playmaker-cli 0.5.1__tar.gz → 0.7.0__tar.gz

{playmaker_cli-0.5.1 → playmaker_cli-0.7.0}/CHANGELOG.md

@@ -5,6 +5,77 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.7.0] - 2026-08-05
+
+### Fixed
+
+- **A batch label is a name you reuse, so the summary now belongs to the
+  fan-out rather than the label.** `--batch dashboard` twice used to mean one
+  batch forever: the second fan-out never pinged at all, because exactly-once
+  was guarded by an `O_EXCL` sentinel in `logs/` that nothing ever removed —
+  and with the sentinel gone it would have counted yesterday's sessions too,
+  `4/4 done · codex ✓ · claude ✓ · codex ✓ · agy ✓` for a fan-out of two, with
+  the stale outputs pasted into the combined `/tmp` file. The claim now lives in
+  `state.db` as a `batch_notified` column: `list_batch` returns the sessions not
+  yet reported, and the finisher that wins the claim releases the label for the
+  next fan-out. Cross-process safety is unchanged — the loser's `UPDATE` finds
+  its set already claimed, rolls back and stays quiet.
+
+- **`playmaker kill` drains the batch it empties.** `killed` is terminal like
+  `done` and `failed`, but `kill` never finalised, so killing a fan-out's last
+  live session left nobody to notice the batch had drained: no summary, ever.
+
+### Changed
+
+- **The combined batch file moved out of `/tmp`.** It is what a batch
+  notification opens on click, so it now lands with the outputs it quotes —
+  `~/.playmaker/outputs/batch-<label>.md` rather than
+  `/tmp/playmaker-batch-<label>.md`, a predictable name in a world-writable
+  directory built from a label the user chose. It also meant the test suite
+  wrote outside `tmp_path`.
+
+## [0.6.0] - 2026-07-27
+
+### Added
+
+- **`opencode` agent handler — one lane, ~75 providers.** The ask was z.ai GLM
+  support, but playmaker never talks to a model API: it drives agent CLIs. So
+  rather than teach the claude handler an `ANTHROPIC_BASE_URL` override — which
+  would have *replaced* the Anthropic lane rather than added one — the handler
+  wraps [opencode](https://opencode.ai), whose `--model provider/model` already
+  reaches GLM (`zai-coding-plan/glm-5.2`), local LMStudio/MLX models, and
+  everything else on models.dev. Oneshot via `opencode run --format json`,
+  resume via `-s <id>`; every JSONL event carries `sessionID`, so a detached
+  dispatch records the id from the first line rather than at completion.
+
+  Four behaviours worth knowing. opencode resolves its working directory from
+  `process.env.PWD`, which `subprocess.Popen(cwd=…)` does not update — left
+  alone it ignores `--cwd` and writes into whatever directory the coach was
+  sitting in, so the handler passes `--dir` and corrects `PWD`.
+  opencode ≥1.18 keeps transcripts in SQLite
+  (`opencode.db`), not one file per session, so playmaker writes a pointer at
+  `~/.playmaker/opencode/<id>.session` and reads the `session`/`message`/`part`
+  tables live — which keeps `thread --follow` current rather than frozen.
+  Cost comes from opencode's own per-session accounting rather than the stream,
+  because `run --format json` can exit before its final `step_finish`
+  ([opencode#26855](https://github.com/anomalyco/opencode/issues/26855)).
+  And `[agents.opencode] model` matters more than the usual `--model` default:
+  left unset, opencode falls back to whatever its own `opencode.json` says,
+  which is typically the model you last picked interactively rather than the
+  one you meant to dispatch to.
+
+  Permissions follow agy, for the same reason: `--auto` is opencode's only
+  lever, so a detached run either auto-approves or comes back having done
+  nothing. `yolo` defaults to `true`; narrow it with the `permission` block in
+  your own opencode.json, which `--auto` still honours for denies.
+- **`zai` quota probe** — GLM Coding Plan windows (5-hour, weekly, and the
+  monthly MCP tool pool) from `api.z.ai/api/monitor/usage/quota/limit`, keyed
+  off the credential opencode already wrote to its `auth.json`, so playmaker
+  still stores no secret of its own. Reported as its own provider rather than
+  under `opencode`, because the quota belongs to the plan — an opencode lane
+  pointed at a local model spends nothing here. Degrades to *unsupported*, not
+  an error row, when no Z.ai credential exists.
+
 ## [0.5.1] - 2026-07-27
 
