@caddis/cli 0.4.29 → 0.4.32

package/bundles/antigravity-plugin/skills/catchup/SKILL.md

@@ -5,6 +5,10 @@ description: Where were we? A quick list of what this session was doing and what
 
 # /catchup — where are we?
 
+If $ARGUMENTS is empty, follow the catchup procedure below. If a name is given, follow
+the pop-and-resume procedure at the end of this file. The name selects this mode; the
+procedure still pops the last parked frame.
+
 Answer one question: **what were we doing, and what is still open?**
 
 Not named `/recap` on purpose — Claude Code has a built-in `/recap` that produces a one-line
@@ -58,5 +62,36 @@ on is worse than one sentence admitting there is nothing to recap.
   folder, or `claude --resume` to pick an older session. Mention this only if the recap is thin,
   or if they ask for detail the files do not hold.
 - To **write** a durable resume doc rather than read one: `/caddis:handoff`.
-- To **pop a parked workstream** off the digress stack: `/caddis:resume`. That is a different
-  mechanism — it restores work deliberately set aside, not the last session's state.
+- To **pop a parked workstream** off the digress stack: `/caddis:catchup [name]`. This restores work deliberately set aside.
+
+# /caddis:catchup [name] — pop the parked task and pick it back up
+
+A digression is finished; return to the task you parked with `/digress`. This pops the top frame off the
+workstream stack, restates where you were, realigns `relay.md`, and immediately continues the work — the
+user should not have to remember or re-state anything.
+
+## Step 1 — read the stack
+Read `.caddis/workstreams.json`. If it is **absent, unparseable, `version != 1`, or `stack` is empty**,
+say exactly `Nothing is parked.` and stop. Do nothing else.
+
+## Step 2 — pop the top frame
+The top of stack is the **last** element of `stack` (LIFO — most recently parked). Remove it and write the
+rest of the file back (preserve `version` and any other frames + unknown fields). This is the one write
+this command makes to the state file.
+
+## Step 3 — restate + realign relay.md
+Restate the popped frame to the user: its `plan`, `phase`, and `resumePointer` (and `repo` if set — the
+parked task lives in another repo, so say which). Then edit `.caddis/relay.md`'s `## Next step` section
+**in place** so it matches the popped frame's `resumePointer` — preserve every other section of relay.md
+untouched. (If relay.md or that section is absent, skip this edit silently; the restatement above is enough.)
+
+## Step 4 — resume the work
+Begin executing the `resumePointer` immediately. Ask nothing — the frame already carries the next action.
+If the parked plan lives in another `repo`, note that the user may need to open that repo first, then
+proceed there.
+
+## Rules
+- **Never** run a destructive or history-rewriting git action (no `git checkout`, `git reset`, `git stash`,
+  branch switches). Resuming is metadata-only.
+- Pop exactly one frame per invocation (the LIFO top). Run `/caddis:catchup [name]` again to pop the next.
+- Only real paths and verified facts — restate the frame as written; do not embellish the resume point.

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

@@ -97,7 +97,7 @@ as a diff that silently is not read. Name a provider explicitly only when you ha
 - **0 — REVIEW: CLEAN** → no blocking issues. Proceed.
 - **1 — REVIEW: BLOCKING** → read the printed findings, then **FIX each blocking item** (or explicitly
   justify why one is a false positive). Re-run until CLEAN.
-- **2 — error** → no verdict parsed, or a git/endpoint failure. Read stderr; do NOT treat as clean.
+- **4 — error** → no verdict parsed, or a git/endpoint failure. Read stderr; do NOT treat as clean.
 - **3 — misconfigured** → `REVIEW_API_KEY` is unset. Set it (see Prerequisite) and re-run.
 
 ## Rules

package/bundles/antigravity-plugin/skills/data-lineage/SKILL.md

@@ -26,7 +26,7 @@ Run the End-to-End Data Lineage generation procedure:
 3. **Convert to Confluence Storage Format XHTML:**
    - Run the PlantUML-enabled converter:
      ```powershell
-     python scripts/convert_doc_to_confluence.py docs/<app>/<app>-data-lineage.md --output docs/<app>/<app>-data-lineage.html --diagram-type plantuml
+     python "${CLAUDE_PLUGIN_ROOT}/scripts/convert_doc_to_confluence.py" docs/<app>/<app>-data-lineage.md --output docs/<app>/<app>-data-lineage.html --diagram-type plantuml
      ```
    - Verify that all diagrams convert to native `<ac:structured-macro ac:name="plantuml">` blocks with scale 1.2, natural splines, auto-wrapped edge labels, crimson red borders on impacted components, and embedded Creole legends.
    - Verify that output validates as 100% well-formed XML.

package/bundles/antigravity-plugin/skills/digress/SKILL.md

@@ -7,7 +7,7 @@ description: Park the current workstream on the stack and switch to a new task
 
 You are about to leave the current task for a related-but-different one (a design decision, a sub-feature,
 or a blocker that must be fixed first). This command **records the original task on a workstream stack** so
-neither you nor the user has to remember and re-state it later — `/resume` pops it back with its exact
+neither you nor the user has to remember and re-state it later — `/catchup [name]` pops it back with its exact
 resume point, and every SessionStart surfaces it (`⛏ Parked workstream: …`).
 
 The reason for the detour is **$ARGUMENTS** (if empty, derive a one-line reason yourself from what the

package/bundles/antigravity-plugin/skills/handoff/SKILL.md

@@ -51,7 +51,7 @@ clean phase boundary and lose or double-do the work.
 
 **Digression check.** If this session abandoned a mid-flight plan for a *different* one (relay.md or the
 tracker shows plan A in progress, but the work moved to plan B), suggest the user run `/digress` to park
-plan A on the workstream stack — so its exact resume point survives and `/resume` can pop it back later.
+plan A on the workstream stack — so its exact resume point survives and `/catchup [name]` can pop it back later.
 
 ## Step 2b — Advisory-Hub handoff audit (conditional — skip silently if not applicable)
 
@@ -113,6 +113,13 @@ Nothing to say is the normal answer. Say nothing and move on — this must not b
 > `/caddis:migrate-dir` does that.
 
 ```markdown
+---
+type: relay
+workstream: <name>
+branch: <branch>
+generated: { by: <model>, at: <ISO time> }
+---
+
 # Relay — <feature>
 **Updated:** <ISO timestamp>
 

package/bundles/antigravity-plugin/skills/hld/SKILL.md

@@ -32,7 +32,7 @@ Run the High-Level Design (HLD) generation procedure:
 4. **Convert to Confluence Storage Format XHTML:**
    - Run the PlantUML-enabled converter:
      ```powershell
-     python scripts/convert_doc_to_confluence.py docs/<app>/<app>-hld.md --output docs/<app>/<app>-hld.html --diagram-type plantuml
+     python "${CLAUDE_PLUGIN_ROOT}/scripts/convert_doc_to_confluence.py" docs/<app>/<app>-hld.md --output docs/<app>/<app>-hld.html --diagram-type plantuml
      ```
    - Verify that all diagrams are converted to native `<ac:structured-macro ac:name="plantuml">` blocks with scale 1.2, natural splines, auto-wrapped edge labels, crimson red borders on impacted components, and embedded Creole legends.
    - Ensure multi-stage operational flows use horizontal side-by-side stages (`flowchart LR`) and architecture topologies follow clean 3-tier hierarchies (`flowchart TD`).

package/bundles/antigravity-plugin/skills/implement/SKILL.md

@@ -51,7 +51,9 @@ python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_gate.py" verdict-gate  --plan <plan
 python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_gate.py" tracker-vs-git --plan <plan>
 ```
 
-Obey the exit code: **0** proceed · **1** STOP and report why · **2** proceed but record the note it
+Use the shared scale in `caddis_exit.py`: 0 clean, 1 blocked, 2 advisory, 3 could not run,
+4 error or bad usage. For `caddis_gate.py`, obey the exit code: **0** proceed · **1** STOP and report
+why · **2** proceed but record the note it
 printed · **3** wrong lane — the launch command is on stdout, spawn it · **4** the launch command is
 malformed, do not run it.
 

