dino-requesting-code-review
/dino-requesting-code-review
Zestawia dane o właścicielach i ryzyku plików do prośby o przegląd, zanim powstanie opis pull requesta dla recenzentów.
- Autor
- artk0de ↗
- Repozytorium
- tearags-mcp ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 10.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 1
Brak sygnałów. Repozytorium zmienione 10.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Zestawia dane o właścicielach i ryzyku plików do prośby o przegląd, zanim powstanie opis pull requesta dla recenzentów.
Wyzwalacze z opisu
Oryginalny opis (EN)
Prepare code-review request bundle: per-file ownership, churn, connected taskIds from tea-rags git signals before composing review request — reviewers know who to ping, where risk lives. Triggers on "request review", "open a PR", "ready for review", "готовлю PR", "запрос на ревью", "code review please". NOT for self-review or draft-PR exploration. Wraps /requesting-code-review with tea-rags reviewer-context bundle.
Jak działa
- Iron Rule
- Step 1 — Collect diff file list
- Step 2 — Git-bundle query
- Step 2.5 — Reviewer-hub awareness (codegraph only)
- Step 2.6 — Naming pre-check (codegraph only)
- Step 3 — Build reviewer-context bundle
- Step 3a — Tests at risk (scenarios under threat)
- Step 4 — Invoke /requesting-code-review
- Red Flags — STOP, restart from Step 2
Podgląd SKILL.md
---
name: dino-requesting-code-review
description:
'Prepare code-review request bundle: per-file ownership, churn, connected
taskIds from tea-rags git signals before composing review request — reviewers
know who to ping, where risk lives. Triggers on "request review", "open a PR",
"ready for review", "готовлю PR", "запрос на ревью", "code review please". NOT
for self-review or draft-PR exploration. Wraps /requesting-code-review with
tea-rags reviewer-context bundle.'
---
# dinopowers: requesting-code-review
> **Resources in Cursor:** if a `tea-rags://` resource cannot be read (no
> resource tool, or the read fails), run
> `tea-rags resource "<uri>" --path "$PWD"` in the terminal — same content.
Wrapper over `/requesting-code-review`. Request reaches reviewer with
git-context bundle — file owners, churn, connected tickets — reviewers
pair-match by expertise, trace coordinated changes without re-excavating.
## Iron Rule
**tea-rags git-bundle query MUST run on `git diff --name-only` BEFORE composing
review request** — whenever ≥1 file changed.
Value = right tool (`semantic_search`) + impact rerank (`blastRadius` / custom
fallback `imports: 0.5, churn: 0.3, ownership: 0.2`) + params (brace-expanded
`pathPattern` over diff files, `metaOnly: true`) + bundle format (per-file
ownership/churn/taskIds, NOT blast-radius verdict).
Diff empty → skip wrapper, invoke `/requesting-code-review` directly. Never
fabricate.
**Chaining rule:** [CHAINING.md](../../CHAINING.md) — every onward `/Y` →
`/dino-Y`. NEVER bypass wrapper.
**Requires superpowers:** `/requesting-code-review` — availability per
[CHAINING.md](../../CHAINING.md) "Missing superpowers" steps 1–3. Not in the
skills list ≠ unavailable (the list is truncated): run
`ls $CODEX_HOME/plugins/cache/*/superpowers/*/skills/requesting-code-review/SKILL.md`
as written first — a hand-built path failing proves nothing; only that glob
matching nothing → stop and ask the user to install superpowers
(https://github.com/obra/superpowers). NEVER present this enrichment alone as
the full workflow.
**Freshness:** [FRESHNESS.md](../../FRESHNESS.md) (owner
`tea-rags://guide/freshness`). Lag ≠ unusable — never Bash `cat`/`grep`/`Read`
over code; `treeState` row → `find_symbol`.
## Step 1 — Collect diff file list
`git diff --name-only <base>...HEAD` (branch) or `git diff --staged`
(pre-commit):
| Source | Example |
| --------------------- | ------------------------------------- |
| Branch against base | `git diff main...HEAD --name-only` |
| Uncommitted + staged | `git diff --name-only HEAD` |
| Specific commit range | `git diff <sha1>..<sha2> --name-only` |
Output: `diffFiles` (M/A, exclude pure D) + `intent` (one sentence, what diff
does). Pure deletions only / empty diff → skip wrapper (reviewers need
code-in-diff).