 ### Fixed

{playmaker_cli-0.5.1 → playmaker_cli-0.7.0}/PKG-INFO

@@ -1,14 +1,14 @@
 Metadata-Version: 2.4
 Name: playmaker-cli
-Version: 0.5.1
-Summary: Playing-coach CLI for orchestrating Claude Code, Codex and Antigravity sub-agents in parallel.
+Version: 0.7.0
+Summary: Playing-coach CLI for orchestrating Claude Code, Codex, Antigravity and opencode sub-agents in parallel.
 Project-URL: Homepage, https://github.com/vladsafedev/playmaker
 Project-URL: Repository, https://github.com/vladsafedev/playmaker
 Project-URL: Issues, https://github.com/vladsafedev/playmaker/issues
 Author-email: Vladislav Shulyugin <vladislav.shulyugin@gmail.com>
 License-Expression: MIT
 License-File: LICENSE
-Keywords: agents,ai,antigravity,claude,claude-code,cli,codex,gemini,orchestration,subagents
+Keywords: agents,ai,antigravity,claude,claude-code,cli,codex,gemini,glm,opencode,orchestration,subagents
 Classifier: Development Status :: 3 - Alpha
 Classifier: Environment :: Console
 Classifier: Intended Audience :: Developers
@@ -28,11 +28,11 @@ Description-Content-Type: text/markdown
 # playmaker
 
 [![CI](https://github.com/vladsafedev/playmaker/actions/workflows/ci.yml/badge.svg)](https://github.com/vladsafedev/playmaker/actions/workflows/ci.yml)
-[![PyPI](https://img.shields.io/pypi/v/playmaker-cli.svg)](https://pypi.org/project/playmaker-cli/)
-[![Python](https://img.shields.io/pypi/pyversions/playmaker-cli.svg)](https://pypi.org/project/playmaker-cli/)
+[![PyPI](https://img.shields.io/pypi/v/playmaker-cli.svg?cacheSeconds=3600)](https://pypi.org/project/playmaker-cli/)
+[![Python](https://img.shields.io/pypi/pyversions/playmaker-cli.svg?cacheSeconds=3600)](https://pypi.org/project/playmaker-cli/)
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
 
-**Run Claude Code, Codex and Antigravity as parallel sub-agents from one terminal — and spend three separate quotas instead of one.**
+**Run Claude Code, Codex, Antigravity and opencode as parallel sub-agents from one terminal — and spend separate quotas instead of one.**
 
 You stay in your Claude Code session doing the part only you can do. `playmaker`
 fans the rest out to other agent CLIs as detached processes, tracks them,
@@ -74,10 +74,12 @@ in one serial session:
 1. **Wall-clock speed.** A task that decomposes into 3–5 independent
    work-streams (schema, backend, frontend, tests, docs) finishes 2–4× faster
    when each stream runs as its own parallel agent.
-2. **Provider arbitrage.** Codex and Antigravity quotas are entirely separate
-   pools from your Anthropic plan. Every slice you hand them is capacity your
-   main session never spends — and Antigravity's roster includes Claude
-   Sonnet/Opus, so even Claude-quality work can run on Google's pool.
+2. **Provider arbitrage.** Codex, Antigravity and opencode quotas are entirely
+   separate pools from your Anthropic plan. Every slice you hand them is
+   capacity your main session never spends — and Antigravity's roster includes
+   Claude Sonnet/Opus, so even Claude-quality work can run on Google's pool.
+   `opencode` widens this the most: one CLI fronting ~75 providers, from a z.ai
+   GLM coding plan to models running locally on your own machine.
 3. **Bucket arbitrage inside one plan.** Headless `claude -p` draws on the same
    subscription as your interactive session, but per-model weekly buckets are
    separate — dispatching `--model sonnet` spends Sonnet's usually-idle bucket
@@ -134,6 +136,16 @@ The other agents differ, because their CLIs do:
   either auto-approves or comes back having done nothing. It therefore
   defaults to `yolo = true`; layer `sandbox = true` on top for agy's own
   terminal restrictions.
+- **opencode** is the same story: its only lever is `--auto`, so it also
+  defaults to `yolo = true`. The granular control lives in opencode's own
+  config rather than in playmaker — and `--auto` still honours it, because it
+  auto-approves only what you have not explicitly denied:
+
+  ```jsonc
+  // ~/.config/opencode/opencode.json
+  { "permission": { "edit": "allow", "bash": "allow", "webfetch": "deny" } }
+  ```
+
 - **gemini** (legacy) runs with `--yolo`.
 
 ## Install
@@ -170,6 +182,7 @@ playmaker agents          # which agent CLIs are reachable
 | **Claude Code** | `npm i -g @anthropic-ai/claude-code` | `--model sonnet` / `opus` / `haiku` |
 | **Codex CLI** | `npm i -g @openai/codex` | the model roster depends on your plan; omit `--model` to use the account default |
 | **Antigravity (`agy`)** | bundled with [Antigravity](https://antigravity.google) | `--model claude-opus-4-6-thinking` — the roster moves, so read it from `agy models` |
+| **opencode** | `brew install sst/tap/opencode` (or see [opencode.ai](https://opencode.ai)) | `--model provider/model`, e.g. `zai-coding-plan/glm-5.2`; roster from `opencode models`, providers from `opencode auth login` |
 | **Gemini CLI** (legacy) | `npm i -g @google/gemini-cli` | still supported, superseded by `agy` |
 
 At least one is required; `playmaker agents` tells you which it can see.
@@ -233,9 +246,11 @@ flowchart LR
     P --> A1["claude -p<br/>--model sonnet"]
     P --> A2["codex exec"]
     P --> A3["agy -p"]
+    P --> A4["opencode run<br/>-m zai-coding-plan/glm-5.2"]
     A1 --> S[("state.db<br/>outputs/ + logs/")]
     A2 --> S
     A3 --> S
+    A4 --> S
     S -->|"list / thread / summary"| C
     S -.->|"batch drained"| N["one ping"]
 ```
@@ -248,8 +263,10 @@ bookkeeping in between.
 ├── state.db          SQLite — sessions, status, pids, models, output paths
 ├── config.toml
 ├── agents/           optional agent profile markdown (claude.md, codex.md, agy.md…)
-├── outputs/          final output per session — .md, or .json if the agent returned JSON
+├── outputs/          final output per session — .md, or .json if the agent returned JSON,
+│                     plus batch-<label>.md, every output in one fan-out combined
 ├── logs/             subprocess stdout for detached runs
+├── opencode/         pointer per opencode session (its transcript lives in SQLite)
 └── quotas.json       latest capacity snapshot
 ```
 
@@ -261,6 +278,7 @@ and locates the session file the tool writes locally. Empirically:
 | Claude | `~/.claude/projects/<cwd-with-slashes-as-dashes>/<id>.jsonl` |
 | Codex | `~/.codex/sessions/<YYYY>/<MM>/<DD>/rollout-<ts>-<thread_id>.jsonl` |
 | Antigravity | `~/.gemini/antigravity-cli/brain/<conversation-id>/.system_generated/logs/transcript_full.jsonl` |
+| opencode | SQLite — `~/.local/share/opencode/opencode.db` (`session` / `message` / `part`); playmaker keeps a pointer at `~/.playmaker/opencode/<id>.session` |
 | Gemini | `~/.gemini/tmp/<cwd-basename>/chats/session-<ts>-<short_id>.{json,jsonl}` |
 
 `thread` and `summary` normalize all of them into the same turn list, so every
@@ -276,6 +294,21 @@ dispatch; and a wrong `--model` is a silent failure on both **codex** (reports
 `turn.failed` while exiting 0) and **agy** (runs its default model instead) —
 playmaker turns both into real errors.
 
+**opencode** deserves its own note, because one CLI is many providers. Models
+are `provider/model` strings from `opencode models`, and if you don't pass
+`--model` opencode falls back to the default in *its* config — often whatever
+you last picked interactively. Set the lane's default once:
+
+```toml
+[agents.opencode]
+model = "zai-coding-plan/glm-5.2"
+```
+
+It also has agy's working-directory problem in a different costume: opencode
+reads `process.env.PWD`, which a subprocess `cwd` does not update, so left
+alone it would ignore `--cwd` and write into the directory *you* were standing
+in. playmaker passes `--dir` and fixes up `PWD`, so `--cwd` means what it says.
+
 ## Notifications
 
