@caddis/cli 0.4.64 → 0.4.66

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

@@ -70,6 +70,10 @@ on is worse than one sentence admitting there is nothing to recap.
 
 ## Workstreams and open work
 
+Open items live in three stores (the 3-store model): this session's tasks (saved at handoff to
+`.caddis/relay/<name>.tasks.md`), the open parking-lot items (`/caddis:park`), and the roadmap and the plan
+Tracker for the order of work. Catchup reads those three and nothing else.
+
 **No argument:** this is the last step of the catchup procedure above, not a separate one.
 After the list, add the workstream index and the open work:
 0. The roadmap on origin is the order of work. List the roadmaps, then show each one's status line:
@@ -82,10 +86,8 @@ After the list, add the workstream index and the open work:
 1. Refresh and show the index: `python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_workstreams.py" render-index`,
    then display `.caddis/relay.md`. If it refuses a legacy `.caddis/relay.md`, display that file
    as it is and suggest `/handoff`, which migrates it.
-2. Show open work for the current workstream:
-   ```
-   python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_todo.py" open-work
-   ```
+2. Show the open parking-lot items, the durable backlog: follow `/caddis:park list` (List mode). Print its table
+   and its counts; if the repo has no `.caddis/parking-lot/`, say `No parking-lot items.` and go on.
 Before saying `Nothing is parked.`, check for an old `/digress` stack: if `.caddis/workstreams.json`
 exists, run `caddis_workstreams.py migrate-legacy` (it keeps a backup and reports conflicts
 instead of overwriting), then show the index again. Only with no relays and no old stack, state:
@@ -97,7 +99,11 @@ instead of overwriting), then show the index again. Only with no relays and no o
    `python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_relay_git.py" fetch <name>`. It reads only
    `refs/caddis/relay/<name>`, never checks out or merges, and refuses to replace a different local
    relay. Pass `--replace` only if the user confirms the local one may be overwritten.
-3. Restate where you were and resume the workstream.
+3. Load the saved tasks, the same way `/caddis:fleet-open` does: if `.caddis/relay/<name>.tasks.md` exists (table
+   `| Old # | Status | Subject |`), make one `TaskCreate` per row with the Subject as its subject, then a
+   `TaskUpdate` to `in_progress` for each row with Status `in_progress`. No file means 0 tasks. If this session
+   already holds tasks (`TaskList`), do not create a row whose subject is already there. Count them.
+4. Restate where you were and resume the workstream; name the number of tasks you loaded.
 
 ## Rules
 - **Never** run a destructive or history-rewriting git action (no `git checkout`, `git reset`, `git stash`,

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

@@ -1,12 +1,12 @@
 ---
 name: cross-review
-description: Cross-vendor code review of the current diff — a second-vendor model (DeepSeek/GLM/any OpenAI-compatible endpoint) reviews your changes to catch bugs a same-vendor reviewer misses
+description: Cross-vendor code review of the current diff — a second-vendor model (agy Argon by default; DeepSeek, GLM or any OpenAI-compatible endpoint) reviews your changes to catch bugs a same-vendor reviewer misses
 ---
 
 # /caddis:cross-review — a second-vendor set of eyes on your diff
 
 Run a cross-vendor review of the current changes with a DIFFERENT model family than the one that wrote them
-(default **DeepSeek** — the cheapest + most architecturally distinct-from-Claude option). Use after a phase
+(default **agy Argon** since 2026-10-21 — it needs no key; DeepSeek and GLM are fallbacks). Use after a phase
 is green and before you commit, or for a second opinion on a risky diff. This is the *cross-vendor*
 complement to `/caddis:code-review`, not a replacement.
 
@@ -16,7 +16,7 @@ complement to `/caddis:code-review`, not a replacement.
 python "$TOOL" --check-config      # resolve $TOOL first, see "Run it" below
 ```
 
-Exit 0 means ready; it prints which providers have a key. **Exit 3 means nothing is configured —
+Exit 0 means ready; it prints which providers are available (agy-argon needs no key, only agy on PATH and caddis_spend.py). **Exit 3 means nothing is configured —
 STOP and ask the user now.** Do not proceed to the review, and do not silently skip it.
 
 Ask with `AskUserQuestion`: which provider to set up (DeepSeek, GLM, or another OpenAI-compatible
@@ -65,7 +65,7 @@ fi
 > copied weeks earlier silently won. `caddis_gate.py vendor-drift` now fails on the same condition.
 ```
 python "$TOOL" $ARGUMENTS                                # pass the user's flags THROUGH, raw
-python "$TOOL"                                           # no arguments: DeepSeek eyes on the working tree
+python "$TOOL"                                           # no arguments: agy Argon eyes on the working tree
 ```
 
 **`$ARGUMENTS` goes through raw, and must not be re-prefixed.** Callers are told to type the flag
@@ -74,7 +74,7 @@ writes. This line used to prefix the flag itself, which turned that into a doubl
 `--range --range main..HEAD` and made every plan-emitted call fail. Whatever the user typed is already a complete flag list:
 
