fix: pass hook event name to speckit.git.commit so AI agents read config

[mnriem]

Problem

When speckit.git.commit fires as a hook (e.g., after speckit.tasks), AI agents read the command template's "Behavior" section and treat it as manual instructions — but never actually read .specify/extensions/git/git-config.yml. They assume the default config (all events disabled) and fabricate messages like:

The after_tasks git hook is disabled in configuration, so no auto-commit occurred.

This happens even when the user has set after_tasks.enabled: true and default: true in their config.

The user's workaround in #2702 confirms this — explicitly attaching the config file with @ forces the AI to read it, and then it works.

Root Cause

1. speckit.git.commit.md — The old template had a descriptive "Behavior" section that AI agents interpreted without actually reading the config file
2. format_hook_message() — The hook event name (e.g., after_tasks) was not passed through to the command invocation, so agents had to infer it from context (and often failed)

Changes

• extensions/git/commands/speckit.git.commit.md — Rewrote with explicit step-by-step instructions that mandate reading the config file (Step 2: "You MUST read the file") before deciding whether to commit
• src/specify_cli/extensions.py — Added Hook event: {event_name} to format_hook_message() output and appended event name to EXECUTE_COMMAND_INVOCATION for both optional and mandatory hooks
• Tests — Updated assertions in test_extensions.py and test_integration_claude.py to match the new hook message format

Testing

Full test suite: 2999 passed, 1 skipped, 0 failures

Fixes #2279
Fixes #2702

[Copilot]

Pull request overview

This PR improves extension hook messaging and the git auto-commit command template so AI agents reliably know which hook event triggered /speckit.git.commit and are prompted to read the live git extension config before deciding whether to commit.

Changes:

• Add the hook event name to hook messages and append it to the suggested command invocation so agents don’t have to infer context.
• Rewrite speckit.git.commit’s command template into explicit, step-by-step instructions emphasizing reading .specify/extensions/git/git-config.yml.
• Update Claude + general hook rendering tests to match the new message format.

Show a summary per file

File	Description
tests/test_extensions.py	Updates hook message assertions to expect “Hook event: …” and invocations including the event name.
tests/integrations/test_integration_claude.py	Updates Claude hook rendering assertion for the new event-aware invocation format.
src/specify_cli/extensions.py	Extends format_hook_message() output to include the hook event and append it to the invocation.
extensions/git/commands/speckit.git.commit.md	Replaces the old “Behavior” section with prescriptive instructions to read config and execute commits correctly.

Copilot's findings

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

• Files reviewed: 4/4 changed files
• Comments generated: 3

[Copilot]

Copilot's findings

• Files reviewed: 4/4 changed files
• Comments generated: 2

[Copilot]

Copilot's findings

• Files reviewed: 1/1 changed files
• Comments generated: 2

[Copilot]

Copilot's findings

• Files reviewed: 1/1 changed files
• Comments generated: 2

[Copilot]

Copilot's findings

• Files reviewed: 1/1 changed files
• Comments generated: 1

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[Copilot]

Copilot's findings

• Files reviewed: 1/1 changed files
• Comments generated: 2

[mnriem]

Closing this PR — the change addresses the right problem at the wrong layer.

After review, I don't think this belongs in the default speckit.git.commit template.

The commit decision is already deterministic. auto-commit.sh / auto-commit.ps1 already read git-config.yml, resolve the event key, fall back to auto_commit.default, choose the custom message: or the default format, and skip cleanly when git is unavailable / not a repo / there are no changes. The agent never needs to read or parse the YAML — the script is the source of truth.

Customization is already covered. Custom commit messages are a config concern (the message: key), and anyone wanting the agent to do more (e.g. generate a semantic message) can prepend instructions via a preset on speckit.git.commit. That's precisely what presets are for. Baking per-commit config interpretation into the default template makes every user pay tokens and latency on every hook fire for behavior a minority wants and that the config/preset layer already handles.

The original bug was narrower than the fix. #2279/#2702 weren't "config is hard to override" — they were "the agent read the template's descriptive prose, concluded 'hook disabled,' and never ran the script." The remedy for that is to make the template do less (just: determine event → run the script), not to add a four-step config-parsing routine that re-derives logic the script already owns.

Net: the elaborate rewrite increases cost and the surface for the agent to get config interpretation wrong, while the real fix — if anything is needed — is a near-zero trim of the misleading prose. Closing in favor of that lighter approach; if the hallucination reports recur, a small follow-up scoped to just that trim would be the way to go.