 Every detached dispatch pings when it finishes. With
@@ -321,11 +354,16 @@ Antigravity (agy)
   Gemini weekly     ███████████████████░ 95% left
   Claude/GPT 5h     ██████████████████░░ 90% left  resets in 2h 05m
   Claude/GPT weekly ██████████████░░░░░░ 70% left  resets in 5d 1h
+
+Z.ai (GLM, via opencode)  Max
+  Session     ████████████████████ 100% left
+  Weekly      ███████████████████░ 95% left   resets in 1d 11h
+  MCP tools   ███████████████████░ 99% left   resets in 25d 11h
 ```
 
 The `Weekly` and `Sonnet` rows above are the point: they are **separate
-buckets**. So is every agy row. Routing a subtask is choosing which of them to
-spend.
+buckets**. So is every agy row, and so is the whole Z.ai block. Routing a
+subtask is choosing which of them to spend.
 
 - **Claude** — OAuth usage API; token from the Claude Code Keychain entry.
 - **Codex** — ChatGPT `wham/usage` API; token from `~/.codex/auth.json`.
@@ -336,6 +374,12 @@ spend.
   `retrieveUserQuota` on the Antigravity backend, which surfaces only coarse
   Gemini buckets and is flagged *daemon offline*. Approach ported from
   [steipete/CodexBar](https://github.com/steipete/CodexBar).
+- **Z.ai** — GLM Coding Plan usage API; key from opencode's
+  `~/.local/share/opencode/auth.json` (or `$ZAI_API_KEY`). Reported as its own
+  provider because the quota belongs to the plan, not to opencode — an opencode
+  lane pointed at a local model spends nothing here. Shows *unsupported* rather
+  than an error when no Z.ai credential exists. `MCP tools` is the monthly
+  web-search/reader pool, not inference.
 
 Reading these at *model* granularity is the point: they are the load-balancing
 input the coach skill uses to route each subtask.

{playmaker_cli-0.5.1 → playmaker_cli-0.7.0}/README.md

@@ -1,11 +1,11 @@
 # playmaker
 
 [![CI](https://github.com/vladsafedev/playmaker/actions/workflows/ci.yml/badge.svg)](https://github.com/vladsafedev/playmaker/actions/workflows/ci.yml)
-[![PyPI](https://img.shields.io/pypi/v/playmaker-cli.svg)](https://pypi.org/project/playmaker-cli/)
-[![Python](https://img.shields.io/pypi/pyversions/playmaker-cli.svg)](https://pypi.org/project/playmaker-cli/)
+[![PyPI](https://img.shields.io/pypi/v/playmaker-cli.svg?cacheSeconds=3600)](https://pypi.org/project/playmaker-cli/)
+[![Python](https://img.shields.io/pypi/pyversions/playmaker-cli.svg?cacheSeconds=3600)](https://pypi.org/project/playmaker-cli/)
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
 
-**Run Claude Code, Codex and Antigravity as parallel sub-agents from one terminal — and spend three separate quotas instead of one.**
+**Run Claude Code, Codex, Antigravity and opencode as parallel sub-agents from one terminal — and spend separate quotas instead of one.**
 
 You stay in your Claude Code session doing the part only you can do. `playmaker`
 fans the rest out to other agent CLIs as detached processes, tracks them,
@@ -47,10 +47,12 @@ in one serial session:
 1. **Wall-clock speed.** A task that decomposes into 3–5 independent
    work-streams (schema, backend, frontend, tests, docs) finishes 2–4× faster
    when each stream runs as its own parallel agent.
-2. **Provider arbitrage.** Codex and Antigravity quotas are entirely separate
-   pools from your Anthropic plan. Every slice you hand them is capacity your
-   main session never spends — and Antigravity's roster includes Claude
-   Sonnet/Opus, so even Claude-quality work can run on Google's pool.
+2. **Provider arbitrage.** Codex, Antigravity and opencode quotas are entirely
+   separate pools from your Anthropic plan. Every slice you hand them is
+   capacity your main session never spends — and Antigravity's roster includes
+   Claude Sonnet/Opus, so even Claude-quality work can run on Google's pool.
+   `opencode` widens this the most: one CLI fronting ~75 providers, from a z.ai
+   GLM coding plan to models running locally on your own machine.
 3. **Bucket arbitrage inside one plan.** Headless `claude -p` draws on the same
    subscription as your interactive session, but per-model weekly buckets are
    separate — dispatching `--model sonnet` spends Sonnet's usually-idle bucket
@@ -107,6 +109,16 @@ The other agents differ, because their CLIs do:
   either auto-approves or comes back having done nothing. It therefore
   defaults to `yolo = true`; layer `sandbox = true` on top for agy's own
   terminal restrictions.
+- **opencode** is the same story: its only lever is `--auto`, so it also