 ```
-/caddis:cross-review                          -> the working tree, DeepSeek
+/caddis:cross-review                          -> the working tree, agy Argon
 /caddis:cross-review --range main..HEAD       -> that range
 /caddis:cross-review --provider glm           -> GLM eyes
 ```
@@ -86,8 +86,8 @@ now means the narrower case — a **single file** too big to split, or more batc
 see it, the fix is usually to review that one file separately, not to narrow `--range`; don't raise
 `--max-diff-chars` blind.
 
-**Do not alternate providers on a schedule.** The order is the owner policy: **DeepSeek, then agy Argon,
-then GLM** (`REVIEW_ORDER` in `oss_review.py`). A named `--provider` goes first; on a provider failure
+**Do not alternate providers on a schedule.** The order is the owner policy (21 Oct 2026): **agy Argon,
+then DeepSeek, then GLM** (`REVIEW_ORDER` in `oss_review.py`). A named `--provider` goes first; on a provider failure
 (HTTP 429, any 5xx, a timeout, a connection error, an empty reply) the tool moves to the next one in
 that order and prints which provider gave the verdict. A provider with no key is skipped. agy runs
 through `caddis_spend.py run` (the spend cap); if the cap refuses, agy is skipped. On a 429 the tool
@@ -95,6 +95,9 @@ waits for `Retry-After` once (at most 60 s); it never splits the diff because of
 reply. A custom endpoint (`--base-url`/`--model`) never falls back. Name a provider only when you have
 a reason.
 
+**Change the default.** The first provider comes from `--provider`, then the process env `REVIEW_PROVIDER`, then
+`REVIEW_PROVIDER` in `~/.caddis/keys.env`, then agy Argon. The others follow in `REVIEW_ORDER`. To make DeepSeek the default again, add `REVIEW_PROVIDER=deepseek` to ~/.caddis/keys.env.
+
 ## Interpret the exit code
 - **0 — REVIEW: CLEAN** → no blocking issues. Proceed.
 - **1 — REVIEW: BLOCKING** → read the printed findings, then **FIX each blocking item** (or explicitly

package/bundles/antigravity-plugin/skills/feature-plan/SKILL.md

@@ -375,7 +375,7 @@ A `risk: high` phase gets two independent review runs: run by the roadmap/plan r
 
 Small work that does not meet a trigger skips its block.
 
-**Trigger: Always.** Add one open-work row for the plan in `.caddis/open-work.md`. Set its cells with `python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_rows.py" set --file <open-work.md> --id <row id> --col Status=todo`, never with a regex edit.
+**Trigger: Always.** The plan's own Tracker is its row of record: give every phase one Tracker row, and when the repo has a roadmap (`caddis_roadmap.py list` prints a slug), name the plan in a roadmap milestone through `/caddis:roadmap` (a docs PR). Work the plan defers on purpose goes to a parking-lot item with `/caddis:park`, and its Tracker row names the file. Open items live in three stores (the 3-store model): the session task list, parking-lot items, and the roadmap and Tracker. The older to-do and repo-backlog stores are retired: do not create them.
 
 **Trigger: The plan records a decision.** Record it with `/caddis:decide`.
 

package/bundles/antigravity-plugin/skills/fleet-close/SKILL.md

@@ -72,7 +72,10 @@ FLEET CLOSE from <coordinator>: write your handoff and ack.
 Then these close rules, in the message:
 1. Run `/caddis:handoff` for **your own** relay only. Never touch another session's files. Skip its
    Step 7 (the `Push this note for another machine?` question): the restart sheet carries your prompt,
-   and a question nobody answers would stop you before you ack.
+   and a question nobody answers would stop you before you ack. Run the handoff's Step 9 as
+   `resume-block ... --fleet`: it prints your Session, Workstream and prompt, and
+   `Start line: from the restart sheet`. The restart sheet holds the only start line. The ack stays a
+   separate reply (rule 5).
 2. In a shared checkout (its `AGENTS.md` says `Shared checkout: yes`, or `git status` shows changes you
    did not make), skip the steps the handoff checklist skips there.
 3. If you run watchers: copy each watcher script to `.caddis/relay/<your name>-files/watchers/`, and fill
@@ -132,7 +135,7 @@ remove that session from the `--no-ack` list before the next `write`.
 
 ## Step 6 — your own handoff and ack
 
-Run `/caddis:handoff` for this session (the coordinator's own relay). Then write your own ack (the
+Run `/caddis:handoff` for this session (the coordinator's own relay), with Step 9 as `resume-block ... --fleet`. Then write your own ack (the
 `FLEET-ACK` block with your fields and your resume prompt) to a temp file and save it with
 `ack add`, as in Step 4. The coordinator's ack is never skipped. Then record this session's presence
 (your session id is printed at session start as `[caddis] this session's id: <id>`; never guess one):

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

@@ -33,11 +33,10 @@ them into its commit, so check first, before Step 1.
    `WORKTREE: <path>`. That worktree is this session's only; no other session's edits can enter it.
    Then, in the steps below:
    - Step 1: tell the `knowledge-transfer` subagent to edit files ONLY under the `WORKTREE:` path.
-   - Step 3, point 4: run `caddis_rows.py set --file <worktree>/.caddis/open-work.md ...`, and make the
-     plan's Tracker edit in `<worktree>/.caddis/plans/<feature>.md`, not in this checkout.
+   - Step 3, point 4: make the plan's Tracker edit in `<worktree>/.caddis/plans/<feature>.md`, and write
+     any new parking-lot item under `<worktree>/.caddis/parking-lot/`, not in this checkout.
    - Step 4: skip `caddis_tidy.py --apply` (report `skipped: shared checkout`).
-   - Step 6: skip `caddis_todo.py reconcile` (report `skipped: shared checkout`; it writes tracked
-     `.caddis/todo-list/` files and has no worktree option).
+   - Step 6: retired; nothing to skip (row 8 reports `skipped: retired (see the 3-store model)`).
    - Step 6b: skip `session_state.py prune` (report `skipped: shared checkout`).
    - Step 6c: run `--reindex` in the worktree (it edits the tracked `DOC-MAP.md`).
    - The relay stays in this checkout: `.caddis/relay/` is local and gitignored.
@@ -53,24 +52,27 @@ them into its commit, so check first, before Step 1.
 ## Checklist
 The handoff is a fixed checklist. Every row ends as `done`, `skipped: <reason>` or `failed: <error>`, and
 Step 8 prints all 13 in this order. Never drop a row from the report, even when its step did not apply.
+Rows 3 and 8 are retired (the 3-store model, 21 Oct 2026): they stay in the list so the row numbers and the
+13-row report do not move for a fleet that mixes caddis versions, and each always reports
+`skipped: retired (see the 3-store model)`. A later release removes them with their scripts.
 
 | # | Row | What it is | Shared checkout (Step 0) |
 |---|---|---|---|
 | 1 | `knowledge-transfer` | Step 1 | edits only in the Step 0 worktree |
 | 2 | `plan-tracker` | the plan's Tracker row (Rules) | edit in the Step 0 worktree |
-| 3 | `open-work-row` | Step 3, point 4: `caddis_rows.py set` | `--file <worktree>/...` |
+| 3 | `open-work-row` | retired: always `skipped: retired (see the 3-store model)` | never runs |
 | 4 | `relay` | Step 3: the relay, written from scratch with current state only | the relay is local; runs |
 | 5 | `tasks-file` | Step 3c | runs |
 | 6 | `kb-index` | Step 6c: refresh the KB index | in the Step 0 worktree |
 | 7 | `violations-report` | Step 6c: report standing violations, each with a one-line fix command | runs (report only) |
-| 8 | `todo-reconcile` | Step 6: `caddis_todo.py reconcile` | `skipped: shared checkout` |
+| 8 | `todo-reconcile` | retired: always `skipped: retired (see the 3-store model)` | never runs |
 | 9 | `usage-record` | Step 6b: `session_state.py usage` | runs |
 | 10 | `tidy-apply` | Step 4: `caddis_tidy.py --apply` | `skipped: shared checkout` |
 | 11 | `state-prune` | Step 6b: `session_state.py prune` | `skipped: shared checkout` |
 | 12 | `prune-list` | Step 6b: `caddis_tidy.py --prune`, list only | runs (lists, never applies) |
 | 13 | `handover-check` | Step 5 | runs |
 
-`tidy-apply`, `state-prune` and `todo-reconcile` run only on `SHARED: no`. Handoff never runs
+`tidy-apply` and `state-prune` run only on `SHARED: no`. Handoff never runs
 `caddis_tidy.py --prune --apply` on its own, in any checkout; only the user can ask for it (Step 6b). In a
 shared checkout the report also lists every tracked file Step 1 changed, under `Docs PR needed:`.
 
@@ -157,11 +159,19 @@ history, and the relay has a 4,000-character budget (`caddis_budget.py`).
 >    to the user and stop.
 > 3. **Read the stored version's hash** before you compose: `caddis_workstreams.py hash <name>`
 >    (prints `none` for a new workstream).
-> 4. **Open-work rows** (only when `.caddis/open-work.md` exists and its frontmatter says
->    `type: open-work`): first set each row this session
->    finished, by id and column name, never by hand:
->    `python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_rows.py" set --file .caddis/open-work.md --id <row id> --col Status=done --col Evidence=<evidence>`.
->    In a shared checkout (Step 0), use `--file <worktree>/.caddis/open-work.md` instead.
+> 4. **Place every open item in its store (the 3-store model).** An open item lives in the store that
+>    matches how long it lives:
+>    - **This session's own tasks:** the native task list. Step 3c saves it to
+>      `.caddis/relay/<name>.tasks.md`, a snapshot that the next handoff replaces. It is never a backlog.
+>    - **An item that must outlive the next handoff** (an owner decision, deferred work, a watch item):
+>      file it with `/caddis:park`, once, and name its slug in the relay's `## Open questions / blockers`.
+>      Park only work we will do later (`park.md` Step 1). A parking-lot file is tracked, so in a shared
+>      checkout (Step 0) write it in the Step 0 worktree.
+>    - **The order of work:** the plan Tracker and the roadmap (below). Set the Tracker row of each phase this
+>      session finished (Rules).
+>    Row 3 (`open-work-row`) is retired: report it exactly as `skipped: retired (see the 3-store model)`.
+>    The `.caddis/open-work.md` and `.caddis/todo-list/` stores are retired. Never write them. A repo that
+>    still holds an old file keeps it as its own document; caddis no longer reads or writes it.
 >    **Order of work:** when the repo has a roadmap (`caddis_roadmap.py list` prints a slug), the
