cp-skill-revise-iterative

## EXECUTE NOW **Target: $ARGUMENTS** (exactly one note path or filename — if empty, ask which note) ### Step 0: Resolve the file Resolve `$ARGUMENTS` to a full path. If it's a bare filename, search `kb/notes/` for a match. Read the file to confirm it exists and capture the original content. ### Step 1: Initialize ``` i=0 source_file=<resolved path> ``` Copy the original to `${source_file}.0` — this is the baseline for the first revision pass. ### Step 2: Revision loop Run the following loop. **Max iterations: 5** (hard cap to prevent runaway). #### 2a. Run non-interactive revision ```bash claude -p "Revise $(basename ${source_file}).$i for flow, readability, and cohesion. Preserve all semantic content, claims, and structure. You MAY improve the title (# heading) and description frontmatter field if you see a clear improvement — but do not change tags, type, status, or traits. Focus on: sentence-level clarity, paragraph transitions, removing filler, tightening prose." \ --allowedTools "Read,Edit,Write" \ --max-turns 10 ``` Run this from the directory containing the file so relative paths work. #### 2b. Verify the revision happened Read `${source_file}.$i` after the claude call. If the file is unchanged from before the call (content identical to pre-call state), the revision failed — report the failure and bail out. Do not increment or retry. #### 2c. Semantic fidelity check Compare `${source_file}.$i` (revised) against the previous version (`${source_file}....