package/bundles/antigravity-plugin/skills/install-statusline/SKILL.md

@@ -0,0 +1,105 @@
+---
+name: install-statusline
+description: Install the caddis status line into Claude Code and agy — one renderer, once per machine
+---
+
+# /install-statusline — set up the status bar
+
+Wire the caddis status line into every coding agent on this machine. One Python renderer serves
+both hosts. It installs at **user scope**, so every project gets the line and no project holds a
+copy that can drift.
+
+## What the user gets
+
+Claude Code:
+
+```
+⌂ claudster-source | saajunaid/caddis ⎇ main +2 ●1 ?3 ↑1 | ◆ Opus 5 (1M context) v2.1.239 |
+◉ 412k/1M [███░░░░░]  41%  | ⚡ 386k cached | effort:high | style:Plain | thinking |
+5h:12% 7d:40% | 💻 57% (37 GB) | ⏺ relay | ◷ 07:58
+```
+
+agy (Antigravity):
+
+```
+▛▜ agy | ● idle | 🧠 Gemini 3.6 Flash (High) (high) | 📁 claudster-source | ⎇ main ?3 |
+◉ 48.5k/1M [░░░░░░░░]  5%  | ⚡ 40k cached | q 62% | 💻 57% (37 GB) | ⏺ relay | ◷ 08:01
+```
+
+Both lines come from `caddis_statusline.py`. It has two profiles, because the two hosts send
+different field names — agy's payload is a sibling of Claude Code's, not a copy.
+
+## Run it
+
+Find the renderer in this install and run it. Try these paths in order and use the first that exists:
+
+- `${CLAUDE_PLUGIN_ROOT}/scripts/caddis_statusline.py`
+- `claude-harness/scripts/caddis_statusline.py` (working inside the caddis source repo)
+
+Then:
+
+```bash
+python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_statusline.py" --install          # both hosts
+python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_statusline.py" --install --host claude
+python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_statusline.py" --install --host agy
+python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_statusline.py" --install --dry-run # print, write nothing
+```
+
+`--install` does four things:
+
+1. Copies the renderer to `~/.caddis/statusline.py` — a **stable** path. Pointing the host
+   settings at the plugin cache would break on every caddis upgrade, because the version number
+   is in that path.
+2. Creates `~/.caddis/statusline.json` with the default toggles. It never overwrites an existing one.
+3. Backs up each host's `settings.json` to `settings.json.bak-caddis-statusline`.
+4. Sets `statusLine` in `~/.claude/settings.json` and `~/.gemini/antigravity-cli/settings.json`,
+   preserving every other key. A host that is not installed is skipped, not failed.
+
+Report the tool's own output to the user. Then tell them the line appears on the **next** session —
+neither host re-reads `settings.json` mid-session.
+
+## Verify
+
+```bash
+python ~/.caddis/statusline.py --check                        # what is wired where
+python ~/.caddis/statusline.py --sample --profile claude      # preview, no host needed
+python ~/.caddis/statusline.py --sample --profile agy
+```
+
+`--check` exits non-zero when a host's `statusLine` is missing or points somewhere else.
+
+## Change the look
+
+Edit `~/.caddis/statusline.json`. Every key is optional; a missing file means defaults.
+
+| key | default | what it controls |
+|---|---|---|
+| `color` | `true` | ANSI colour |
+| `icons` | `true` | Unicode segment icons. These are plain Unicode, **not** Nerd Font glyphs |
+| `progress_bar` | `true` | the `[███░░░░░]` context bar |
+| `bar_width` | `8` | bar cells |
+| `token_counts` | `true` | `◉ 412k/1M` |
+| `cached_tokens` | `true` | `⚡ 386k cached` — bold and bright by design, never width-gated |
+| `ram` | `true` | `💻 57% (37 GB)` |
+| `ram_min_width` | `0` | hide RAM below this terminal width; `0` always shows it |
+| `clock` | `true` | `◷ 07:58` |
+| `relay` | `true` | `⏺ relay` when a parked resume pointer exists |
+| `rate_limits` | `true` | `5h:12% 7d:40%` (Claude Code only) |
+| `session_id` | `false` | `#a1b2c3d4` |
+| `git` | `true` | branch, staged/unstaged/untracked counts, ahead/behind |
+| `separator` | `"\|"` | segment separator |
+
+Config changes take effect on the next render. No reinstall needed.
+
+## Notes for you, not the user
+
+- Do **not** hand-edit either `settings.json` to wire this. `--install` reads, merges and writes
+  back, so it cannot drop an unrelated key. Hand edits have dropped `trustedWorkspaces` before.
+- The renderer never crashes the host. A bad payload, a missing field, or an unreadable config
+  degrades to fewer segments; a hard failure degrades to the directory name.
+- RAM on Windows uses `ctypes` `GlobalMemoryStatusEx`, an in-process call. Do not replace it with
+  `Get-CimInstance Win32_OperatingSystem` — that cost about 300 ms on **every** render in the
+  PowerShell line this replaced.
+- If the user asks why their old `~/.claude/statusline-custom.sh` or
+  `~/.gemini/statusline/statusline.ps1` no longer runs: nothing points at them any more. They are
+  superseded, not deleted. Both are safe to remove once the user is happy with the new line.