+  defaults to `yolo = true`. The granular control lives in opencode's own
+  config rather than in playmaker — and `--auto` still honours it, because it
+  auto-approves only what you have not explicitly denied:
+
+  ```jsonc
+  // ~/.config/opencode/opencode.json
+  { "permission": { "edit": "allow", "bash": "allow", "webfetch": "deny" } }
+  ```
+
 - **gemini** (legacy) runs with `--yolo`.
 
 ## Install
@@ -143,6 +155,7 @@ playmaker agents          # which agent CLIs are reachable
 | **Claude Code** | `npm i -g @anthropic-ai/claude-code` | `--model sonnet` / `opus` / `haiku` |
 | **Codex CLI** | `npm i -g @openai/codex` | the model roster depends on your plan; omit `--model` to use the account default |
 | **Antigravity (`agy`)** | bundled with [Antigravity](https://antigravity.google) | `--model claude-opus-4-6-thinking` — the roster moves, so read it from `agy models` |
+| **opencode** | `brew install sst/tap/opencode` (or see [opencode.ai](https://opencode.ai)) | `--model provider/model`, e.g. `zai-coding-plan/glm-5.2`; roster from `opencode models`, providers from `opencode auth login` |
 | **Gemini CLI** (legacy) | `npm i -g @google/gemini-cli` | still supported, superseded by `agy` |
 
 At least one is required; `playmaker agents` tells you which it can see.
@@ -206,9 +219,11 @@ flowchart LR
     P --> A1["claude -p<br/>--model sonnet"]
     P --> A2["codex exec"]
     P --> A3["agy -p"]
+    P --> A4["opencode run<br/>-m zai-coding-plan/glm-5.2"]
     A1 --> S[("state.db<br/>outputs/ + logs/")]
     A2 --> S
     A3 --> S
+    A4 --> S
     S -->|"list / thread / summary"| C
     S -.->|"batch drained"| N["one ping"]
 ```
@@ -221,8 +236,10 @@ bookkeeping in between.
 ├── state.db          SQLite — sessions, status, pids, models, output paths
 ├── config.toml
 ├── agents/           optional agent profile markdown (claude.md, codex.md, agy.md…)
-├── outputs/          final output per session — .md, or .json if the agent returned JSON
+├── outputs/          final output per session — .md, or .json if the agent returned JSON,
+│                     plus batch-<label>.md, every output in one fan-out combined
 ├── logs/             subprocess stdout for detached runs
+├── opencode/         pointer per opencode session (its transcript lives in SQLite)
 └── quotas.json       latest capacity snapshot
 ```
 
@@ -234,6 +251,7 @@ and locates the session file the tool writes locally. Empirically:
 | Claude | `~/.claude/projects/<cwd-with-slashes-as-dashes>/<id>.jsonl` |
 | Codex | `~/.codex/sessions/<YYYY>/<MM>/<DD>/rollout-<ts>-<thread_id>.jsonl` |
 | Antigravity | `~/.gemini/antigravity-cli/brain/<conversation-id>/.system_generated/logs/transcript_full.jsonl` |
+| opencode | SQLite — `~/.local/share/opencode/opencode.db` (`session` / `message` / `part`); playmaker keeps a pointer at `~/.playmaker/opencode/<id>.session` |
 | Gemini | `~/.gemini/tmp/<cwd-basename>/chats/session-<ts>-<short_id>.{json,jsonl}` |
 
 `thread` and `summary` normalize all of them into the same turn list, so every
@@ -249,6 +267,21 @@ dispatch; and a wrong `--model` is a silent failure on both **codex** (reports
 `turn.failed` while exiting 0) and **agy** (runs its default model instead) —
 playmaker turns both into real errors.
 
+**opencode** deserves its own note, because one CLI is many providers. Models
+are `provider/model` strings from `opencode models`, and if you don't pass
+`--model` opencode falls back to the default in *its* config — often whatever
+you last picked interactively. Set the lane's default once:
+
+```toml
+[agents.opencode]
+model = "zai-coding-plan/glm-5.2"
+```
+
+It also has agy's working-directory problem in a different costume: opencode
+reads `process.env.PWD`, which a subprocess `cwd` does not update, so left
+alone it would ignore `--cwd` and write into the directory *you* were standing
+in. playmaker passes `--dir` and fixes up `PWD`, so `--cwd` means what it says.
+
 ## Notifications
 
 Every detached dispatch pings when it finishes. With
@@ -294,11 +327,16 @@ Antigravity (agy)
   Gemini weekly     ███████████████████░ 95% left
   Claude/GPT 5h     ██████████████████░░ 90% left  resets in 2h 05m
   Claude/GPT weekly ██████████████░░░░░░ 70% left  resets in 5d 1h
+
+Z.ai (GLM, via opencode)  Max
+  Session     ████████████████████ 100% left
+  Weekly      ███████████████████░ 95% left   resets in 1d 11h
+  MCP tools   ███████████████████░ 99% left   resets in 25d 11h
 ```
 
