@caddis/cli 0.4.9 → 0.4.10

package/bundles/antigravity-plugin/agents/code-reviewer.md

@@ -31,7 +31,7 @@ You are a principal engineer doing a focused code review. You read the change an
 ## Return format (always end with this)
 ```
 review:
-  verdict: approved | changes-requested
+  verdict: CLEAN | BLOCKING
   blocking:
     - file: <path:line>   issue: <what>   fix: <concrete suggestion>
   should_fix:
@@ -40,7 +40,7 @@ review:
     - file: <path:line>   note: <what>
   good: <one line on what's done well, if anything>
 ```
-`verdict: approved` requires an empty `blocking` list. Be specific with `file:line`. Do not fix; report.
+`verdict: CLEAN` requires an empty `blocking` list. Be specific with `file:line`. Do not fix; report.
 
 **Then, as the very last line of your output (nothing after it), emit the machine verdict:**
 `REVIEW: CLEAN` (empty `blocking` list) or `REVIEW: BLOCKING` (one or more blocking items). Automated

package/bundles/antigravity-plugin/plugin.json

@@ -1,5 +1,5 @@
 {
   "name": "caddis",
-  "version": "1.3.96",
+  "version": "1.3.97",
   "description": "caddis — agent-agnostic dev harness as an Antigravity (agy) plugin: TDD skills + commands, single-sourced with the Claude Code plugin."
 }

package/bundles/antigravity-plugin/scripts/enforcement_inventory.py

@@ -10,7 +10,7 @@ model change, context pressure, a cheaper lane, or a distracted session — and
 guarantees, which is worse than reading as advice.
 
 The things that actually held under pressure were the ones with exit codes: pytest,
-validate_pool, the hooks, oss_review's exit 2.
+validate_pool, the hooks, oss_review's exit 4.
 
 So this script measures the ratio of rules that a machine can fail on, to rules that only a
 model can choose to honour. That number is the health metric for the plugin. It is expected to

package/bundles/antigravity-plugin/scripts/oss_review.py

@@ -53,10 +53,18 @@ import urllib.error
 import urllib.request
 from pathlib import Path
 
+# ONE SCALE ACROSS caddis (see claude-harness/README.md):
+#   0 clean · 1 blocked · 2 advisory note · 3 could not run · 4 malformed
+# ERROR moved 2 -> 4 on 2026-09-09. It sat on 2, which caddis_gate uses for "proceed, this is
+# worth saying" — the exact opposite, and in the dangerous direction: a caller treating 2 as
+# advisory silently accepts a review that never ran. That is the 2026-08-10 shape, where a stale
+# copy of this file returned CLEAN on a database write path nobody had reviewed.
+# 2 is left FREE here. This tool has no advisory verdict; a slot that means one thing everywhere
+# is worth more than a slot reused for something else.
 EXIT_CLEAN = 0
 EXIT_BLOCKING = 1
-EXIT_ERROR = 2
-EXIT_CONFIG = 3
+EXIT_CONFIG = 3     # could not run: no key, no provider, bad configuration
+EXIT_ERROR = 4      # ran, produced nothing usable: no verdict line, transport failure, unsplittable file
 
 # Provider presets — the SINGLE place a renamed model id or moved endpoint is edited. Adding a new
 # provider (Qwen, a local vLLM, …) is one new row. Callers can always bypass this via env/flags.

package/bundles/antigravity-plugin/skills/cross-review/SKILL.md

@@ -81,7 +81,7 @@ writes. This line used to prefix the flag itself, which turned that into a doubl
 Optional: `--base-url <url>`, `--model <id>` (env always overrides the preset).
 
 **A large diff is batched, not refused.** Over `REVIEW_MAX_DIFF_CHARS` (default 60,000) the tool splits
-on whole-file boundaries and reviews each batch, and the verdict is CLEAN only if every batch is. Exit 2
+on whole-file boundaries and reviews each batch, and the verdict is CLEAN only if every batch is. Exit **4**
 now means the narrower case — a **single file** too big to split, or more batches than the cap. When you
 see it, the fix is usually to review that one file separately, not to narrow `--range`; don't raise
 `--max-diff-chars` blind.
@@ -102,7 +102,10 @@ as a diff that silently is not read. Name a provider explicitly only when you ha
 
 ## Rules
 - Read-only second opinion — the tool never edits, commits, or pushes. YOU apply fixes in the main thread.
-- Treat exit 2/3 as blocking-unknown, never as approval (the tool is fail-closed by design).
+- Treat exit **3 or 4** as blocking-unknown, never as approval (the tool is fail-closed by
+  design). 3 = could not run, 4 = ran and produced nothing usable. **Exit 2 is not
+  emitted by this tool** — across caddis 2 always means an advisory note, and a review
+  has no advisory verdict.
 - A different vendor means a different style; weigh its findings on merit, don't cargo-cult them.
 - **Never use `find`, `Get-ChildItem -Recurse`, or any other filesystem-wide search to locate
   `oss_review.py`.** Check only the two paths above. On Git Bash under Windows, `find /` (or any

package/bundles/antigravity-plugin/skills/spawn-session/SKILL.md

@@ -348,7 +348,7 @@ immediately.
 
 | Verdict | Condition | You write |
 |---|---|---|
-| **PASS** | every answer correct, citations check out | a short go-ahead naming the first task |
+| **ACCEPT** | every answer correct, citations check out | a short go-ahead naming the first task |
 | **ACCEPT WITH CORRECTIONS** | broadly right; one or two facts stale or a citation wrong | the corrections, each with the file that proves it |
 | **REJECT** | a settled decision misremembered, a withdrawn figure quoted, or a hazard misunderstood | an instruction to re-read named sections and re-answer. **Do not let it start work** |
 
@@ -383,7 +383,7 @@ Round 1's verdict sets round 2's size:
 
 | Round 1 | Round 2 task |
 |---|---|
-| PASS | a normal next item; review the outcome |
+| ACCEPT | a normal next item; review the outcome |
 | ACCEPT WITH CORRECTIONS | a **small, reversible** item that writes nothing permanent; review closely |
 | REJECT | no task — re-read and re-answer |
 

package/bundles/manifest.json

@@ -1,7 +1,7 @@
 {
-  "poolVersion": "1.3.96",
+  "poolVersion": "1.3.97",
   "bundles": {
-    "antigravity-plugin": "1.3.96",
+    "antigravity-plugin": "1.3.97",
     "antigravity-plugin-extras": "1.3.28",
     "codex": "unknown"
   }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@caddis/cli",
-  "version": "0.4.9",
+  "version": "0.4.10",
   "description": "Install, update and diagnose the caddis dev harness across every coding agent you use (Claude Code, agy/Antigravity, and more).",
   "keywords": [
     "caddis",