package/bundles/antigravity-plugin/skills/lld/SKILL.md

@@ -39,7 +39,7 @@ Run the Lower-Level Design (LLD) generation procedure:
 4. **Convert to Confluence Storage Format XHTML:**
    - Run the PlantUML-enabled converter:
      ```powershell
-     python scripts/convert_doc_to_confluence.py docs/<app>/<app>-lld.md --output docs/<app>/<app>-lld.html --diagram-type plantuml
+     python "${CLAUDE_PLUGIN_ROOT}/scripts/convert_doc_to_confluence.py" docs/<app>/<app>-lld.md --output docs/<app>/<app>-lld.html --diagram-type plantuml
      ```
    - Verify that all diagrams are converted to native `<ac:structured-macro ac:name="plantuml">` blocks with scale 1.2, natural splines, auto-wrapped edge labels, crimson red borders on impacted components, and embedded Creole legends.
    - Verify that output validates as well-formed XML.

package/bundles/antigravity-plugin/skills/mmr/SKILL.md

@@ -6,12 +6,20 @@ description: Multi-Model Review & Runner (MMR) — run an ad-hoc task in an isol
 # /caddis:mmr — Multi-Model Review & Runner
 
 Execute an ad-hoc task across model lanes (GLM-5.3 → Codex Sol → Claude) inside an isolated
-git worktree with automated test gates and second-vendor review (DeepSeek). Destroys the worktree
-immediately on completion (zero worktree leaks) and leaves the output on a clean branch with a
-simple Verdict Card.
+git worktree with automated test gates and second-vendor review (DeepSeek). A full PASS commits
+code changes and removes the worktree. A FAIL or PENDING keeps the worktree for inspection or
+Claude-lane work. The result appears in a Verdict Card.
 
 When invoked without a task prompt (or with a git range), `/mmr` runs a direct cross-model diff review.
 
+The user typed: **$ARGUMENTS**. Route it to one of the commands below: `keep`, `drop` or `status`
+as given; a git range such as `origin/main..HEAD` to `review --range`; empty to `review`;
+anything else is the task for `run`. Pass supported run flags such as `--type` with the task.
+An unknown flag returns a usage error.
+
+The exit codes follow `caddis_exit.py`: 0 clean, 1 blocked, 2 advisory, 3 could not run,
+4 error or bad usage. In particular, `caddis_adhoc.py` returns 4 for bad usage.
+
 ## Natural Language Triggers
 You can trigger MMR naturally without typing the full slash command:
 - *"Run this on another lane: &lt;task&gt;"* &rarr; `/mmr "<task>"`
@@ -27,21 +35,22 @@ You can trigger MMR naturally without typing the full slash command:
 
