fix: doctor false-positive on managed-block hooks, setup output polish

[liam-ai-reality]

Summary

• doctor false-positive 'schema drift' on codex hook with managed-block append #102 — doctor false-positive on Codex hook with user content above block: check_hook previously compared the full on-disk hook file byte-for-byte against render_hook_script, which always returns a fresh-install layout (shebang + blank + managed block). When a user already had content in .git/hooks/pre-commit, install correctly appended the managed block after that content — but doctor then flagged schema drift because the full files don't match. Fix: extract only the managed block portion from both actual and expected (by scanning for MANAGED_START/MANAGED_END markers) and compare those. Byte-equality is still tried first for surfaces without managed blocks (Claude Code).
• klasp setup --dry-run banner appears on line 2 instead of line 1 #106 — dry-run banner appeared after the gate-count line: moved --dry-run mode println before the klasp setup — detected N gate(s) line so users immediately know no I/O will occur.
• klasp setup prints stale "Next:" instructions for steps it already ran #105 — stale "Next: klasp init" footer in setup output: render_plan was called in setup, which appends a "Next:" footer suggesting klasp init — but setup already ran init. Switched to render_plan_no_next (added in this workspace) for the setup code path.

Test plan

• cargo test --package klasp -- doctor_codex_hook_with_user_content_above_block_exits_0 — previously failed, now passes
• cargo test --package klasp — 501 passed, 0 failed
• cargo test --package klasp --test setup — 12 passed
• cargo test --package klasp --test doctor — all doctor tests pass

🤖 Generated with Claude Code

[liam-ai-reality]

Code review

No issues found. Checked for bugs and CLAUDE.md compliance.

[liam-ai-reality]

Code review

Summary: Three targeted bug fixes with regression tests. No new abstractions beyond what the fixes require.

Area	Finding
doctor.rs check_hook (#102)	Managed-block extraction logic is correct — handles out-of-order markers, missing trailing newline, and Claude Code surfaces (no markers → byte-equality only).
render.rs render_plan_no_next (#105)	Bool-flag refactor is idiomatic and preserves all existing callers unchanged.
setup.rs output order (#106)	Swap is correct.
Tests	Both regression tests are well-scoped and exercise the exact failure shapes the bugs produced.

Critical: none

Suggestions: none

Verdict: 🟢 all-green

[liam-ai-reality]

Review remediation

Step 10 (triage-followups) skipped — code-review returned 🟢 all-green with zero validated findings. No deferred issues. No fixes required.