## Step 2 — Git-bundle query
ONE `semantic_search` — same idiom as `/dino-writing-plans`:
```
project: <alias> ← `tea-rags://guide/freshness` "Addressing the Codebase"
path: <your working directory> ← the tree (worktree / branch checkout)
query: <intent from Step 1>
pathPattern: "{diffFile1,diffFile2,...}" ← brace expansion
rerank: "blastRadius" ← codegraph on; else { custom: { imports: 0.5, churn: 0.3, ownership: 0.2 } }
limit: <diffFiles.length * 3>
metaOnly: true
fields: ["relativePath", "imports", "git"] ← metaOnly alone drops raw git.* (taskIds, blame*)
```
Do NOT substitute:
| Wrong tool | Why wrong |
| --------------------------------------------------- | ---------------------------------------------------------------------------------------------------- |
| `hybrid_search` | Same rerank, but its BM25 leg re-ranks by query-token overlap — signal scan wants semantic + signals |
| Other named preset (`"codeReview"` / `"ownership"`) | `codeReview` misses `imports`; `ownership` misses churn + taskIds in one call |
| `git blame` / `git log --format` per-file | Manual git commands are slower and miss the indexed overlay |
| `find_similar` | Finds code analogs, not diff metadata |
Do NOT pass:
- `metaOnly: false` — bundle = signals; content already in PR diff
- different weights — must match project idiom
- `filter` narrowing — pathPattern scopes already
Empty results (files too new) → skip bundle, invoke `/requesting-code-review`
with note "diff files not yet indexed — no git-context available". Never
fabricate.
## Step 2.5 — Reviewer-hub awareness (codegraph only)
Codegraph on = graph tools (`get_callers`, `trace_path`, …) in tool list; unsure
→ read `tea-rags://project/<alias>/metrics?section=registry`. On → per changed
symbol `get_callers symbolId=<id>` (exact id via `find_symbol` first). Callers
depend on change → their `blameDominantAuthor`s = stakeholders. Add "Affected
callers / suggested reviewers" line to bundle.
Off → skip; Step 2/3 ownership bundle stands; note caller-impact not computed.
NEVER invent caller lists.
## Step 2.6 — Naming pre-check (codegraph only)
`review_changes(changes={ base: <Step 1 base> }, sections: ["naming"])` — ONE
call reviews every name diff's added lines declare (values, types, constants);
never hand-list into `names[]`. Base = target branch for branch diff (`main`);
read at merge-base with HEAD (target moving on adds nothing). Uncommitted work →
omit (default `HEAD`). Worktree branch → pass `path=<worktree>` (alias alone
reads main checkout).
Findings already quoted in the task / context = what this call returns — still
list the `review_changes` call in your sequence, then act on them.
Settle naming before reviewer spends comment:
- `MISFIT` → rename to `suggestion` before sending (or list under risk flags
with holder). Type MISFIT with `role.evidence: "directory"` → rename only when
type belongs to `role.examples`' family, else keep name
- `COLLISION` → rename, or justify against `existing` declaration
- `NEW_TERM` → one-line justification in bundle, or take offered alternative
- finding with `genericName` → rename or justify
Answer under `review` {workTree, base, mergeBase, changedFiles, checked,
conforming, novel, findings, notJudged, truncated?}; findings flat
`{relPath, line, name, kind, type?, verdict, …}`. Your answer prints this line,
filled from the findings you hold (plan-only too), every finding with
`relPath:line` AND the action taken on it (pick one side of each `|`), both
counts always. Echoing a finding with no action = MISFIT left for the reviewer:
`Naming: MISFIT <name> (<relPath:line>) → renamed to <suggestion> before sending | kept, risk flag for <holder>; COLLISION <name> (<relPath:line>) vs <existing> → renamed | justified: <why>; NEW_TERM <name> (<relPath:line>) — <justification>; conforming <N>, novel <M>`
Path or count not in the prompt → write `<relPath:line from review>` /
`novel <from review>`, never drop the slot (`novel` = nothing to compare, no
action). Reading: `/data-driven-generation` Step 5 "Naming (lexicon)".
## Step 3 — Build reviewer-context bundle
Aggregate by `relativePath`. Per file — raw values (`blame*`, `recent*`,
`commitCount`, `ageDays`, `taskIds`, `bugFixRate`) from payload `git.file.*`
(present only because Step 2 names `"git"` in `fields`; a field the index lacks
is absent — never fabricate); labels from `rankingOverlay.file.*`:
- `blameDominantAuthor` + `blameDominantAuthorPct` — live-line owner (approves
on current code)
- `recentDominantAuthor` + `recentDominantAuthorPct` — recent committer
(mentally loaded, fastest turnaround)
- `blameContributorCount` (live owners), `recentContributorCount` (recent
committers)
- `commitCount` + `ageDays`
- `taskIds` (connected tickets — highest-value signal, never omit)
- `bugFixRate` (risk signal)
Reviewers by `blame*`; fast turnaround by `recent*`.
