fix(hooks): make the recall opener survive tool use

[plind-junior]

real-world verification found the last gap: on tool-heavy prompts the model explored the codebase first and dropped the soft "open your reply with" phrasing from its final message — recall ran but was invisible. all three banners now state the opener as a hard output contract ("your final reply MUST open with the exact words … even if you use tools or explore the codebase first").

verified in production before this pr: on a live project, the exact prompt that previously failed now opens with "From vouch memory:", grounds in cited ids, and still does its tool exploration afterwards. tests updated to pin the strengthened phrasing; full gate green (pytest, mypy, ruff).

Summary by CodeRabbit

• Bug Fixes
  • Improved Claude responses when no relevant vouch context is found, including a consistent fallback message.
  • Strengthened response-opening instructions so the required vouch memory notice remains clear even when tools or code exploration are used.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 769ba8d1-f713-4487-b416-8ce265e1f0e2

📥 Commits

Reviewing files that changed from the base of the PR and between 756ff19 and 280b4c7.

📒 Files selected for processing (2)

• src/vouch/hooks.py
• tests/test_hooks.py

---

Walkthrough

build_claude_prompt_hook now injects mandatory exact opener instructions for empty and non-empty retrieved context, including tool-use persistence and an irrelevant-context fallback. Tests update assertions for the stricter memory-banner wording.

Changes

Claude prompt contract updates

Layer / File(s)	Summary
Strengthen injected opener contracts src/vouch/hooks.py	Empty-context prompts require Nothing in vouch on this., while standard prompts require From vouch memory: and define the irrelevant-context fallback.
Update opener assertions tests/test_hooks.py	Tests require the stricter MUST open with the exact words From vouch memory: wording in both scenarios.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related PRs

• vouchdev/vouch#518: Both changes modify build_claude_prompt_hook behavior for empty or absent injected context.

Suggested labels: retrieval

Suggested reviewers: tet-9

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly reflects the main change: strengthening the recall opener so it persists through tool use.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/banner-survives-tool-use

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}