 >    roadmap on origin is the order of work. Write the relay's "Next step (exact)" as
 >    `next: top open item of roadmap <slug> (caddis_roadmap.py status --roadmap <slug>)`. Never copy a
@@ -209,7 +219,8 @@ Paths are relative to the repo root.
 <Commands run this session and their result: pass / fail / not-run. Be honest.>
 
 ## Open questions / blockers
-<Decisions needed, ambiguities, anything unverified. "None" if truly none.>
+<Decisions needed, ambiguities, anything unverified. "None" if truly none. An item that must outlive the
+next handoff is filed with `/caddis:park`: name its slug here, never copy its text.>
 
 ## Learnings captured
 knowledge-transfer: <✓ ran → files written | ✗ skipped → reason>
@@ -228,7 +239,7 @@ STATUS: <CONTINUE | OWNER-GATE <P>-OG<n> | BLOCKED <owner> | FAILED <reason>>
 
 Small work that does not meet a trigger skips its block.
 
-**Trigger: Always.** The relay ends with a status line, the last non-empty line, and it is exactly one of `STATUS: CONTINUE`, `STATUS: OWNER-GATE <P>-OG<n>`, `STATUS: BLOCKED <owner>` or `STATUS: FAILED <reason>`. In Step 3, point 5, write the relay with `python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_workstreams.py" put <name> <temp-file> --require-status --expected-hash <hash>` instead of `write` (in short: put --require-status), with the same hash Step 3 read, so a peer's newer relay is never overwritten. It rejects a relay with no valid status line, and renders the index. When the repo has `.caddis/open-work.md` with `type: open-work` in its frontmatter, also set each open-work row's status with `python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_rows.py" set --file <open-work.md> --id <row id> --col Status=<status>` (in a shared checkout, the copy in the Step 0 worktree). When the status is `done`, also pass `--col Evidence=<commit, PR or test>`: caddis_rows.py refuses `done` with no Evidence.
+**Trigger: Always.** The relay ends with a status line, the last non-empty line, and it is exactly one of `STATUS: CONTINUE`, `STATUS: OWNER-GATE <P>-OG<n>`, `STATUS: BLOCKED <owner>` or `STATUS: FAILED <reason>`. In Step 3, point 5, write the relay with `python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_workstreams.py" put <name> <temp-file> --require-status --expected-hash <hash>` instead of `write` (in short: put --require-status), with the same hash Step 3 read, so a peer's newer relay is never overwritten. It rejects a relay with no valid status line, and renders the index.
 
 ## Step 3c - Save the open tasks
 The task list dies with the session unless it is saved. Write this session's open tasks beside
@@ -258,7 +269,7 @@ be blocked by a script it does not have. Include its report in your output. A **
 (`done/<name>` already exists) is reported by the script, not fatal — surface it, do not fail the
 handoff over it. Exit 2 means a plan is held because its Tracker still has open rows (`held (open Tracker rows): ...` lines, from Phase 3): show those lines; finish, park (`.caddis/parking-lot/<file>.md` in the row) or mark the rows `dropped: <reason>`, or set the plan status back. Do not fail the handoff over it.
 
-After step 4, for each item this session closed (a phase `<plan>#<n>`, a plan slug, or an open-work row id), run `mentions` to check whether another relay's Next step still names it:
+After step 4, for each item this session closed (a phase `<plan>#<n>`, a plan slug, or a parking-lot slug), run `mentions` to check whether another relay's Next step still names it:
 ```
 python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_workstreams.py" mentions --item "<item>" --except <name>
 ```
@@ -280,12 +291,10 @@ incoming session looking for a prompt that was never written. Fix what it lists,
 It also notes any generated artefact committed before its generator last changed. That is advisory,
 not blocking — but do not quote a report the note names without rebuilding it first.
 
-## Step 6 — reconcile task rows from session-state
-Run task reconciliation against the newest session-state file:
-```
-python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_todo.py" reconcile --session-state .caddis/session-state/<session-id>.md
-```
-This marks ad-hoc to-do items done by metadata ID; it never edits the Tracker and ignores phase rows.
+## Step 6 — retired (task reconcile)
+This step is retired. It used to reconcile the `todo-list/` store, which is retired too (the 3-store model),
+so the step does nothing: do not run any script for it. Report row 8 (`todo-reconcile`) as
+`skipped: retired (see the 3-store model)`. The script stays in place for older installs only.
 
 ## Step 6b — record usage and tidy session state
 The Stop hook only writes the session-state file. These two used to run on every turn; they run
@@ -350,18 +359,18 @@ index or staged files — after the secret filter passes. Report its one-line re
 On the other machine, `/catchup <name>` fetches it.
 
 ## Step 8 — the checklist report
-End the handoff with this block, every row filled with `done`, `skipped: <reason>` or `failed: <error>`:
+Print this block, every row filled with `done`, `skipped: <reason>` or `failed: <error>`:
 
 ```
 === HANDOFF CHECKLIST ===
 1 knowledge-transfer   done | skipped: <reason> | failed: <error>
 2 plan-tracker         <same three forms>
-3 open-work-row        <same three forms>
+3 open-work-row        skipped: retired (see the 3-store model)
 4 relay                <same three forms>
 5 tasks-file           <same three forms>
 6 kb-index             <same three forms>
 7 violations-report    <same three forms>
-8 todo-reconcile       <same three forms>
+8 todo-reconcile       skipped: retired (see the 3-store model)
 9 usage-record         <same three forms>
 10 tidy-apply          <same three forms>
 11 state-prune         <same three forms>
@@ -372,6 +381,32 @@ Shared checkout: yes|no <reason>
 
 Then, in a shared checkout, `Docs PR needed:` with the `PR:` line and the file list from Step 0.
 
+Step 9 prints after this report. It is the last output of the handoff.
+
+## Step 9 — print the resume prompt (always the last output)
+The owner should never have to ask for the resume prompt. After the checklist report, run:
+```
+python "${CLAUDE_PLUGIN_ROOT}/scripts/caddis_fleet.py" resume-block --name <name> --session-id <session-id> --root .
+```
+`<name>` is the workstream name from Step 3. It reads the relay's `## Resume prompt` block verbatim, and
+the session name (`session-state/<session-id>.name` when `whoami` wrote it, else the first 8 characters of
+the id). It prints, and you print its output as it is, with nothing after it:
+```
+=== RESUME ===
+Session: <session name> (<session id>)
+Workstream: <name>
+<the resume prompt, in a fenced block>
+Start line:
+claude --name <role>-<MMDD>-<HH> '<the resume prompt on one line>'
+```
+- **Fleet close.** When this handoff runs because a peer sent `FLEET CLOSE from <coordinator>`, add
+  `--fleet`. The Start line is then `Start line: from the restart sheet`: the restart sheet is the only
+  start line in a fleet close. The `FLEET-ACK` reply is a separate message and does not change.
+- **Exit 1** (no `## Resume prompt` block, or no relay) prints nothing on stdout. The handoff is not
+  complete: fix the relay (Step 3, point 5) and run the command again. Never write the block by hand.
+- Workstream with no `whoami` record: the new name uses the workstream name as its role. Pass
+  `--role <role>` to set another.
+
 ## Rules
 - **One next action, by priority ladder.** `## Next step` names exactly ONE action — not a menu. Pick it in
   order: (1) interrupted mid-phase work → finish + commit it; else (2) the active plan's next not-started

