Mariomarquezt/multi-flow-skills / skills/sliced-review
sliced-review
FlaggedReview a whole codebase with several models by cutting it into small file-listed slices ordered by blast radius, then triage every finding with three judges, and optionally fix. Use when the user asks to review or audit a whole project or repo with other models, wants a second opinion on a codebase, wants a review split into slices or tiers, has a whole-repo review that died on credits, or wants a check before a launch or handover. Modes (`3`, `5`, `8`, `fast`, `fix …`, `fix fast …`, `fix only`, `report`, `no wrap`) are in the body. Not for a plan before code exists (/grill-me-codex, /codex-review) or a single diff, branch or PR (/code-review).
CHANGELOG.md
View on GitHub# Changelog
The version lives in `SKILL.md` frontmatter (`version:`) and as a git tag. Compare yours with
`grep ^version ~/.claude/skills/sliced-review/SKILL.md` to see if you are behind.
## Repo move — 2026-11-19
This skill now lives in the `multi-flow-skills` bundle (renamed from `sliced-review`; the old URL forwards). Models moved to `skills/_shared/models.md`; Sol is GPT-6.2 Sol and Astra 6.2 is an extra lane. The version in `SKILL.md` is the current one.
## 2.3.1 — 2026-09-04
- Fixed five-model cross-check ring: Sol → Opus → Grok → GLM Flash → Composer → Sol. Grok and
Composer never check each other — both bill to Cursor, and a fix and its check must be on
different subscriptions. Same rule stated for the `fix fast` modes and in `runners.md`.
## 2.3.0 — 2026-09-04
- **`fix only` mode** (§11i): fix a review that already ran without re-running it. Skips
Steps 1–9 and 11a, takes the `open` findings from `findings.json`, re-verifies each against the
current tree (closing what a later commit already fixed), then runs the 11g pipeline — split by
file, fix, rotated cross-check, settled-tree pass, one commit. Variants `fix only 5`,
`fix only high`; composes with `report` and `no wrap`. Scorecard is not rebuilt.
## 2.2.0 — 2026-09-04
- **Review report, Markdown + HTML.** `templates/build-report.py` (stdlib only) renders
`docs/reviews/report-<date>.md` and a self-contained `report-<date>.html` from `findings.json`,
the scorecard and each lane's `ledger.tsv`: counts, the ranked scorecard, a lanes table
(total time, avg per slice, slowest slice, slice status per model) and every finding grouped by
severity with file, status, finder and verified flag.
- **`report` suffix** (`/sliced-review 5 report`, `fix fast 3 report no wrap`) builds it without
asking. Without the suffix the skill asks once at the end of Step 12 (new §12b).
## 2.1.1 — 2026-09-04
- Scorecard: time is explicit per model — total time, average per slice, slowest slice (with its
id), all from `ledger.tsv`. Time is shown beside the score but never changes the rank.
## 2.1.0 — 2026-09-04
- **Model scorecard.** Every multi-model run now ends Step 10 by writing
`docs/reviews/scorecard-<date>.md` (template in `templates/scorecard.md`): one row per model
with reported / verified / refuted counts, precision, verified findings by severity
(critical · high · medium · low), unique finds, missed critical/high, a weighted score
(8 · 4 · 2 · 1) and the ledger's wall clock and cost. Models are ranked by score; ties break by
unique finds, then precision. Grades use findings that survived triage at the triager's
severity, never raw claims or the model's own labels.
- New honesty rule: never rank models on raw counts or self-reported severity.
## 2.0.1 — 2026-09-04
(2.0.0 was published for a few minutes the same day and withdrawn; 2.0.1 is the release.)
Rebuilt around a measured five-model reviewer stack. Seven models ran the same five slices of a
~20k-line codebase head to head; the roster and effort levels below are what that produced.
- **Invocation modes** (new section at the top of `SKILL.md`, parsed before anything else):
bare `/sliced-review` and `3` = core three (GPT-5.6 Sol xhigh, Grok 4.6 high, Opus 5 high);
`5` adds GLM 5.3 Flash and Composer 2.5; `fast` = Composer alone.
- **`fix` modes** (§11g): `fix`, `fix 3`, `fix 5`, `fix high` also triage, apply fixes with the
reviewing models, cross-check every fix with a DIFFERENT model, and commit.
- **`fix fast` modes** (§11h): same, pipelined — each slice's fixes start as soon as that slice's
review lands, with a cap on concurrent Cursor consumers.
- **`no wrap` suffix** skips the closing wrap-up step.
- **Effort is pinned, never inherited.** Every runner takes an explicit `--effort`
(`low|medium|high|xhigh|max`). Grok stays at `high`: xhigh returned half the findings and zero
HIGHs.
- **Route pinning** in `references/runners.md`: one correct route per model; the Cursor wrapper
hard-refuses anything but Grok and Composer. Codex runs require `-o/--output-last-message`.
- **Step 12 — record the findings.** Every review ends by writing `docs/reviews/findings.json`
(schema in `templates/findings.json`) so findings survive past the `out/` folder.
- **FIELD-NOTES**: four-reviewer run on 42k lines; the "never ran on Cursor" claim retracted.
## 1.0.0 — 2026-08-22
Initial public release: blast-radius-ordered slices, cost ledger, activity-aware watchdog with a
hard wall-clock cap, quota guard, `chain.sh` queue runner, triple review with cross-reviewed
fixes and capped adversarial rounds.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |