agentpool-cli 0.1.11__tar.gz → 0.1.13__tar.gz

{agentpool_cli-0.1.11 → agentpool_cli-0.1.13}/.github/workflows/release.yml

@@ -6,12 +6,13 @@ on:
       - "v*"
   workflow_dispatch:
 
-permissions:
-  contents: write
+permissions: {}
 
 jobs:
   release:
     runs-on: ubuntu-latest
+    permissions:
+      contents: write
 
     steps:
       - name: Check out
@@ -33,6 +34,7 @@ jobs:
       - name: Run release checks
         run: |
           python -m pytest -q
+          python scripts/check_release_metadata.py
           agentpool models validate --path src/agentpool/provider_model_catalog.json --json
           agentpool config validate --json
           agentpool smoke --provider fake-question --repo . --json
@@ -82,7 +84,9 @@ jobs:
     needs: release
     runs-on: ubuntu-latest
     # Gated on the PUBLISH_TO_PYPI repo variable so tagging never publishes by accident.
-    # Requires a configured PyPI Trusted Publisher and a `pypi` environment.
+    # PyPI Trusted Publisher must be:
+    # project=agentpool-cli, owner=sidduHERE, repo=agentpool, workflow=release.yml, environment=pypi.
+    # Requires the GitHub `pypi` environment and no PyPI token secret.
     if: startsWith(github.ref, 'refs/tags/') && vars.PUBLISH_TO_PYPI == 'true'
     environment:
       name: pypi
@@ -99,3 +103,5 @@ jobs:
 
       - name: Publish to PyPI (Trusted Publishing)
         uses: pypa/gh-action-pypi-publish@release/v1
+        with:
+          print-hash: true

{agentpool_cli-0.1.11 → agentpool_cli-0.1.13}/CHANGELOG.md

@@ -2,6 +2,32 @@
 
 ## Unreleased
 
+## 0.1.13 - 2026-06-24
+
+- Add `poll_worker` as a fast MCP progress snapshot and make
+  `observe_worker(timeout_seconds=0/1)` behave as a true fast poll.
+- Bound MCP observe waits below common host executor timeouts and return timeout
+  metadata instead of leaving coordinators with dropped tool calls.
+- Honor `include_recent_log` for MCP observation, refresh
+  `summary.partial.md` while workers are running, and avoid full transcript
+  scans on observe/poll artifact manifests.
+- Accept terminal state names such as `COMPLETED`, `FAILED`, and `CANCELLED` in
+  `wait_for` and pass observe deadlines into runtime capture calls.
+- Mark `terminate_worker` as explicit, side-effecting, idempotent cleanup in
+  MCP annotations without treating it as user-data destruction.
+
+## 0.1.12 - 2026-06-06
+
+- Add optional Terminal Control runtime support while keeping tmux as the
+  default worker runtime.
+- Fix Claude Code steering submission by using provider-specific submit keys
+  for both initial prompts and live `send_worker_message` calls.
+- Restore MCP capacity/cache compatibility aliases, add searchable MCP tool
+  descriptions and annotations, and make Cursor Agent CLI detection robust when
+  MCP hosts launch AgentPool with a minimal `PATH`.
+- Add real-provider and fake-provider coverage for tmux and Terminal Control,
+  plus focused regression coverage for MCP aliases and provider binary lookup.
+
 ## 0.1.11 - 2026-06-05
 
 - Refresh provider model catalogs from current Claude Code, Codex, Cursor,

{agentpool_cli-0.1.11 → agentpool_cli-0.1.13}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agentpool-cli
-Version: 0.1.11
+Version: 0.1.13
 Summary: Make full use of every coding-agent subscription you pay for: a local CLI + MCP server that surfaces live usage limits and offloads work to providers with headroom.
 Author: AgentPool contributors
 License-Expression: MIT
@@ -40,7 +40,7 @@ informed, never automatic.
 
 The v0.1 alpha posture is conservative:
 
-- tmux is the required runtime.
+- tmux is the default runtime; Terminal Control is optional when configured.
 - Provider selection is explicit; `provider=auto` is rejected.
 - Usage/capacity summaries are confidence-tagged and keyed by provider id.
 - CodexBar and ccusage are optional usage helpers when installed or configured.
@@ -51,8 +51,10 @@ The v0.1 alpha posture is conservative:
 
 - Python 3.11 or newer.
 - tmux on `PATH`.
+- Optional: `termctrl` on `PATH` for the Terminal Control runtime.
 - Git for worktree isolation and diff collection.
-- macOS or Linux. Windows is not a v0.1 target except through WSL-like shells.
+- macOS or Linux for live runtimes. Windows is not a v0.1 target except through
+  WSL-like shells.
 
