Skip to content

Commit e9ca9d9

Browse files
remigiusz-antczakRemigiusz Antczakclaude
authored
feat(skill): 1.108.0 — wave 29 review-calibration anti-slop lens (#268) + fix SKILL.md frontmatter version drift (#272) (#292)
#268 — method.md anti-slop: a settled style/preference nit the team has already declined (kept in a committed, path-scoped review-calibration record, in git, not a hosted memory service) is dropped so the review doesn't re-raise what the owner dismissed. The record NEVER suppresses a security, logic, null-deref, or data-validation/data-loss finding -- calibration silences preference noise, not correctness or safety. No provenance risk (no standards named). #272 (fix) — the three lockstep skills' SKILL.md metadata.version frontmatter had lagged the authoritative VERSION file at 1.100.0 for seven releases; bumped to match (1.108.0). The frontmatter version now moves with the lockstep bump going forward. Lockstep version bump (deep-code-review, agentic-delivery, idea-critic, plugin, AND the three SKILL.md frontmatter) to 1.108.0; one new eval (dcr 113); SHA256SUMS regenerated. Tests: 67/67 CI gates pass; evals.json valid; routing clean; install.sh syntax OK; privacy + provenance self-scan clean; frontmatter version == VERSION on all 3 lockstep skills. Co-authored-by: Remigiusz Antczak <remigiusz@antczak.io> Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
1 parent 2e62d3e commit e9ca9d9

11 files changed

Lines changed: 52 additions & 15 deletions

File tree

‎.claude-plugin/plugin.json‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -2,7 +2,7 @@
22
"name": "deep-code-review",
33
"description": "Perun — bring the thunder to your codebase. A portable, evidence-grounded quality bar for building with AI agents: the flagship deep-code-review skill (default install is review-only), plus opt-in overlays — gated delivery, an idea critic, a suite conductor, and stage-aware product, growth, positioning, business-ops, and output-safety specialists.",
44
"skills": "./.claude/skills",
5-
"version": "1.107.0",
5+
"version": "1.108.0",
66
"author": {
77
"name": "deep-code-review contributors"
88
},

‎.claude/skills/agentic-delivery/SKILL.md‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -10,7 +10,7 @@ description: >-
1010
license: MIT
1111
metadata:
1212
author: deep-code-review contributors
13-
version: "1.100.0"
13+
version: "1.108.0"
1414
---
1515

1616
# Agentic delivery
Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1 +1 @@
1-
1.107.0
1+
1.108.0

‎.claude/skills/deep-code-review/SKILL.md‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -11,7 +11,7 @@ description: >-
1111
license: MIT
1212
metadata:
1313
author: deep-code-review contributors
14-
version: "1.100.0"
14+
version: "1.108.0"
1515
---
1616

1717
# Deep Code Review
Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1 +1 @@
1-
1.107.0
1+
1.108.0

‎.claude/skills/deep-code-review/evals/evals.json‎

Lines changed: 11 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1194,6 +1194,17 @@
11941194
"Reruns at most once and only after diagnosis; a reproduced hang/timeout is the bug to file, not to rerun",
11951195
"Cancels a hang with evidence (last log line, elapsed) so it isn't mistaken for a flake"
11961196
]
1197+
},
1198+
{
1199+
"id": "review-calibration-suppresses-settled-nits-never-security-or-logic",
1200+
"prompt": "Our team keeps a committed .review-calibration file of nits we've decided we don't care about (e.g. 'we allow console.log in scripts/'). This PR re-introduces a console.log in scripts/ AND has an unscoped SQL query that could leak another tenant's rows. The calibration file says 'ignore console.log in scripts/'. What does the reviewer post?",
1201+
"expected_output": "Suppress the console.log nit, post the SQL finding. The review-calibration record legitimately silences a SETTLED STYLE/PREFERENCE nit the team has already declined -- the console.log in scripts/ is exactly that, so re-raising it is noise (anti-slop: a finding that changes no owner action). But the record NEVER suppresses a security, logic, null-deref, or data-validation/data-loss finding: the unscoped SQL query is a cross-tenant data leak (a security + data finding), so it is raised at full severity regardless of any calibration entry -- a calibration file that silenced it would be a dangerous misuse. Calibration is a committed, path-scoped record of style/preference decisions read to cut repeat noise, not a hosted memory service and not a mechanism to dismiss correctness or safety. (method.md.)",
1202+
"expectations": [
1203+
"Suppresses the settled style nit (console.log) per the committed calibration record -- cuts repeat noise",
1204+
"Posts the unscoped-SQL cross-tenant leak at full severity -- calibration NEVER suppresses security/logic/data findings",
1205+
"Frames calibration as a committed path-scoped style/preference record, not a hosted memory service and not a way to dismiss correctness/safety",
1206+
"Does not let a calibration entry override a security/data finding"
1207+
]
11971208
}
11981209
]
11991210
}

‎.claude/skills/deep-code-review/references/method.md‎

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -463,6 +463,12 @@ owner action is not a finding. Drop or demote to Nit/Info:
463463
- restyling, renaming, or "consider maybe" with no defect;
464464
- a second copy of a fact the project's own gate already enforces and the
465465
review already confirmed green;
466+
- a **settled nit** the team has already seen and declined — kept in a committed, path-scoped
467+
**review-calibration** record (accepted / rejected findings for this repo, in git, not a
468+
hosted memory service) that the reviewer reads to avoid re-raising what the owner already
469+
dismissed. **The record NEVER suppresses a security, logic, null-deref, or
470+
data-validation/data-loss finding** — those are re-raised every time regardless of prior
471+
dismissals; calibration silences *style/preference* noise, not correctness or safety;
466472
- a security finding in a **chronically-noisy class** — DoS, rate-limiting,
467473
resource-exhaustion, generic input-validation with no reached sink, open-redirect — with
468474
**no demonstrated impact path**: demoted to a non-blocking `unverified`/Nit **lead** —

‎.claude/skills/idea-critic/SKILL.md‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -9,7 +9,7 @@ description: >-
99
license: MIT
1010
metadata:
1111
author: deep-code-review contributors
12-
version: "1.100.0"
12+
version: "1.108.0"
1313
---
1414

1515
# Idea critic

‎.claude/skills/idea-critic/VERSION‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1 +1 @@
1-
1.107.0
1+
1.108.0

‎CHANGELOG.md‎

Lines changed: 20 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -3,6 +3,26 @@
33
All notable changes to this repository are documented here. Format loosely
44
follows Keep a Changelog; versioning follows Semantic Versioning.
55

6+
## [1.108.0] — 2026-09-17
7+
8+
Wave 29 of the dogfooding batch: #268 review-calibration (an anti-slop lens) + #272, the
9+
SKILL.md frontmatter version drift (a real housekeeping bug — the three lockstep skills'
10+
frontmatter had been stuck at 1.100.0 since it stopped moving with the bump).
11+
12+
### Added — deep-code-review
13+
- **`method.md` anti-slop — a review-calibration record suppresses settled nits, never
14+
security/logic** (#268). A settled style/preference nit the team has already declined — kept
15+
in a committed, path-scoped review-calibration record (in git, not a hosted memory service) —
16+
is dropped so the review doesn't re-raise what the owner already dismissed. The record
17+
**never** suppresses a security, logic, null-deref, or data-validation/data-loss finding;
18+
calibration silences preference noise, not correctness or safety. One new eval (113 total).
19+
20+
### Fixed
21+
- **SKILL.md frontmatter version drift** (#272). The three lockstep skills' `metadata.version`
22+
frontmatter had lagged the authoritative `VERSION` file since 1.100.0; bumped to match
23+
(1.108.0) in all three. A CI gate that fails on `metadata.version` ≠ `VERSION`, so the drift
24+
can't recur silently, is filed as #290.
25+
626
## [1.107.0] — 2026-09-17
727

828
Wave 28 of the dogfooding batch: three owner-filed CI-diagnosis hygiene lessons (#285, #286,

0 commit comments

Comments
 (0)