fix(cline): stop unrelated prose from suppressing the hook command note

[jawwad-ali]

Problem

ClineIntegration._inject_hook_command_note guards idempotency with a whole-document substring scan:

if "replace dots" in content:
    return content

The shared SkillsIntegration._inject_hook_command_note helper (base.py:1637-1642) instead compares only the line immediately above each match:

previous_lines = content[:m.start()].splitlines()
if previous_lines and previous_lines[-1] == indent + note:
    return m.group(0)

Cline never picked up that change.

Reproduction on current main

A. unrelated prose -> note injected? False      (base: True)
B. two hook sections, one already noted -> notes: 1   (want 2)
C. indented instruction after a blank line:
   '## Hooks\n\n  <note>\n\n  - For each executable hook, ...'   <-- spurious blank line

(a) Any command or extension markdown whose prose happens to contain the phrase "replace dots" — e.g. "When normalizing table names, replace dots with underscores." — loses the note entirely. The generated workflow then tells the agent to emit /speckit.git.commit, a dotted command Cline never registers (Cline installs speckit-git-commit.md). The note exists precisely to prevent that.

(b) A document with one already-noted hook section never gets a note on a second, un-noted one — exactly the case the base helper was changed to handle.

(c) Secondary: cline's regex captures indentation with ^(\s*) where the base uses ^([ \t]*). Because \s matches newlines, the captured "indent" can swallow a preceding blank line, which is then re-emitted between the note and the instruction. The per-instruction check cannot line up until this is fixed, so both belong in one change.

Fix

Adopt the base helper's per-instruction check and its [ \t]* capture. After the fix:

A. unrelated prose -> note injected? True
B. two sections -> notes: 2 (want 2)
C. indented+blank -> note sits directly above the instruction, indentation preserved
D. idempotent -> True

Verification

• Fail-before / pass-after: 3 new-vs-baseline failures with the source reverted → 35 passed with the fix.
• Idempotency is asserted explicitly (re-running on already-noted content is a no-op) — the property the old guard was there to provide is kept.
• Full tests/integrations scope: 18 failed, 2627 passed — identical to the clean-main baseline's 18 (all Windows symlink-privilege), plus my 3 new tests.
• uvx ruff@0.15.0 check src tests → clean

No breaking change. Content that already has the note above every instruction is returned byte-identical; only documents that were previously missing a note now gain one.

---

Written with assistance from Claude Code. Bug found, reproduced, and verified by me on current main.