package/bundles/antigravity-plugin/skills/roadmap/assets/plan-resume-prompt.md

@@ -15,7 +15,7 @@ Use the scripts in that folder; do not look for them in another caddis install.
 
 1. The plan `.caddis/plans/{{slug}}.md`: its goal, its decisions, its tracker. Every decision binds you.
 2. The relay `.caddis/relay/{{slug}}.md`. Its "Next step" says where to start.
-3. The open-work list `.caddis/open-work.md`, when it exists.
+3. The open parking-lot items (`.caddis/parking-lot/*.md` with `status: open`) that name this plan, when any.
 
 ## 2. Step 0: reconcile
 
@@ -47,8 +47,8 @@ publish a branch that does). Commit the phase there, with its tracker row set to
 - The runner reviews `risk: high` phases. A phase marked `**Risk:** high`: do not halt for it and do not run the
   review yourself. When its exit gate is green, set its row `done` and end with `STATUS: CONTINUE`. The runner then
   reviews it twice and, on findings, starts one fix session.
-- Set the tracker row of the batch to `done` (with its evidence) when its exit gate is green. Set open-work rows with
-  `python <scripts>/caddis_rows.py set` (by row id and column name).
+- Set the tracker row of the batch to `done` (with its evidence) when its exit gate is green. Work you defer on
+  purpose is filed once with `/caddis:park` (a parking-lot item); never write a to-do or backlog file.
 
 ## 5. Write the relay
 

package/bundles/antigravity-plugin/skills/use-model/SKILL.md

@@ -17,7 +17,9 @@ command is the quick-reference + do-it-now version.
 | **Primary** | `claude` | your Anthropic plan |
 | **OSS lane** | `claude-oss <provider> [claude args…]` | the provider's plan/pay-per-token |
 | **Convenience alias** | `claude-glm [claude args…]` | same as `claude-oss glm` |
-| **Cross-vendor review** | `/caddis:cross-review` (`oss_review.py`) | the reviewer's plan |
+| **Cross-vendor review** | `/caddis:cross-review` (`oss_review.py`; default reviewer agy Argon, then DeepSeek, then GLM) | the reviewer's plan |
+
+To make DeepSeek the default again, add `REVIEW_PROVIDER=deepseek` to ~/.caddis/keys.env.
 
 `claude-oss` / `claude-glm` are launchers at `claude-harness/scripts/claude-oss.{sh,ps1}`. Each
 resolves the endpoint/model/key, sets `ANTHROPIC_BASE_URL` / `ANTHROPIC_MODEL` / `ANTHROPIC_AUTH_TOKEN`

package/bundles/antigravity-plugin-extras/plugin.json

@@ -1,5 +1,5 @@
 {
   "name": "caddis-extras",
-  "version": "1.3.45",
+  "version": "1.3.46",
   "description": "caddis-extras — the long-tail skills for the Antigravity (agy) caddis plugin; install alongside caddis for the full set."
 }

package/bundles/antigravity-plugin-extras/skills/_registry.md

@@ -30,7 +30,7 @@
 | Caching Patterns | `caching-patterns/` | Caching strategies for Streamlit and FastAPI applications |
 | Case Management | `case-management/` | >- |
 | Code Explainer | `code-explainer/` | Explain code with diagrams, analogies and step-by-step walkthroughs. Use when teaching a codebase or answering "how does this work?" |
-| Cross Review | `cross-review/` | Have a different vendor's model (DeepSeek, GLM, any OpenAI-compatible endpoint) review the current diff. Use after a phase is green and before commit/merge, or for a second opinion on a risky diff. |
+| Cross Review | `cross-review/` | Have a different vendor's model (agy Argon by default; DeepSeek, GLM, any OpenAI-compatible endpoint) review the current diff. Use after a phase is green and before commit/merge, or for a second opinion on a risky diff. |
 | Llm Application Dev | `llm-application-dev/` | LLM applications: prompt engineering, RAG and LLM integration. Use for AI-powered features, chatbots or LLM automation. |
 | Mcp Builder | `mcp-builder/` | Build high-quality MCP (Model Context Protocol) servers that let LLMs use external APIs and services, in Python (FastMCP) or Node/TypeScript (MCP SDK). |
 | Observability | `observability/` | Structured logging, correlation IDs, PII redaction, OpenTelemetry tracing, metrics (RED), SLO alerting, health probes and Sentry, in Python and TypeScript. Use when adding logs, traces, metrics or error tracking. |