 ## Install
 
@@ -72,9 +74,21 @@ agentpool setup codex
 agentpool doctor --deep --privacy
 ```
 
-The `agentpool-cli` package installs on macOS, Linux, and Windows, but the
-runtime requires `tmux`, so the supported runtime is macOS or Linux (Windows
-via WSL).
+The `agentpool-cli` package installs on macOS, Linux, and Windows, but live
+terminal runtimes are supported on macOS or Linux (Windows via WSL).
+
+Optional Terminal Control config:
+
+```yaml
+runtime:
+  default: tmux
+  terminal_control:
+    enabled: true
+    binary: termctrl
+    session_prefix: agentpool
+    cols: 120
+    rows: 36
+```
 
 Install from source:
 
@@ -134,7 +148,7 @@ agentpool spawn \
   --task "Inspect the project and ask one clarifying question." \
   --isolation read_only
 
-agentpool observe <session-id> --wait-for completed,error,question,approval_prompt --timeout 120 --json
+agentpool observe <session-id> --wait-for completed,error,question,approval_prompt --timeout 60 --json
 agentpool send <session-id> "Continue with the smallest useful check."
 agentpool artifacts <session-id> --json
 agentpool transcript <session-id> --tail-lines 80 --json
@@ -185,8 +199,8 @@ agentpool usage-summary --refresh --backend ccusage --provider claude-code --jso
 ordered and it is not a recommendation list. Each row includes `usable`,
 `unusable_reason`, quota windows, confidence, age/staleness, and reset timing when
 the provider exposes it. The older CLI `capacity-summary` command is retained
-as a human convenience alias; the MCP surface only exposes `get_usage_snapshot`
-and the opt-in `get_usage_summary` tool.
+as a human convenience alias; MCP also exposes `get_capacity_summary` as a
+compatibility alias for `get_usage_summary`.
 
 The default buffer is `policy.min_remaining_percent = 10`. If any reported
 quota window is below that buffer, the provider row is marked unusable for the
@@ -294,9 +308,9 @@ Use `--absolute-command` if the MCP host does not inherit your shell `PATH`.
 Verified per-host steps live in [docs/mcp-clients.md](docs/mcp-clients.md).
 Team templates: [.cursor/mcp.json.example](.cursor/mcp.json.example),
 [.mcp.json.example](.mcp.json.example), and [docs/examples/README.md](docs/examples/README.md).
-MCP Registry draft metadata: [server.json](server.json). It intentionally omits
-package entries until `agentpool-cli` exists on PyPI. Release checklist:
-[docs/release.md](docs/release.md).
+MCP Registry metadata: [server.json](server.json). It advertises the
+`agentpool-cli` PyPI package and should be bumped with each release. Release
+checklist: [docs/release.md](docs/release.md).
 Provider setup guides:
 [Cursor](docs/setup-cursor.md),
 [Cursor Agent CLI](docs/setup-cursor-cli.md),
@@ -316,7 +330,7 @@ Then use tools for live operations. The user-owned preferences file also shows
 up through `agentpool preferences` and `get_delegation_preferences()`. It may
 say to use your native subagent system instead of AgentPool for some tasks. The
 default MCP toolset is deliberately small: inventory, usage snapshot, usage
-summary, provider models, preferences, spawn, observe, send, interrupt,
+summary, provider models, preferences, spawn, observe, poll, send, interrupt,
 collect, artifact manifest, transcript paging, and terminate. Add opt-in
 toolsets with `agentpool mcp --toolsets default,stats,sessions,leases,worktrees`.
 

{agentpool_cli-0.1.11 → agentpool_cli-0.1.13}/README.md

@@ -19,7 +19,7 @@ informed, never automatic.
 
 The v0.1 alpha posture is conservative:
 
-- tmux is the required runtime.
+- tmux is the default runtime; Terminal Control is optional when configured.
 - Provider selection is explicit; `provider=auto` is rejected.
 - Usage/capacity summaries are confidence-tagged and keyed by provider id.
 - CodexBar and ccusage are optional usage helpers when installed or configured.
@@ -30,8 +30,10 @@ The v0.1 alpha posture is conservative:
 
 - Python 3.11 or newer.
 - tmux on `PATH`.
+- Optional: `termctrl` on `PATH` for the Terminal Control runtime.
 - Git for worktree isolation and diff collection.
-- macOS or Linux. Windows is not a v0.1 target except through WSL-like shells.
+- macOS or Linux for live runtimes. Windows is not a v0.1 target except through
+  WSL-like shells.
 
 ## Install
 
@@ -51,9 +53,21 @@ agentpool setup codex
 agentpool doctor --deep --privacy
 ```
 
-The `agentpool-cli` package installs on macOS, Linux, and Windows, but the
-runtime requires `tmux`, so the supported runtime is macOS or Linux (Windows
-via WSL).
+The `agentpool-cli` package installs on macOS, Linux, and Windows, but live
+terminal runtimes are supported on macOS or Linux (Windows via WSL).
+
+Optional Terminal Control config:
+
+```yaml
+runtime:
+  default: tmux
+  terminal_control:
+    enabled: true
+    binary: termctrl
+    session_prefix: agentpool
+    cols: 120
+    rows: 36
+```
 
 Install from source:
 
@@ -113,7 +127,7 @@ agentpool spawn \
   --task "Inspect the project and ask one clarifying question." \
   --isolation read_only
 
-agentpool observe <session-id> --wait-for completed,error,question,approval_prompt --timeout 120 --json
+agentpool observe <session-id> --wait-for completed,error,question,approval_prompt --timeout 60 --json
 agentpool send <session-id> "Continue with the smallest useful check."
 agentpool artifacts <session-id> --json
 agentpool transcript <session-id> --tail-lines 80 --json
@@ -164,8 +178,8 @@ agentpool usage-summary --refresh --backend ccusage --provider claude-code --jso
 ordered and it is not a recommendation list. Each row includes `usable`,
 `unusable_reason`, quota windows, confidence, age/staleness, and reset timing when
 the provider exposes it. The older CLI `capacity-summary` command is retained
-as a human convenience alias; the MCP surface only exposes `get_usage_snapshot`
-and the opt-in `get_usage_summary` tool.
+as a human convenience alias; MCP also exposes `get_capacity_summary` as a
+compatibility alias for `get_usage_summary`.
 
 The default buffer is `policy.min_remaining_percent = 10`. If any reported
 quota window is below that buffer, the provider row is marked unusable for the
@@ -273,9 +287,9 @@ Use `--absolute-command` if the MCP host does not inherit your shell `PATH`.
 Verified per-host steps live in [docs/mcp-clients.md](docs/mcp-clients.md).
 Team templates: [.cursor/mcp.json.example](.cursor/mcp.json.example),
 [.mcp.json.example](.mcp.json.example), and [docs/examples/README.md](docs/examples/README.md).
-MCP Registry draft metadata: [server.json](server.json). It intentionally omits
-package entries until `agentpool-cli` exists on PyPI. Release checklist:
-[docs/release.md](docs/release.md).
+MCP Registry metadata: [server.json](server.json). It advertises the
+`agentpool-cli` PyPI package and should be bumped with each release. Release
+checklist: [docs/release.md](docs/release.md).
 Provider setup guides:
 [Cursor](docs/setup-cursor.md),
 [Cursor Agent CLI](docs/setup-cursor-cli.md),
@@ -295,7 +309,7 @@ Then use tools for live operations. The user-owned preferences file also shows
 up through `agentpool preferences` and `get_delegation_preferences()`. It may
 say to use your native subagent system instead of AgentPool for some tasks. The
 default MCP toolset is deliberately small: inventory, usage snapshot, usage
-summary, provider models, preferences, spawn, observe, send, interrupt,
+summary, provider models, preferences, spawn, observe, poll, send, interrupt,
 collect, artifact manifest, transcript paging, and terminate. Add opt-in
 toolsets with `agentpool mcp --toolsets default,stats,sessions,leases,worktrees`.
 