 The `Weekly` and `Sonnet` rows above are the point: they are **separate
-buckets**. So is every agy row. Routing a subtask is choosing which of them to
-spend.
+buckets**. So is every agy row, and so is the whole Z.ai block. Routing a
+subtask is choosing which of them to spend.
 
 - **Claude** — OAuth usage API; token from the Claude Code Keychain entry.
 - **Codex** — ChatGPT `wham/usage` API; token from `~/.codex/auth.json`.
@@ -309,6 +347,12 @@ spend.
   `retrieveUserQuota` on the Antigravity backend, which surfaces only coarse
   Gemini buckets and is flagged *daemon offline*. Approach ported from
   [steipete/CodexBar](https://github.com/steipete/CodexBar).
+- **Z.ai** — GLM Coding Plan usage API; key from opencode's
+  `~/.local/share/opencode/auth.json` (or `$ZAI_API_KEY`). Reported as its own
+  provider because the quota belongs to the plan, not to opencode — an opencode
+  lane pointed at a local model spends nothing here. Shows *unsupported* rather
+  than an error when no Z.ai credential exists. `MCP tools` is the monthly
+  web-search/reader pool, not inference.
 
 Reading these at *model* granularity is the point: they are the load-balancing
 input the coach skill uses to route each subtask.

{playmaker_cli-0.5.1 → playmaker_cli-0.7.0}/pyproject.toml

@@ -1,7 +1,7 @@
 [project]
 name = "playmaker-cli"
-version = "0.5.1"
-description = "Playing-coach CLI for orchestrating Claude Code, Codex and Antigravity sub-agents in parallel."
+version = "0.7.0"
+description = "Playing-coach CLI for orchestrating Claude Code, Codex, Antigravity and opencode sub-agents in parallel."
 readme = "README.md"
 requires-python = ">=3.11"
 license = "MIT"
@@ -19,6 +19,8 @@ keywords = [
     "codex",
     "antigravity",
     "gemini",
+    "opencode",
+    "glm",
     "subagents",
 ]
 classifiers = [

{playmaker_cli-0.5.1 → playmaker_cli-0.7.0}/skills/playmaker-coach/SKILL.md

@@ -1,11 +1,11 @@
 ---
 name: playmaker-coach
-description: Playing-coach orchestration of Claude/Codex/Antigravity (agy) sub-agents via the `playmaker` CLI and in-session Task sub-agents. Triggered when the user gives a complex multi-component task where decomposition gives >2x parallel speedup (e.g. "build admin dashboard with schema, backend, FE, tests, docs"). NOT triggered for single-file refactor, single bug fix, single component, or simple Q&A — those stay in this thread without delegation.
+description: Playing-coach orchestration of Claude/Codex/Antigravity (agy)/opencode sub-agents via the `playmaker` CLI and in-session Task sub-agents. Triggered when the user gives a complex multi-component task where decomposition gives >2x parallel speedup (e.g. "build admin dashboard with schema, backend, FE, tests, docs"). NOT triggered for single-file refactor, single bug fix, single component, or simple Q&A — those stay in this thread without delegation.
 ---
 
 # playmaker-coach — playing-coach orchestration
 
-Use the `playmaker` CLI as a facade to dispatch sub-tasks to Codex / Antigravity (`agy`) / a sibling Claude, monitor them, read their threads, review their diffs, and feed back. The coach (this thread) does its own portion of the work in parallel.
+Use the `playmaker` CLI as a facade to dispatch sub-tasks to Codex / Antigravity (`agy`) / opencode / a sibling Claude, monitor them, read their threads, review their diffs, and feed back. The coach (this thread) does its own portion of the work in parallel.
 
 > **`agy` (Antigravity CLI)** does not only serve Google models: alongside the Gemini Flash and Pro tiers, `agy models` carries **Claude Sonnet and Opus** and a **GPT-OSS** mid-tier. Opus via agy runs on *Google's* quota pool — top-tier Claude work that does not touch the Anthropic subscription's scarce Opus weekly bucket: `--model claude-opus-4-6-thinking`.
 >
@@ -23,7 +23,7 @@ All Claude work — coach, internal sub-agents, external `claude -p` — draws f
 
 3. **External dispatch — `playmaker dispatch <agent>`.** Separate OS processes, tracked in playmaker (`list`/`watch`/`thread`/`continue`):
    - **`claude -p` (sibling Claude):** same subscription. Its key lever is the model bucket — **Sonnet is a separate weekly bucket from Opus**, usually idle while Opus depletes, so default **`--model sonnet`** to spare the scarce Opus bucket (**`--model haiku`** for trivial mechanical work). It **can write files** — playmaker runs it in `acceptEdits`, so it edits and runs commands freely inside `--cwd` and is refused outside it (see §10). Use it over an internal sub-agent when you want a **tracked, detached work-stream** you can monitor/continue independently of the coach's turn.
-   - **`codex` / `agy`:** each on its own subscription/quota — the right home for **write-heavy** parallel implementation that can leave the Anthropic subscription. `agy` is special: besides Gemini tiers it carries **Claude Sonnet/Opus 4.6 (Thinking)** on Google's pool, so even "must be Claude-quality" work can leave the Anthropic quota.
+   - **`codex` / `agy` / `opencode`:** each on its own subscription/quota — the right home for **write-heavy** parallel implementation that can leave the Anthropic subscription. `agy` is special: besides Gemini tiers it carries **Claude Sonnet/Opus 4.6 (Thinking)** on Google's pool, so even "must be Claude-quality" work can leave the Anthropic quota. `opencode` is the widest lane: one CLI over ~75 providers, addressed as `provider/model` — a GLM coding plan (`zai-coding-plan/glm-5.2`), or a model running locally on this machine, which spends **no** subscription quota at all.
 
 **Routing cheat-sheet for Claude-side work:**
 
@@ -32,8 +32,10 @@ All Claude work — coach, internal sub-agents, external `claude -p` — draws f
 | writes files and the coach integrates the result directly | **internal sub-agent** (Task tool) | in-session, write-capable, returns into context |
 | is an independent stream you'll monitor / continue separately | **external `dispatch claude --model sonnet`** | tracked & detached; Sonnet's own weekly bucket spares Opus |
 | is heavy reasoning only the coach can do | **coach** | top-tier Opus, serial |
-| is write-heavy and can leave Claude | **codex / agy** | their own quotas |
+| is write-heavy and can leave Claude | **codex / agy / opencode** | their own quotas |
 | needs top-tier Claude but the Anthropic Opus weekly is precious | **`dispatch agy --model claude-opus-4-6-thinking`** | Opus quality on Google's pool |
+| is bulk work and every subscription is running low | **`dispatch opencode --model zai-coding-plan/glm-5.2`** | a separate GLM plan, untouched by the others |
+| is mechanical and privacy-sensitive, or all quotas are spent | **`dispatch opencode --model <local provider>/<model>`** | runs on this machine; costs no quota, just wall-clock |
 
 **Sonnet is your cheap parallel Claude worker** — a separate weekly bucket that usually sits idle while Opus depletes. Reach for it (via `dispatch claude --model sonnet`, or by pointing an internal sub-agent at Sonnet) instead of burning Opus on mid-tier work.
 
@@ -61,6 +63,7 @@ playmaker quotas --refresh          # current capacity, broken out per model
 - Claude: two non-coach ways to run it — an **internal sub-agent** (Task tool; in-session, write-capable, result returns to the coach) and an **external `claude -p` dispatch** (tracked, detached stream). Both draw on the subscription; the difference is where results land, not cost. See "Execution lanes". Default external Claude to `--model sonnet` — its weekly bucket is separate from Opus and usually idle, so it spares the scarce coach (Opus) bucket.
 - Antigravity (`agy`): one Google pool split across model families — Gemini Flash tiers for cheap bulk work, Gemini 3.1 Pro for hard Gemini work, **Claude Sonnet/Opus 4.6 (Thinking)** as Anthropic-quality capacity that spends *Google's* quota, GPT-OSS 120B as a spare mid-tier. `playmaker quotas` shows the **full categorized breakdown** — `Gemini 5h` / `Gemini weekly` and `Claude/GPT 5h` / `Claude/GPT weekly`. Two things share a bucket: all Gemini models draw the Gemini bucket, and Claude *and* GPT-OSS share the Claude/GPT bucket. So dispatching Opus 4.6 via agy spends the same `Claude/GPT` bucket as Sonnet or GPT-OSS — watch the `Claude/GPT 5h` window when fanning out several agy-Claude jobs. (This needs agy's local daemon up — normally true when any agy process is running; if `playmaker quotas` tags agy "daemon offline" it fell back to a coarse Gemini-only view.)
 - Codex: top-tier vs lighter modes (where applicable).
+- Z.ai (GLM): shown as its own provider because the plan is what has the quota, not the CLI — `Session` (5h) and `Weekly` windows plus a monthly `MCP tools` pool that only web-search/reader calls touch. It appears whenever `opencode auth login` has a Z.AI credential, and reads *unsupported* when it doesn't. An `opencode` dispatch pointed at a **local** model spends none of it, so local lanes never show up in this table at all.
 
 If `quotas.json` is more than 1h old (or shows errors), say so before relying on the numbers.
 
@@ -159,6 +162,8 @@ Two exceptions to "always pass `--model`":
 - **codex** — its model roster depends on the account plan, and an unavailable name fails the whole dispatch (`codex turn failed: … not supported when using Codex with a ChatGPT account`). Omitting `--model` uses whatever that account actually has, which is usually what you want.
 - **agy** — its own default is a top-tier model, so omitting `--model` on a dispatch meant to be cheap silently spends the expensive bucket. Always pass it here.
 
+**opencode is `provider/model`, and the default is a trap.** Names look like `zai-coding-plan/glm-5.2` or `lmstudio/qwen/qwen3-coder-30b` — **run `opencode models` and copy a line** rather than writing one from memory; playmaker validates against that roster and fails the dispatch on a name it doesn't contain. Omitting `--model` does *not* fall back to something sensible: it uses whatever is set in the user's own `~/.config/opencode/opencode.json`, which is typically the last model they picked interactively — often a small local one. So for opencode, always pass `--model` unless `[agents.opencode] model` is already set in `~/.playmaker/config.toml`.
+
 **agy prompt discipline:** the agy agent's shell lives in a private scratch directory, not the workspace. playmaker automatically prepends a workspace preamble to every agy dispatch, but reinforce it: word file instructions with paths relative to the workspace root or absolute paths, never "in the current directory".
 