package/bundles/antigravity-plugin-extras/skills/agent-orchestration/SKILL.md

@@ -458,7 +458,7 @@ LiteLLM cutover. **This table is the single place the intent→model mapping liv
 | **deepest** | Subtle correctness, security invariants, novel architecture, judgment-heavy seams | `opus` | `/model opus`; `Agent(model: "opus")` | **High** — Anthropic plan quota |
 | **broad** | Wide parallel coverage of a large but not-subtle surface; the workhorse | `sonnet` | `/model sonnet`; `Agent(model: "sonnet")` | Medium — the default; budget for several |
 | **mechanical** | Fully-specced, repeat-of-a-known-pattern, no judgment required | `haiku` | `/model haiku`; `Agent(model: "haiku")` | Low |
-| **cross-vendor-reviewer** | A *different vendor's* blind spots on a diff — **code review only** | `deepseek` (default), `glm` | `/caddis:cross-review`, `/glm`, `/deepseek` | **Scarce** — separate starter-tier subscription, NOT Anthropic quota. Ration it. |
+| **cross-vendor-reviewer** | A *different vendor's* blind spots on a diff — **code review only** | `agy-argon` (default), `deepseek`, `glm` | `/caddis:cross-review`, `/glm`, `/deepseek` | **Scarce** — separate starter-tier subscription, NOT Anthropic quota. Ration it. |
 | **synthesis / long-horizon** | Orchestrating many lanes, or work that can't self-verify across a large codebase | `fable` | `/model fable` | **Very high** — ~2× opus's rate-limit burn. RARE. |
 
 **Update this table, not the callers**, when a model is added, retired, or re-tiered. A plan that says
@@ -512,7 +512,7 @@ Invent no new machinery. Everything needed already ships:
 | Per-phase model + lane assignment in a plan | `/caddis:feature-plan` — every phase already carries **Model**, **Lane**, and a cross-review provider |
 
 **Reviewer rule (standing):** the reviewer is always a different vendor than whoever wrote the code.
-Claude or GLM implemented → DeepSeek reviews (or swap). Never same-vendor.
+Claude or GLM implemented → agy Argon reviews by default (DeepSeek, then GLM, as fallbacks). Never same-vendor.
 
 Full command/billing decision table: `docs/guide/multi-model-workflow.md`.
 

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

@@ -1,12 +1,12 @@
 ---
 name: cross-review
-description: Have a different vendor's model (DeepSeek, GLM, any OpenAI-compatible endpoint) review the current diff. Use after a phase is green and before commit/merge, or for a second opinion on a risky diff.
+description: Have a different vendor's model (agy Argon by default; DeepSeek, GLM, any OpenAI-compatible endpoint) review the current diff. Use after a phase is green and before commit/merge, or for a second opinion on a risky diff.
 ---
 
 # Cross-Review — a second-vendor set of eyes on your diff
 
 A different vendor's model has different blind spots. Claude reviewing Claude's diff shares those
-blind spots; a **DeepSeek** (or GLM, or any OpenAI-compatible) reviewer catches a different class of
+blind spots; an **agy Argon** (default), DeepSeek, GLM or other OpenAI-compatible reviewer catches a different class of
 issue. This skill runs that second reviewer over the current changes and acts on its verdict.
 