 ### 1. Run an Ad-Hoc Task
 ```bash
-python scripts/caddis_adhoc.py run "<task description>" [--type backend|tests|docs|ui|hard]
+python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_adhoc.py" run "<task description>" [--type backend|tests|docs|ui|hard]
 # Shorthand:
-python scripts/caddis_adhoc.py "<task description>"
+python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_adhoc.py" "<task description>"
 ```
 - **Sandbox Isolation:** Creates an ephemeral 1-phase plan at `.caddis/orchestrator/adhoc/<slug>.md`.
 - **Pre-flight Checks:** Verifies dirty tree and API keys. Warns if local edits exist.
 - **Failover Cascade:** Automatically cascades from GLM to Codex to Claude if a lane is out of budget.
 - **Local Test Gates:** Runs project test gates before review.
 - **Cross-Vendor Review:** DeepSeek or GLM reviews the diff.
-- **Immediate Worktree Cleanup:** Stashes/commits results to `lane/adhoc-<slug>` and immediately removes the worktree. Zero open worktrees remain.
-- **Verdict Card:** Prints PASS/FAIL summary with diff stats and next actions.
+- **Worktree result:** PASS commits code changes to `lane/adhoc-<slug>` and removes the worktree. FAIL keeps it. A Claude-lane `PENDING` keeps it and prints the `NEXT:` step. Gates that never ran show `Gates: not run`.
+- **No-op and gate failures:** A run with no code changes fails and keeps its worktree. A pytest gate that collects no tests also fails.
+- **Verdict Card:** Prints PASS, FAIL, or PENDING with diff stats and next actions.
 
 ### 2. Keep the Result
 ```bash
-python scripts/caddis_adhoc.py keep
+python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_adhoc.py" keep
 ```
 - Keeps the output branch `lane/adhoc-<slug>`.
 - **Main Branch Protection:** If you are on `main` or `master`, direct merge is disabled to protect trunk. Recommends running `/ship-pr`.
@@ -49,15 +58,15 @@ python scripts/caddis_adhoc.py keep
 
 ### 3. Drop the Result
 ```bash
-python scripts/caddis_adhoc.py drop
+python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_adhoc.py" drop
 ```
-- Deletes the branch `lane/adhoc-<slug>`.
+- Removes a kept worktree, prunes its registration, and deletes the task branch.
 - Deletes the ephemeral plan file and run logs.
 - Workspace remains clean and untouched.
 
 ### 4. Check Status
 ```bash
-python scripts/caddis_adhoc.py status
+python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_adhoc.py" status
 ```
 - Displays API key readiness for all lanes (`Anthropic`, `DeepSeek`, `GLM`, `OpenAI`).
 - Shows the latest task status, model used, and gates/review verdicts.
@@ -65,11 +74,11 @@ python scripts/caddis_adhoc.py status
 
 ### 5. Review Diff Only
 ```bash
-python scripts/caddis_adhoc.py review [--range <range>]
+python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_adhoc.py" review [--range <range>]
 # Shorthand (no args):
-python scripts/caddis_adhoc.py
+python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_adhoc.py"
 ```
-- Fast cross-model review of working tree or git range (e.g., `origin/main..HEAD`).
+- Fast cross-model review of working tree or git range (e.g., `origin/main..HEAD`). A range can also be passed directly as the first argument.
 
 ---
 

package/bundles/antigravity-plugin/skills/resume/SKILL.md

@@ -1,36 +1,6 @@
 ---
 name: resume
-description: Pop the most recently parked workstream off the stack and resume it at its exact resume point
+description: Renamed to /caddis:catchup (alias, removed next release)
 ---
 