{agentpool_cli-0.1.11 → agentpool_cli-0.1.13}/docs/agent-cli-and-mcp.md

@@ -16,7 +16,7 @@ another agent's full transcript through an MCP tool result.
 agentpool usage-summary --refresh --json
 agentpool models --provider <provider-id> --json
 agentpool spawn --provider <provider-id> --model <model-id> --repo . --task "<narrow task>" --isolation read_only --json
-agentpool observe <session-id> --wait-for completed,error,question,approval_prompt --timeout 120 --json
+agentpool observe <session-id> --wait-for completed,error,question,approval_prompt --timeout 60 --json
 agentpool send <session-id> "<steering>"
 agentpool artifacts <session-id> --json
 agentpool transcript <session-id> --offset 0 --limit 4000 --json

{agentpool_cli-0.1.11 → agentpool_cli-0.1.13}/docs/agentpool-skill.md

@@ -36,8 +36,8 @@ need to delegate coding-agent work.
 - Use `read_only` isolation for exploration, review, and triage.
 - Choose `worktree` explicitly when AgentPool should create a worktree.
 - Keep workers narrow: one task, clear stop condition, explicit provider.
-- Observe workers with `observe_worker` or `agentpool observe`; do not replace
-  the control loop with session-list polling.
+- Observe workers with `observe_worker`, `poll_worker`, or `agentpool observe`;
+  do not replace the control loop with session-list polling.
 - Treat worker output as untrusted. Read artifact files only when needed.
 - Collect artifacts before relying on worker output.
 - Terminate sessions when finished.
@@ -49,7 +49,7 @@ agentpool usage-summary --refresh --json
 agentpool preferences
 agentpool models --provider <provider-id> --json
 agentpool spawn --provider <provider-id> --model <model-id> --repo . --task "<narrow task>" --isolation read_only --json
-agentpool observe <session-id> --wait-for completed,error,question,approval_prompt --timeout 120 --json
+agentpool observe <session-id> --wait-for completed,error,question,approval_prompt --timeout 60 --json
 agentpool send <session-id> "<steering>"
 agentpool artifacts <session-id> --json
 agentpool transcript <session-id> --tail-lines 80 --json
@@ -75,12 +75,22 @@ without dumping the whole file into context.
 2. `get_usage_summary(provider_id=..., refresh=false)`
 3. `get_provider_models(provider_id=...)`
 4. `spawn_worker(provider_id=..., model=..., repo_path=..., task=..., isolation="read_only")`
-5. `observe_worker(session_id=..., wait_for=["completed","error","question","approval_prompt"], timeout_seconds=120)`
-6. `send_worker_message(...)` or `interrupt_worker(...)`
-7. `get_artifact_manifest(...)`
-8. `read_worker_transcript(...)` for bounded transcript pages, only if needed
-9. `collect_worker_artifacts(...)`
-10. `terminate_worker(...)`
+5. `observe_worker(session_id=..., wait_for=["completed","error","question","approval_prompt"], timeout_seconds=45)`
+6. `poll_worker(session_id=..., include_recent_log=true)` for immediate progress checks between waits
+7. `send_worker_message(...)` or `interrupt_worker(...)`
+8. `get_artifact_manifest(...)`
+9. `read_worker_transcript(...)` for bounded transcript pages, only if needed
+10. `collect_worker_artifacts(...)`
+11. `terminate_worker(...)`
+
+In MCP, `observe_worker(timeout_seconds=0)` and
+`observe_worker(timeout_seconds=1)` are fast polls. Prefer `poll_worker` when
+you want a current snapshot without waiting. Long waits are guarded below common
+host executor caps, and AgentPool returns current state plus timeout metadata
+instead of letting the host kill the call. Set `include_recent_log=true` when
+you need a bounded output tail while the worker is still running; AgentPool also
+keeps `summary.partial.md` in the artifact manifest for salvageable progress
+notes before final `summary.md` or `result.md` exists.
 
 Use opt-in MCP toolsets for extra surfaces:
 

{agentpool_cli-0.1.11 → agentpool_cli-0.1.13}/docs/architecture.md

@@ -15,7 +15,8 @@ Core modules:
 - `models.py`: Pydantic response and persistence models.
 - `config.py`: `~/.agentpool/config.yaml` loader and defaults.
 - `providers/`: conservative CLI detection and launch templates.
-- `runtimes/tmux.py`: required v0.1 runtime.
+- `runtimes/tmux.py`: default worker runtime.
+- `runtimes/terminal_control.py`: optional named-session Terminal Control runtime.
 - `store.py`: SQLite sessions, events, usage snapshots, artifacts, and leases.
 - `session_manager.py`: policy enforcement and worker lifecycle orchestration.
 - `mcp_server.py`: MCP tools and resources.

{agentpool_cli-0.1.11 → agentpool_cli-0.1.13}/docs/examples.md

@@ -20,7 +20,7 @@ agentpool spawn \
 Observe and steer:
 
 ```bash
-agentpool observe <session-id> --wait-for question,completed,error --timeout 120 --json
+agentpool observe <session-id> --wait-for question,completed,error --timeout 60 --json
 agentpool send <session-id> "Inspect migrations first. Stay read-only."
 agentpool session show <session-id> --json
 agentpool collect <session-id> --json

{agentpool_cli-0.1.11 → agentpool_cli-0.1.13}/docs/install.md

@@ -1,16 +1,17 @@
 # Install AgentPool
 
 AgentPool is a local CLI and MCP server. Install it on the same machine where
-your coding-agent CLIs are logged in and where `tmux` is available.
+your coding-agent CLIs are logged in and where your selected runtime is
+available.
 
 ## Requirements
 
 - Python 3.11 or newer.
-- `tmux` on `PATH`.
+- `tmux` on `PATH` for the default runtime.
+- Optional: `termctrl` on `PATH` for the Terminal Control runtime.
 - Git on `PATH`.
-- macOS or Linux for the runtime. The `agentpool-cli` package installs on
-  Windows too, but `tmux` is not native there, so Windows is supported only
-  through WSL.
+- macOS or Linux for live runtimes. The `agentpool-cli` package installs on
+  Windows too, but terminal runtimes are supported only through WSL.
 
 ## Install From PyPI
 

{agentpool_cli-0.1.11 → agentpool_cli-0.1.13}/docs/mcp-clients.md

@@ -260,8 +260,9 @@ Once connected, a good first prompt is:
 Read agentpool://skill.md and agentpool://preferences.md. Then call
 get_delegation_preferences(), get_usage_summary(refresh=false),
 get_provider_models(), and get_inventory() before
-spawning any workers. After spawn_worker, use observe_worker for the worker
-control loop; do not poll get_session/list_sessions as a substitute. Use
+spawning any workers. After spawn_worker, use observe_worker for waits and
+poll_worker for immediate progress checks; do not poll get_session/list_sessions
+as a substitute. Use
 read_worker_transcript with offset/limit only when you need bounded transcript
 pages. If Codex shows update or hook-review startup menus, answer the numbered
 menu choice explicitly before observing again. Choose providers explicitly; do

{agentpool_cli-0.1.11 → agentpool_cli-0.1.13}/docs/mcp-tools.md

@@ -32,11 +32,14 @@ The default toolset is the smallest worker lifecycle surface:
 
 - `get_inventory`
 - `get_usage_summary`
+- `get_capacity_summary`
 - `get_usage_snapshot`
+- `get_cached_usage_snapshot`
 - `get_provider_models`
 - `get_delegation_preferences`
 - `spawn_worker`
 - `observe_worker`
+- `poll_worker`
 - `send_worker_message`
 - `interrupt_worker`
 - `collect_worker_artifacts`
@@ -44,11 +47,10 @@ The default toolset is the smallest worker lifecycle surface:
 - `read_worker_transcript`
 - `terminate_worker`
 
-Removed MCP aliases:
+Compatibility aliases:
 
-- `send_message`
-- `get_capacity_summary`
-- `get_cached_usage_snapshot`
+- `get_capacity_summary`: alias for `get_usage_summary`
+- `get_cached_usage_snapshot`: alias for `get_usage_snapshot(refresh=false)`
 
 Use `get_usage_summary(refresh=false)` for the compact provider-capacity view.
 Use `get_usage_snapshot(refresh=false)` only when you need raw snapshots; cached
@@ -84,12 +86,29 @@ Recommended loop:
 1. `get_usage_summary(provider_id=..., refresh=false)`
 2. `get_provider_models(provider_id=...)`
 3. `spawn_worker(provider_id=..., model=..., repo_path=..., task=..., isolation="read_only")`
-4. `observe_worker(session_id=..., wait_for=["completed","error","question","approval_prompt"], timeout_seconds=120)`
+4. `observe_worker(session_id=..., wait_for=["completed","error","question","approval_prompt"], timeout_seconds=45)`
 5. `send_worker_message(...)` or `interrupt_worker(...)` when needed
-6. `get_artifact_manifest(...)`
-7. `read_worker_transcript(session_id=..., offset=..., limit=...)` only when a bounded transcript page is needed
-8. `collect_worker_artifacts(...)`
-9. `terminate_worker(...)` when the session is no longer useful
+6. `poll_worker(session_id=..., include_recent_log=true)` for immediate progress checks between waits
+7. `get_artifact_manifest(...)`
+8. `read_worker_transcript(session_id=..., offset=..., limit=...)` only when a bounded transcript page is needed
+9. `collect_worker_artifacts(...)`
+10. `terminate_worker(...)` when the session is no longer useful
+
+`observe_worker` accepts both event names and terminal state names in
+`wait_for`, so `["completed"]` and `["COMPLETED"]` both work. Unknown
+`wait_for` values are rejected instead of silently becoming a poll. In MCP, long
+observe waits are guarded below common host executor caps; if a larger
+`timeout_seconds` is requested, AgentPool returns a timeout payload with
+`requested_timeout_seconds`, `effective_timeout_seconds`, and the current worker
+metadata rather than letting the outer host drop the tool call. Passing
+`timeout_seconds=0` or `1` is treated as a fast poll. Prefer `poll_worker` when
+you explicitly want "show me the current frame now."
+
+When `include_recent_log=true`, `observe_worker` and `poll_worker` include a
+bounded recent screen/log tail in `worker_output`, even before final artifacts
+exist. Each observe also refreshes `summary.partial.md` in the artifact
+manifest. `summary.md` and `result.md` remain final-result files and are written
+when a result marker is found or when artifacts are collected.
 
 `spawn_worker.task` must be the concrete delegated instruction. Placeholder text
 such as `Improve documentation in @filename` is rejected.
@@ -107,9 +126,10 @@ user's provider config.
 
 ## Output Detail And Lockdown
 
-Worker text is untrusted model output. `observe_worker` and
+Worker text is untrusted model output. `observe_worker`, `poll_worker`, and
 `collect_worker_artifacts` default to `detail="summary"`, which returns state,
-readiness metadata, and artifact paths without inline worker text.
+readiness metadata, and artifact paths without inline worker text unless an
+observe/poll call explicitly asks for `include_recent_log=true`.
 
 Use `detail="excerpt"` or `detail="full"` only when the coordinator needs text
 inline. Inline worker text is wrapped in a random per-call delimiter:

{agentpool_cli-0.1.11 → agentpool_cli-0.1.13}/docs/model-catalog.md

@@ -23,7 +23,7 @@ Each provider entry can describe:
 - `model_selection`: non-flag pinning mode, such as Droid runtime settings.
 - `reasoning_effort_arg`: the CLI flag used by providers with native effort
   selection, such as Claude Code `--effort` or Droid `--reasoning-effort`.
-- `submit_keys`: tmux keys needed for interactive submit quirks.
+- `submit_keys`: runtime keys needed for interactive submit quirks.
 - `quirks`: operator-facing notes for harness behavior.
 
 ## Override

{agentpool_cli-0.1.11 → agentpool_cli-0.1.13}/docs/provider-adapters.md

@@ -8,7 +8,8 @@ Default real-provider behavior is intentionally limited:
 - Probe version with a short timeout.
 - Report auth as `unknown` unless a safe probe exists.
 - Report usage as `unknown` unless a safe explicit probe exists.
-- Launch through tmux using the provider's normal CLI command.
+- Launch through the selected runtime using the provider's normal CLI command;
+  tmux is the default runtime and Terminal Control is optional.
 - Pin a requested model only when the provider exposes a safe process-local
   mechanism for it.
 
@@ -50,12 +51,17 @@ Model pinning behavior:
   `--reasoning-effort`, which AgentPool forwards when requested.
 - AgentPool does not edit or persist provider credentials or user defaults.
 
-Tmux submit behavior:
+Runtime submit behavior:
 
 - Most providers submit pasted text with terminal Enter.
 - `codex-cli` submits with `C-m` in tmux.
+- `claude-code` receives an explicit `Enter` after pasted steering so
+  multi-line messages ending in a newline are submitted instead of left in the
+  composer.
 - Short menu selections such as startup trust or update prompts still use normal
   Enter so they do not go through provider composer shortcuts.
+- Terminal Control maps the same submit keys onto its named key vocabulary, for
+  example `C-m` to `enter` and `C-c` to `ctrl-c`.
 
 Fake providers under `src/agentpool/fixtures/fake_agents/` are the executable contract for v0.1.
 They are packaged with AgentPool so `agentpool smoke --provider fake-question`

{agentpool_cli-0.1.11 → agentpool_cli-0.1.13}/docs/quickstart.md

@@ -20,8 +20,8 @@ agentpool init
 agentpool doctor --deep --privacy
 ```
 
-`doctor` confirms `tmux`, SQLite, and the artifact root, and lists each
-provider's install/auth state.
+`doctor` confirms runtime availability, SQLite, and the artifact root, and lists
+each provider's install/auth state.
 
 ## 2. Smoke test without a real provider
 
@@ -54,7 +54,7 @@ agentpool spawn --provider <provider-id> --repo . \
 The result includes a top-level `session_id`. Drive it with:
 
 ```bash
-agentpool observe <session-id> --wait-for question,completed,error --timeout 120 --json
+agentpool observe <session-id> --wait-for question,completed,error --timeout 60 --json
 agentpool send <session-id> "Continue with the smallest useful check." --json
 agentpool session show <session-id> --json
 agentpool collect <session-id> --json

{agentpool_cli-0.1.11 → agentpool_cli-0.1.13}/docs/release.md

@@ -7,6 +7,7 @@ metadata, and onboarding verification.
 
 ```bash
 .venv/bin/python -m pytest -q
+.venv/bin/python scripts/check_release_metadata.py
 agentpool models validate --path src/agentpool/provider_model_catalog.json --json
 agentpool config validate --json
 agentpool smoke --provider fake-question --repo . --json
@@ -27,12 +28,13 @@ Update together before tagging:
 - `pyproject.toml` `[project].version`
 - `src/agentpool/__init__.py` `__version__`
 - `server.json` top-level `version`
-- `server.json` `packages[].version` after public PyPI publishing is enabled
+- `server.json` `packages[].version`
 - [CHANGELOG.md](../CHANGELOG.md)
 
 The public PyPI distribution name is `agentpool-cli`. The installed console
-command remains `agentpool`. Do not publish a package named `agentpool`; that
-PyPI name belongs to another project.
+command remains `agentpool`, with an `agentpool-cli` console-script alias for
+`uvx agentpool-cli mcp`. Do not publish a package named `agentpool`; that PyPI
+name belongs to another project.
 
 ## Tag and publish
 
@@ -51,18 +53,37 @@ The [release workflow](../.github/workflows/release.yml) will:
    repository variable `PUBLISH_TO_PYPI` is set to `true` and the `pypi`
    deployment environment is configured. Otherwise the publish job is skipped.
 
-See [Manual one-time setup](#manual-one-time-pypi-setup) below for the steps
-that must be done on PyPI before the publish job can succeed.
+Keep the external setup in
+[PyPI and GitHub trusted-publisher state](#pypi-and-github-trusted-publisher-state)
+in place before tagging.
 
-## Manual one-time PyPI setup
+## PyPI and GitHub trusted-publisher state
 
-These steps are done once, outside CI, before the first automated publish:
+The `agentpool-cli` project is already live on PyPI, and the `v0.1.12`
+distributions were uploaded using Trusted Publishing from
+`sidduHERE/agentpool` via `.github/workflows/release.yml` and the `pypi`
+environment.
 
-1. Reserve / claim the `agentpool-cli` project name on PyPI.
-2. Configure a PyPI **Trusted Publisher (pending publisher)** pointing at this
-   GitHub repository, the `release.yml` workflow, and the `pypi` environment.
-3. Create a GitHub **`pypi` deployment environment** on the repo.
-4. Set the repository variable `PUBLISH_TO_PYPI=true` to arm the publish job.
+Keep these exact external settings:
+
+- PyPI project: `agentpool-cli`
+- PyPI trusted publisher owner: `sidduHERE`
+- PyPI trusted publisher repository: `agentpool`
+- PyPI trusted publisher workflow: `release.yml`
+- PyPI trusted publisher environment: `pypi`
+- GitHub deployment environment: `pypi`
+- GitHub repository variable: `PUBLISH_TO_PYPI=true`
+
+The publish job has `id-token: write` only on the PyPI job and no PyPI token
+secret is required.
+
+Recommended GitHub hardening:
+
+- Add required reviewers to the `pypi` deployment environment so a tag push does
+  not immediately publish without a human release approval.
+- If deployment branch/tag rules are enabled, allow release tags matching `v*`;
+  do not restrict the environment to branch deployments only.
+- Keep PyPI API tokens out of GitHub secrets for this project.
 