 **Bad-model handling (playmaker ≥0.4 does this for you):** a wrong `--model` is the classic silent failure. Codex returns a `turn.failed` while exiting 0 with empty output; agy silently runs its *default* model instead of erroring. playmaker catches both — a codex model/auth failure raises `codex turn failed: …`, and an unknown agy model raises with the valid roster before dispatch. So a dispatch that comes back **failed** with a model message means: fix the `--model` string (for agy, copy it exactly from `agy models`) and re-dispatch. Don't retry the same string.
@@ -230,7 +235,7 @@ If `playmaker dispatch` returns an error (binary missing, auth bad, agent unavai
 Both Claude lanes are on the subscription, so pick by **where the work lives**, not cost:
 - **Coach folds the result in directly → internal sub-agent (Task tool).** In-session, write-capable, returns into context. Default choice for "more Claude."
 - **Independent stream you'll monitor / continue separately → `playmaker dispatch claude --model sonnet`.** Tracked, detached; Sonnet's separate weekly bucket spares Opus.
-- **Work that can leave the Claude family → Codex / agy** (their own quotas).
+- **Work that can leave the Claude family → Codex / agy / opencode** (their own quotas, or none at all for a local model).
 
 If a dispatch comes back with zero file changes, check `playmaker summary <id>`. Two common causes for Claude: the subtask tried to write outside `--cwd` and was refused (re-dispatch with the right `--cwd`), or the run needed a permission the configured mode doesn't grant — a "I need your permission" answer, not a crash. For **agy** specifically, a "done" with no file changes usually means the files landed in agy's private scratch dir (`~/.gemini/antigravity-cli/scratch/`) — the prompt referred to "the current directory" instead of workspace paths; re-dispatch with explicit paths.
 
@@ -244,7 +249,7 @@ If a dispatch comes back with zero file changes, check `playmaker summary <id>`.
 ## Quota awareness
 
 - **Read Claude quota at model granularity:** Sonnet weekly and Opus weekly are **independent buckets**. Coach and internal sub-agents on Opus spend the scarce one; Sonnet usually sits idle. Push mid-tier work to Sonnet (an internal sub-agent set to Sonnet, or `dispatch claude --model sonnet`) to spare Opus.
-- Read quotas at *model* granularity, not *agent* granularity. For agy, the probe reports four windows — `Gemini 5h/weekly` and `Claude/GPT 5h/weekly`; for Codex, top-tier vs lighter modes.
+- Read quotas at *model* granularity, not *agent* granularity. For agy, the probe reports four windows — `Gemini 5h/weekly` and `Claude/GPT 5h/weekly`; for Codex, top-tier vs lighter modes; for opencode, the quota is reported under the **provider** it is pointed at (`Z.ai`), and a local model reports nothing because it spends nothing.
 - Skip a *model* if its `*_left` is below ~10%; reroute the subtask to the same agent with a different model, or to a different agent entirely.
 - If a top-tier `weekly_*_left` is degrading toward the deadline, push as much work as possible to mid- and cheap-tier models on the same provider (which are often nowhere near depleted), instead of switching providers blindly.
 - agy's **5-hour** window is the one that bites during a fan-out: it "smooths aggregate demand", so a burst of agy dispatches drains `Gemini 5h` or `Claude/GPT 5h` well before the weekly. If a 5h window is low, spread the burst or move some slices to Codex/coach.
@@ -269,7 +274,7 @@ playmaker watch                                 # Rich live TUI of sessions
 
 Every command takes `--json` for machine-readable output.
 
-`--model NAME` is forwarded to the agent's native CLI: `claude --model sonnet`, `agy --model claude-opus-4-6-thinking`, `codex -m <whatever that account has>`. Without it the agent CLI uses its own default. Model is stored on the session row, so detached re-runs and `continue` inherit it; `continue --model X` overrides for that one turn.
+`--model NAME` is forwarded to the agent's native CLI: `claude --model sonnet`, `agy --model claude-opus-4-6-thinking`, `codex -m <whatever that account has>`, `opencode -m zai-coding-plan/glm-5.2`. Without it the agent CLI uses its own default. Model is stored on the session row, so detached re-runs and `continue` inherit it; `continue --model X` overrides for that one turn.
 
 **The agy roster is not documented here on purpose** — it changes with Antigravity releases, and so does the spelling convention. Run `agy models` and copy a line. At the time of writing it returns bare slugs in the shape `gemini-3.6-flash-{low,medium,high}`, `gemini-3.1-pro-{low,high}`, `claude-sonnet-4-6`, `claude-opus-4-6-thinking`, `gpt-oss-120b-medium` — but treat that as an example of the *shape*, not a list to copy from.