-# /caddis:resume — pop the parked task and pick it back up
-
-A digression is finished; return to the task you parked with `/digress`. This pops the top frame off the
-workstream stack, restates where you were, realigns `relay.md`, and immediately continues the work — the
-user should not have to remember or re-state anything.
-
-## Step 1 — read the stack
-Read `.caddis/workstreams.json`. If it is **absent, unparseable, `version != 1`, or `stack` is empty**,
-say exactly `Nothing is parked.` and stop. Do nothing else.
-
-## Step 2 — pop the top frame
-The top of stack is the **last** element of `stack` (LIFO — most recently parked). Remove it and write the
-rest of the file back (preserve `version` and any other frames + unknown fields). This is the one write
-this command makes to the state file.
-
-## Step 3 — restate + realign relay.md
-Restate the popped frame to the user: its `plan`, `phase`, and `resumePointer` (and `repo` if set — the
-parked task lives in another repo, so say which). Then edit `.caddis/relay.md`'s `## Next step` section
-**in place** so it matches the popped frame's `resumePointer` — preserve every other section of relay.md
-untouched. (If relay.md or that section is absent, skip this edit silently; the restatement above is enough.)
-
-## Step 4 — resume the work
-Begin executing the `resumePointer` immediately. Ask nothing — the frame already carries the next action.
-If the parked plan lives in another `repo`, note that the user may need to open that repo first, then
-proceed there.
-
-## Rules
-- **Never** run a destructive or history-rewriting git action (no `git checkout`, `git reset`, `git stash`,
-  branch switches). Resuming is metadata-only.
-- Pop exactly one frame per invocation (the LIFO top). Run `/resume` again to pop the next.
-- Only real paths and verified facts — restate the frame as written; do not embellish the resume point.
+This command was renamed to /caddis:catchup. Run /caddis:catchup now in its pop-and-resume mode with the same arguments ($ARGUMENTS); if $ARGUMENTS is empty, pass `last` as the name, so this alias still pops the most recently parked task as /caddis:resume always did. Tell the user once: "/caddis:resume is renamed to /caddis:catchup; the old name will be removed in the next release."

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

@@ -1,100 +1,6 @@
 ---
 name: review
-description: Review someone else's OPEN pull request end to end — what changed, what CI says, what is risky — then approve or request changes. Never approves on its own.
+description: Renamed to /caddis:review-pr (alias, removed next release)
 ---
 