 ## Post-release smoke
 
@@ -87,25 +108,16 @@ uv tool install --force --python python3.11 \
 
 ## MCP Registry
 
-Do not publish [server.json](../server.json) to the MCP Registry until
-`agentpool-cli` is available from PyPI. The committed metadata intentionally
-omits `packages` so it does not advertise an unavailable install method. After
-the first successful PyPI publish, add a `pypi` package entry that maps registry
-consumers to:
-
-```json
-{ "command": "agentpool", "args": ["mcp"] }
-```
-
-Bump `server.json` `packages[].version` together with the project version from
-then on.
+[server.json](../server.json) now advertises the published `agentpool-cli` PyPI
+package. Keep its top-level `version` and `packages[].version` aligned with
+`pyproject.toml` and `src/agentpool/__init__.py`.
 
 Host-specific install UX remains in [docs/mcp-clients.md](mcp-clients.md).
 
 Publish sequence:
 
 1. Publish `agentpool-cli` to PyPI with Trusted Publishing (the release job).
-2. Add the `packages` entry to `server.json`.
+2. Keep [server.json](../server.json) pointed at the same published version.
 3. Keep the README metadata comment `mcp-name: io.github.sidduHERE/agentpool`.
    The namespace casing must match the GitHub username exactly (`sidduHERE`, not
    `sidduhere`); the registry grants `io.github.<login>/*` with the login's

{agentpool_cli-0.1.11 → agentpool_cli-0.1.13}/docs/setup-codex.md

@@ -101,7 +101,7 @@ Read agentpool://skill.md. Then call
 get_usage_summary(provider_id="codex-cli", refresh=false),
 get_provider_models(provider_id="codex-cli"), and get_inventory before spawning
 any workers. After spawn_worker, use
-observe_worker for the worker control loop. If Codex shows an update prompt,
+observe_worker for waits and poll_worker for immediate progress checks. If Codex shows an update prompt,
 send menu choice 2 to skip. If Codex shows a directory trust prompt, send an
 empty submitted message only when trusting that directory is acceptable. If it
 shows hook review, send menu choice 3 unless the user approved trusting hooks.

{agentpool_cli-0.1.11 → agentpool_cli-0.1.13}/docs/usage-detection.md

@@ -7,8 +7,9 @@ Allowed confidence values include `official`, `local_cli`, `local_config`, `prov
 Live probes are only run by explicit usage requests. Inventory remains non-invasive and reports whether a provider supports an explicit usage probe. Successful explicit probes are persisted to SQLite; `agentpool usage --cached`, `get_usage_summary(refresh=false)`, and `get_usage_snapshot(refresh=false)` read the latest persisted snapshots without refreshing providers. MCP refreshes do not run interactive provider TUI probes, because those can interfere with the host agent that is calling AgentPool. CLI callers can use `--no-interactive` on `usage` / `usage-summary`, or set `AGENTPOOL_NO_INTERACTIVE_USAGE=1`, to take the same headless posture. MCP refreshes are also bounded by a short server-side budget; if a provider is slow, AgentPool returns a partial response with an unknown row instead of holding the MCP connection open.
 
 `agentpool usage-summary` returns a `providers` map keyed by provider id. The
-CLI `capacity-summary` command is a human convenience alias; MCP does not expose
-a capacity alias. Each row includes `usable`, `unusable_reason`, `stale`, and
+CLI `capacity-summary` command and MCP `get_capacity_summary` tool are
+compatibility aliases for `usage-summary` / `get_usage_summary`. Each row
+includes `usable`, `unusable_reason`, `stale`, and
 `age_seconds`. `stale` is informational age metadata only; it does not by
 itself make a provider unusable. `usable` is derived from install/auth status,
 provider usage status, confidence, and reported quota windows. The default