk across user SSH setups.

   Files likely to change:

   • extensions/git/scripts/bash/create-new-feature-branch.sh
   • extensions/git/scripts/powershell/create-new-feature-branch.ps1
   • extensions/git/scripts/python/create_new_feature_branch.py
   • tests/extensions/git/test_git_extension_python_parity.py

   Tests to add or update:

   • Add regression tests that assert non-dry-run branch-number discovery invokes fetch with non-interactive environment in all three implementations.
   • Add parity coverage ensuring dry-run and non-dry-run prompt-handling strategy is consistent across Bash/PowerShell/Python.
   • Add a behavior test that simulates credential-challenging remotes and verifies command returns promptly instead of hanging (mock/stub subprocess or git wrapper in deterministic test harness).

   Risks & Considerations

   • PowerShell env scoping must avoid leaking GIT_TERMINAL_PROMPT to later commands.
   • Enforcing non-interactive mode may change user-visible errors from prompt-driven retries to immediate failures; this is desirable for automation but should be documented in test expectations.
   • Ensure no regression in offline/no-git graceful-degradation paths.

   Open Questions

   • [NEEDS CLARIFICATION: Should SSH auth/passphrase prompts also be suppressed in this path for full non-interactive parity, or is HTTPS credential prompt suppression the intended scope?]

   Generated by 🐛 Assess Bug from Labeled Issue for #5199 · copilot · gpt53codex · 22.1 AIC · ⌖ 0.604 AIC · ⊞ 22.4K · ◷
3. github-actions

   added

   severity-medium

   on Oct 30, 2026