Bundle goes INTO request, not a verdict. Metadata only — no raw diff, no raw git
log. **Four sections, ALWAYS all four, named as below** (no data for one → write
"none", never drop it): (1) per-file table with ALL five columns — File, Owner,
Contributors, Commits (age), Related tickets (taskIds; none → "—", never drop
the column), (2) suggested reviewers, (3) coordinated change context (taskIds),
(4) risk flags for reviewer attention (bugFixRate / churn / silo per file):
```
### Reviewer context bundle
**Files changed (per-file owners + history):**
| File | Owner | Contributors | Commits (age) | Related tickets |
|---|---|---|---|---|
| src/a.ts | Alice (92%) | 1 | 23 (142d) | #123, #145 |
| src/b.ts | shared (42%) | 5 | 8 (30d) | — |
**Suggested reviewers by expertise:**
- Alice — primary owner of `src/a.ts` (92% dominance)
- Bob — deep contributor to `src/b.ts` (last 3 commits)
**Coordinated change context:**
- Related ticket #123 also touched src/c.ts, src/d.ts in previous commits — reviewer should check consistency
**Risk flags for reviewer attention:**
- `src/a.ts` has bugFixRate=35% (high) — scrutinize new logic
- `src/e.ts` has been touched 47 times in 90 days — stability concern
- `<name>` (<relPath:line>) kept despite naming MISFIT → `<suggestion>` — <holder> decides (only when not renamed)
**Naming (Step 2.6, codegraph on):** <the Step 2.6 `Naming:` line, verbatim>
```
Bundle > 20 lines → per-file table top 10 by `imports`, note "N more files
omitted". Thin data → empty bundle beats fabricated ownership.
## Step 3a — Tests at risk (scenarios under threat)
Invoke `/tests-as-context` with:
```
recipe: "tests-at-risk"
affectedFiles: <diffFiles from Step 1>
intent: <intent from Step 1>
```
Recipe → ranked DSL leaf test chunks bound to change. Add to bundle one of:
- SKIP (no DSL test chunks indexed):
```
**Scenarios under risk:** unavailable (no DSL test chunks indexed)
```
- Non-empty list:
```
**Scenarios under risk:**
- <file>:<line> — <describe-it path>
- <file>:<line> — <describe-it path>
```
Cap top 8; "N more omitted" if truncated.
- Empty (preflight passed, no semantic match):
```
**Scenarios under risk:** no obvious test bindings found — reviewer should
verify whether new behavior needs new tests
```
NEVER name test runners — generic phrasing; command left to reviewer / CI /
pre-commit hook.
## Step 4 — Invoke /requesting-code-review
Invoke `Skill` tool with `/requesting-code-review`. Prepend bundle as context.
Phrase handoff as:
> "Include this reviewer-context bundle in the review request: …<block>…
> Reviewers can pair-match by ownership and see coordinated-change context
> without re-excavating history.
>
> Chaining rule reminder: when your cycle would next invoke
> `/verification-before-completion` (or any wrapped `/Y`), invoke `/dino-Y`
> instead — see the Chaining rule section above."
`/requesting-code-review` runs standard cycle — wrapper enriches, never
replaces.
## Red Flags — STOP, restart from Step 2
- "Reviewer knows codebase, no bundle" → bundle = PAIR-MATCHING, not education
- "Diff small, skip bundle" → small diff to high-churn file still needs
ownership context
- git blame/log substituted → redo with semantic_search + Step 2 rerank
- Wrong rerank for codegraph state → redo: `"blastRadius"` when codegraph on
(graph tools present), else `{imports: 0.5, churn: 0.3, ownership: 0.2}`
(tea-rags analytics-rerank). `"ownership"` alone misses imports/churn/taskIds;
`"codeReview"` finds reviewable code, not reviewer context
- Request composed before Step 2 → revert, restart
- Raw diff / full git log in bundle → METADATA only
- `D`-status files in bundle → drop; show in diff
- Inner cycle chains raw `/verification-before-completion` → intercept, invoke
`/dino-verification-before-completion`
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · artk0de/tearags-mcp | 100 | MIT |
| artk0de/tearags-mcp | 100 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki cf0c1e3210a0 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100requesting-code-review
Przygotowuje pakiet przeglądu z właścicielami plików, churnem i zadaniami z sygnałów git, zanim powstanie prośba o review. - 100mr-review
Ocenia ryzyka w merge requeście lub gałęzi na podstawie sygnałów z tea-rags i pisze uwagi z konkretną poprawką. - 100loom-requesting-code-review
Przygotowuje prośbę o recenzję kodu z podsumowaniem zmian, wynikami samodzielnych testów i obszarami uwagi dla recenzentów. - 100requesting-code-review
Pomaga poprosić o review lub utworzyć pull request: najpierw self-review, właściwy kontekst i plan testów, także gdy ktoś pisze tylko „open a PR”. - 100requesting-code-review
Przeprowadza przegląd gotowości do produkcji zmiany w MR względem zatwierdzonych wymagań. - 96requesting-code-review
Prosi o przegląd kodu, gdy zadania gałęzi są skończone, albo przed scaleniem.