-# /caddis:review — review an open PR, then decide
-
-Review pull request **$ARGUMENTS** and walk the operator to an approve-or-not decision.
-
-> **This is the only caddis review surface aimed at a PULL REQUEST.** The others review a local
-> diff: `/caddis:gate-review` (adversarial, same vendor), `/caddis:cross-review` (a different
-> vendor's model). `/caddis:ship-pr` and `/caddis:ship-merge` create and merge PRs. This one is for
-> the case none of those covers — somebody else has opened a PR and you have to decide about it.
->
-> Do not reimplement the review itself. Gather the context, then hand the diff to
-> `/caddis:gate-review` (or Claude Code's native `/code-review`, which is better where it exists).
-
-## Absolute rule
-
-**Never approve, request changes, merge, or close on the operator's behalf.** Present the evidence
-and the exact command; the human runs it. An approval is the one human judgement in the pipeline —
-a review tool that also approves has removed the thing it was built to protect.
-
-## 1. Resolve the PR
-
-`$ARGUMENTS` is a PR number, optionally followed by `owner/repo`. With no repo, use the current
-directory's `origin`. If the number is missing, list the open PRs and stop.
-
-```bash
-gh pr view <n> --repo <owner/repo> --json number,title,author,state,isDraft,mergeStateStatus,reviewDecision,baseRefName,headRefName,additions,deletions,changedFiles,url
-```
-
-**Stop and say so, before reviewing anything**, if:
-- the PR is **merged or closed** — there is nothing to decide;
-- it is a **draft** — say so and ask whether to continue;
-- **the operator is the author.** GitHub does not allow approving your own PR. Say it plainly, and
-  say the alternatives: have someone else approve it, or merge with an admin bypass
-  (`gh pr merge <n> --squash --admin`). Do not silently produce a review they cannot submit.
-
-## 2. Gather what the decision needs
-
-Four things, and report them even when they are boring — "CI is green" is information:
-
-```bash
-gh pr checks <n> --repo <owner/repo>            # every check and its state
-gh pr diff  <n> --repo <owner/repo>             # the change itself
-gh pr view  <n> --repo <owner/repo> --comments  # review comments, and any bot review
-```
-
-If an automated review has already commented, **read it and say whether you agree**. Two reviews
-that disagree is a finding; two that agree is worth stating once, not twice.
-
-`mergeStateStatus` explains a blocked PR without guessing: `BLOCKED` (needs a review),
-`BEHIND` (needs *Update branch*), `UNSTABLE` (a check failed), `DIRTY` (conflicts).
-
-## 3. Review the diff
-
-Hand it to `/caddis:gate-review` — or the native `/code-review` on Claude Code. Then, on top of its
-verdict, check the things that have actually cost this fleet time, because a generic reviewer will
-not know to look:
-
-- **A gate that cannot fail.** A check whose failure path cannot execute — piped to `Out-Null`,
-  wrapped in a `catch{}` a native non-zero exit can never trigger, or asserting something too broad
-  to attribute. It manufactures confidence. Worse than no check.
-- **A wall-clock assertion.** A timing threshold on a shared runner measures load, not the code.
-  Assert ordering instead.
-- **Staged directories.** `git add -A` / `git add .` sweeps in files nobody reviewed.
-- **CI shell assumptions.** The runner may be Windows PowerShell 5.1, where a native command's
-  stderr becomes a terminating error under `$ErrorActionPreference='Stop'`, and where `Should -Be`
-  (Pester 4 syntax) fails on a Pester 3 runner.
-- **Deploy-script changes.** If the pipeline runs the deploy script from a checkout it also updates,
-  the change does not take effect on the deploy that delivers it. Budget two deploys.
-- **Verification by proxy.** An HTTP 200 or a version string is not proof a page renders.
-- **Credentials.** A token scoped to the wrong host; a permission raised to make a test pass; a
-  secret in a URL where git will echo it into a log.
-- **A doc the change makes untrue.** Comments and runbooks that were accurate until this diff.
-
-## 4. Report, then hand over
-
-Lead with the verdict. Keep it short enough to read before deciding:
-
-- **What this PR does** — one or two sentences, not a file list.
-- **Blocking issues** — each with file, line, and why it matters. None is a valid answer, and say it.
-- **Should-fix / nits** — grouped, brief.
-- **CI** — green, red, or still running. Name the failing check.
-- **Merge state** — and what would clear it.
-
-Then print the exact commands, and stop:
-
-```bash
-# approve
-gh pr review <n> --repo <owner/repo> --approve
-gh pr merge  <n> --repo <owner/repo> --squash --delete-branch
-
-# or ask for changes
-gh pr review <n> --repo <owner/repo> --request-changes --body "..."
-```
-
-**If you found a blocking issue, say so in the verdict line** rather than burying it under the
-summary. The operator should be able to read the first line and know whether to keep reading.
+This command was renamed to /caddis:review-pr. Run /caddis:review-pr now with the same arguments ($ARGUMENTS), and tell the user once: "/caddis:review is renamed to /caddis:review-pr; the old name will be removed in the next release."