 ## When to use
@@ -20,7 +20,7 @@ issue. This skill runs that second reviewer over the current changes and acts on
 python "$TOOL" --check-config
 ```
 
-Exit 0 prints which providers have a key. **Exit 3 means nothing is configured — stop and ask the
+Exit 0 prints which providers are available (agy-argon needs no key, only agy on PATH and caddis_spend.py). **Exit 3 means nothing is configured — stop and ask the
 user before reading any diff.** Do not proceed, and do not silently skip the review.
 
 Ask with `AskUserQuestion` which provider to set up, then write the key to `~/.caddis/keys.env` as
@@ -36,15 +36,16 @@ If a key exists for a *different* provider than the one requested, the error nam
 exact flag (`--provider glm`). Read the message before concluding anything is missing.
 
 ## Prerequisites (one-time)
-Only the key is mandatory — the tool ships **provider presets** (default `deepseek`, the cheapest + most
-architecturally distinct-from-Claude option; a review costs well under a cent):
+The default reviewer is **agy Argon** (since 2026-10-21; no key, but agy and `caddis_spend.py` must be present). For an HTTP
+provider only the key is mandatory — the tool ships **provider presets** (`deepseek`, `glm`, `openrouter`):
 ```
 REVIEW_API_KEY   = <your provider key>     # REQUIRED (exit 3 without it)
-REVIEW_PROVIDER  = deepseek                 # optional: deepseek | glm | openrouter (default deepseek)
+REVIEW_PROVIDER  = deepseek                 # optional: agy-argon | deepseek | glm | openrouter (default agy-argon)
 REVIEW_BASE_URL  = ...                      # optional: overrides the preset's URL
 REVIEW_MODEL     = ...                      # optional: overrides the preset's model id
 ```
-Switch provider with one flag: `--provider glm` (or `REVIEW_PROVIDER=glm`). **Future-proofing:** model ids
+Switch provider with one flag: `--provider glm` (or `REVIEW_PROVIDER=glm`, in the env or in `~/.caddis/keys.env`; flag, then env,
+then keys file, then agy Argon). To make DeepSeek the default again, add `REVIEW_PROVIDER=deepseek` to ~/.caddis/keys.env. **Future-proofing:** model ids
 churn — the preset table in `oss_review.py` is the one place to edit a rename, and `REVIEW_MODEL` /
 `REVIEW_BASE_URL` always win, so you can point at any new id without touching code.
 
@@ -91,7 +92,7 @@ smaller chunks; don't just raise `--max-diff-chars` without knowing the endpoint
   Fix it, then re-review only the fix diff. After 3 rounds, ask the owner (rule 3).
 - **4 — error** → no verdict parsed, a git failure, or an endpoint/parse failure. Read stderr; do not
   treat this as CLEAN — investigate.
-- **3 — misconfigured** → `REVIEW_API_KEY` is unset. Set it (see Prerequisites) and re-run.
+- **3 — could not run** → no reviewer is available (no agy on PATH and no API key), or every provider failed. Fix it (see Prerequisites) and re-run.
 
 ## Three rules: one attempt, one review
 1. **Think before you send.** Before a task goes to any coder, write its edge cases and check the facts it depends on (one query or one file read).

package/bundles/codex/.codex/skills/workflow/roadmap/assets/plan-resume-prompt.md

@@ -15,7 +15,7 @@ Use the scripts in that folder; do not look for them in another caddis install.
 
 1. The plan `.caddis/plans/{{slug}}.md`: its goal, its decisions, its tracker. Every decision binds you.
 2. The relay `.caddis/relay/{{slug}}.md`. Its "Next step" says where to start.
-3. The open-work list `.caddis/open-work.md`, when it exists.
+3. The open parking-lot items (`.caddis/parking-lot/*.md` with `status: open`) that name this plan, when any.
 
 ## 2. Step 0: reconcile
 
@@ -47,8 +47,8 @@ publish a branch that does). Commit the phase there, with its tracker row set to
 - The runner reviews `risk: high` phases. A phase marked `**Risk:** high`: do not halt for it and do not run the
   review yourself. When its exit gate is green, set its row `done` and end with `STATUS: CONTINUE`. The runner then
   reviews it twice and, on findings, starts one fix session.
-- Set the tracker row of the batch to `done` (with its evidence) when its exit gate is green. Set open-work rows with
-  `python <scripts>/caddis_rows.py set` (by row id and column name).
+- Set the tracker row of the batch to `done` (with its evidence) when its exit gate is green. Work you defer on
+  purpose is filed once with `/caddis:park` (a parking-lot item); never write a to-do or backlog file.
 
 ## 5. Write the relay
 

package/bundles/manifest.json

@@ -1,8 +1,8 @@
 {
-  "poolVersion": "1.3.150",
+  "poolVersion": "1.3.152",
   "bundles": {
-    "antigravity-plugin": "1.3.150",
-    "antigravity-plugin-extras": "1.3.45",
+    "antigravity-plugin": "1.3.152",
+    "antigravity-plugin-extras": "1.3.46",
     "codex": "unknown"
   }
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@caddis/cli",
-  "version": "0.4.64",
+  "version": "0.4.66",
   "description": "Install, update and diagnose the caddis dev harness across every coding agent you use (Claude Code, agy/Antigravity, and more).",
   "keywords": [
     "caddis",