simplify → code-review medium
Call the two skills in this order, in one turn. Do not ask the person anything in between.
Skill(skill: "simplify", args: "$ARGUMENTS") — applies the cleanups to the working tree.
- As soon as simplify finishes,
Skill(skill: "code-review", args: "medium $ARGUMENTS").
The review also covers the code step 1 changed — that is the point of the order
(the review catches defects the cleanup introduced).
Rules
- Never swap the order. Cleanups made after the review are reviewed by no one.
- Always pass
medium to code-review. Left out, it reuses the last level typed.
- Run step 2 even when step 1 changed nothing.
- If step 1 fails or breaks the build, stop and report instead of going on to step 2.
- Commit as each skill and the person's own rules say. This skill sets no commit units of its own.
Final report — the point of this skill
After both steps, always report once, as below. Without it the skill did not run.
Write every text in the report (summary, each reason and resolution, unverified, next) in the language the
person writes to you, not in the language of these instructions.
Write it even when code-review already reported through ReportFindings — then point at each finding
with its short_summary in one line instead of repeating its text.
Where: desk-review's pane
Call mcp__desk-review__review_report with every item:
summary: one sentence — what changed and what the review found.
items: one entry per cleanup and per finding, { stage, status, file, line, reason, resolution }.
stage: "simplify" with status fixed (applied) or skipped (left alone).
stage: "review" with status fixed (a real defect, fixed in this run), found (a real defect, left unfixed)
or dismissed (rejected on verification).
file: the full path, relative to the working directory or absolute. Never just the file name, never shortened with ….
line: the 1-based line the item is about. For a fix, the first changed line after the fix.
reason: always. Why it was fixed (duplication, needless complexity …), why it was skipped
(false positive, out of scope, would change behaviour …), how it breaks (which input, what goes wrong),
or why it was rejected (the evidence from verification).
resolution: always for a review item marked fixed or found. For fixed, what was changed to fix it;
for found, why it was left unfixed (it needs the person's decision, it is out of scope, the fix is a larger
change …). The person reads it to know what is still open.
unverified: tests not run, files not read, doubts left. Empty when there are none.
next: what the person has to decide, as two options with a recommendation. Empty when there is nothing.
The pane draws each item as one line with its reason below; pressing it opens that file in desk-look's diff
pane. So do not paste code into the report, and in your
reply write only the one-line summary — never the items again.
When the tool is missing or refuses
If mcp__desk-review__review_report is not available, or refuses the input twice, write the same report as
Markdown in the session's scratchpad directory (from the system prompt; /tmp/claude-$(id -u)/ without one)
as simplify-review-<YYYYMMDD-HHMMSS>.md, and reply with its full path and the one-line summary.
In that file, put under each item's path:line its reason and, for a real finding, its resolution, then a code block: for fixed the diff hunk of that change
(git diff -U3 -- <file>, language diff), for the others about five lines either side of the line from the
current file. Read the code from the file or the diff; never rewrite it from memory. Do not use tables, and write
"None" for an empty section rather than leaving it out.
1---2name: simplify-review3description: "Runs /simplify on the current changes to apply cleanups, then /code-review medium to look for bugs, and shows one report in desk-review's side pane. Usage: /desk-review:simplify-review [target]. Use when the person asks to simplify and then review, or to clean up and review in one go."4argument-hint: "[target — the current diff when left out]"5disable-model-invocation: true6metadata:7 source: https://skillmd.com/skills/kyongsik-yoon/simplify-review8---910# simplify → code-review medium1112Call the two skills **in this order, in one turn**. Do not ask the person anything in between.13141. `Skill(skill: "simplify", args: "$ARGUMENTS")` — applies the cleanups to the working tree.152. As soon as simplify finishes, `Skill(skill: "code-review", args: "medium $ARGUMENTS")`.16 The review also covers the code step 1 changed — that is the point of the order17 (the review catches defects the cleanup introduced).1819## Rules2021- Never swap the order. Cleanups made after the review are reviewed by no one.22- Always pass `medium` to code-review. Left out, it reuses the last level typed.23- Run step 2 even when step 1 changed nothing.24- If step 1 fails or breaks the build, stop and report instead of going on to step 2.25- Commit as each skill and the person's own rules say. This skill sets no commit units of its own.2627## Final report — the point of this skill2829After both steps, **always** report once, as below. Without it the skill did not run.30Write every text in the report (`summary`, each `reason` and `resolution`, `unverified`, `next`) in the language the31person writes to you, not in the language of these instructions.32Write it even when code-review already reported through `ReportFindings` — then point at each finding33with its `short_summary` in one line instead of repeating its text.3435### Where: desk-review's pane3637Call `mcp__desk-review__review_report` with every item:3839- `summary`: one sentence — what changed and what the review found.40- `items`: one entry per cleanup and per finding, `{ stage, status, file, line, reason, resolution }`.41 - `stage: "simplify"` with `status` `fixed` (applied) or `skipped` (left alone).42 - `stage: "review"` with `status` `fixed` (a real defect, fixed in this run), `found` (a real defect, left unfixed)43 or `dismissed` (rejected on verification).44 - `file`: the full path, relative to the working directory or absolute. Never just the file name, never shortened with `…`.45 - `line`: the 1-based line the item is about. For a fix, the first changed line after the fix.46 - `reason`: **always**. Why it was fixed (duplication, needless complexity …), why it was skipped47 (false positive, out of scope, would change behaviour …), how it breaks (which input, what goes wrong),48 or why it was rejected (the evidence from verification).49 - `resolution`: **always** for a review item marked `fixed` or `found`. For `fixed`, what was changed to fix it;50 for `found`, why it was left unfixed (it needs the person's decision, it is out of scope, the fix is a larger51 change …). The person reads it to know what is still open.52- `unverified`: tests not run, files not read, doubts left. Empty when there are none.53- `next`: what the person has to decide, as two options with a recommendation. Empty when there is nothing.5455The pane draws each item as one line with its reason below; pressing it opens that file in desk-look's diff56pane. So **do not paste code into the report**, and in your57reply write only the one-line summary — never the items again.5859### When the tool is missing or refuses6061If `mcp__desk-review__review_report` is not available, or refuses the input twice, write the same report as62Markdown in the session's scratchpad directory (from the system prompt; `/tmp/claude-$(id -u)/` without one)63as `simplify-review-<YYYYMMDD-HHMMSS>.md`, and reply with its full path and the one-line summary.64In that file, put under each item's `path:line` its reason and, for a real finding, its resolution, then a code block: for `fixed` the diff hunk of that change65(`git diff -U3 -- <file>`, language `diff`), for the others about five lines either side of the line from the66current file. Read the code from the file or the diff; never rewrite it from memory. Do not use tables, and write67"None" for an empty section rather than leaving it out.