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

@@ -0,0 +1,100 @@
+---
+name: review-pr
+description: Review someone else's OPEN pull request end to end — what changed, what CI says, what is risky — then approve or request changes. Never approves on its own.
+---
+
+# /caddis:review-pr — review an open PR, then decide
+
+Review pull request **$ARGUMENTS** and walk the operator to an approve-or-not decision.
+
+> **This is the only caddis review surface aimed at a PULL REQUEST.** The others review a local
+> diff: `/caddis:gate-review` (adversarial, same vendor), `/caddis:cross-review` (a different
+> vendor's model). `/caddis:ship-pr` and `/caddis:ship-merge` create and merge PRs. This one is for
+> the case none of those covers — somebody else has opened a PR and you have to decide about it.
+>
+> Do not reimplement the review itself. Gather the context, then hand the diff to
+> `/caddis:gate-review` (or Claude Code's native `/code-review`, which is better where it exists).
+
+## Absolute rule
+
+**Never approve, request changes, merge, or close on the operator's behalf.** Present the evidence
+and the exact command; the human runs it. An approval is the one human judgement in the pipeline —
+a review tool that also approves has removed the thing it was built to protect.
+
+## 1. Resolve the PR
+
+`$ARGUMENTS` is a PR number, optionally followed by `owner/repo`. With no repo, use the current
+directory's `origin`. If the number is missing, list the open PRs and stop.
+
+```bash
+gh pr view <n> --repo <owner/repo> --json number,title,author,state,isDraft,mergeStateStatus,reviewDecision,baseRefName,headRefName,additions,deletions,changedFiles,url
+```
+
+**Stop and say so, before reviewing anything**, if:
+- the PR is **merged or closed** — there is nothing to decide;
+- it is a **draft** — say so and ask whether to continue;
+- **the operator is the author.** GitHub does not allow approving your own PR. Say it plainly, and
+  say the alternatives: have someone else approve it, or merge with an admin bypass
+  (`gh pr merge <n> --squash --admin`). Do not silently produce a review they cannot submit.
+
+## 2. Gather what the decision needs
+
+Four things, and report them even when they are boring — "CI is green" is information:
+
+```bash
+gh pr checks <n> --repo <owner/repo>            # every check and its state
+gh pr diff  <n> --repo <owner/repo>             # the change itself
+gh pr view  <n> --repo <owner/repo> --comments  # review comments, and any bot review
+```
+
+If an automated review has already commented, **read it and say whether you agree**. Two reviews
+that disagree is a finding; two that agree is worth stating once, not twice.
+
+`mergeStateStatus` explains a blocked PR without guessing: `BLOCKED` (needs a review),
+`BEHIND` (needs *Update branch*), `UNSTABLE` (a check failed), `DIRTY` (conflicts).
+
+## 3. Review the diff
+
+Hand it to `/caddis:gate-review` — or the native `/code-review` on Claude Code. Then, on top of its
+verdict, check the things that have actually cost this fleet time, because a generic reviewer will
+not know to look:
+
+- **A gate that cannot fail.** A check whose failure path cannot execute — piped to `Out-Null`,
+  wrapped in a `catch{}` a native non-zero exit can never trigger, or asserting something too broad
+  to attribute. It manufactures confidence. Worse than no check.
+- **A wall-clock assertion.** A timing threshold on a shared runner measures load, not the code.
+  Assert ordering instead.
+- **Staged directories.** `git add -A` / `git add .` sweeps in files nobody reviewed.
+- **CI shell assumptions.** The runner may be Windows PowerShell 5.1, where a native command's
+  stderr becomes a terminating error under `$ErrorActionPreference='Stop'`, and where `Should -Be`
+  (Pester 4 syntax) fails on a Pester 3 runner.
+- **Deploy-script changes.** If the pipeline runs the deploy script from a checkout it also updates,
+  the change does not take effect on the deploy that delivers it. Budget two deploys.
+- **Verification by proxy.** An HTTP 200 or a version string is not proof a page renders.
+- **Credentials.** A token scoped to the wrong host; a permission raised to make a test pass; a
+  secret in a URL where git will echo it into a log.
+- **A doc the change makes untrue.** Comments and runbooks that were accurate until this diff.
+
+## 4. Report, then hand over
+
+Lead with the verdict. Keep it short enough to read before deciding:
+
+- **What this PR does** — one or two sentences, not a file list.
+- **Blocking issues** — each with file, line, and why it matters. None is a valid answer, and say it.
+- **Should-fix / nits** — grouped, brief.
+- **CI** — green, red, or still running. Name the failing check.
+- **Merge state** — and what would clear it.
+
+Then print the exact commands, and stop:
+
+```bash
+# approve
+gh pr review <n> --repo <owner/repo> --approve
+gh pr merge  <n> --repo <owner/repo> --squash --delete-branch
+
+# or ask for changes
+gh pr review <n> --repo <owner/repo> --request-changes --body "..."
+```
+
+**If you found a blocking issue, say so in the verdict line** rather than burying it under the
+summary. The operator should be able to read the first line and know whether to keep reading.