code-coordinator 0.5.109__tar.gz → 0.5.111__tar.gz

{code_coordinator-0.5.109 → code_coordinator-0.5.111}/.github/workflows/test.yml

@@ -56,35 +56,44 @@ jobs:
   # 2026-08-08 stall (`parked — CI running: windows`): each drive parked on it
   # and handed off, which is how five PRs piled up unmerged.
   #
-  # What that latency bought per-PR was ~nothing. The job is deliberately red
-  # and non-blocking (see below), so nothing gates on its result; its output
-  # is one number (`368 failed, ...`) that only moves when someone works on
-  # #1156. Running it per-merge keeps that baseline just as current.
+  # What that latency bought per-PR was ~nothing while the job was red by
+  # design; running it per-merge keeps the baseline just as current. Now that
+  # it is a real gate (below), per-PR coverage is worth revisiting — and the
+  # right move there is probably to keep it per-PR but shrink it to the import
+  # smoke #1895's own title describes, rather than the full suite.
   #
-  # REVISIT WHEN #1156 LANDS. Once the suite is genuinely green and blocking,
-  # per-PR coverage is worth its duration again — and at that point the right
-  # move is probably to keep it per-PR but shrink it to the import smoke
-  # #1895's own title describes, rather than the full suite.
+  # NO `continue-on-error` — THIS IS A REAL GATE NOW (#2000).
   #
-  # This job is EXPECTED TO BE RED when it lands: `coord.cli` cannot import
-  # on Windows today (`cli.py` -> `coord/interactive.py`'s unconditional
-  # `import fcntl`, a POSIX-only stdlib module). #1156 (CP-1, the import
-  # guards) is what makes this suite green — and #1156 is the one that
-  # removes the `continue-on-error` below. Until then this is a
-  # visible-but-non-blocking signal, not a real gate.
+  # History: when #1895 landed this job this was EXPECTED TO BE RED, because
+  # `coord.cli` could not import on Windows (`cli.py` -> `coord/interactive.py`
+  # unconditionally `import fcntl`, a POSIX-only stdlib module), so both steps
+  # carried `continue-on-error: true` and the job was a visible-but-
+  # non-blocking signal. Two things closed that out:
   #
-  # `continue-on-error` MUST stay on the *step*, not the job (#2026).
-  # At job level it only softens the *workflow run* conclusion; the `windows`
-  # **check** still reports `failure` to the Checks API. GitHub's own branch
-  # protection tolerates that, but `coord merge`'s CI gate does not —
-  # `coord.ci_store._PASSING_CONCLUSIONS` is {success, skipped, neutral} with
-  # no notion of a non-blocking check, so a job-level flag left every
-  # claude-coordinator PR unmergeable and stalled the drive queue overnight
-  # on 2026-08-08 (four finished PRs, everything else green). At step level
-  # the step's failure is recorded but the job exits 0, so the check is
-  # `success` and #1895's actual acceptance criterion — "it does not block
-  # merges while red" — finally holds. The failures stay fully visible in the
-  # job log; that log IS the #1156 specification and the pass/fail baseline.
+  #   1. #1156 (CP-1) landed the import guards, killing the `fcntl` failure.
+  #   2. #2000 put `tzdata; sys_platform == 'win32'` in the BASE `[project]
+  #      dependencies` (NOT a CI-only `pip install`) — Windows ships no IANA
+  #      time-zone database, so `zoneinfo` could not resolve `America/Chicago`
+  #      and the suite died during *collection*. coord/config.py,
+  #      coord/models.py, coord/failure_class.py and coord/machine_pause.py all
+  #      resolve named zones at runtime, so that was a shipped defect for every
+  #      `pip install code-coordinator` Windows user, not a test artefact.
+  #      tests/test_client_base_install.py pins it in the base deps.
+  #
+  # With both fixed the flags are gone and a Windows failure reddens this job
+  # for real. tests/test_windows_job_is_a_real_gate_2000.py is the regression
+  # guard against them creeping back.
+  #
+  # Why re-arming this cannot re-block the repo: `if: github.event_name ==
+  # 'push'` (below) means the job is SKIPPED on every `pull_request` event, and
+  # `skipped` is already in `coord.ci_store._PASSING_CONCLUSIONS` — so a red
+  # Windows run shows up on main, never on a PR's merge gate. That guard is
+  # what makes removing `continue-on-error` safe; it did not exist during the
+  # 2026-08-08 stall, when a *job-level* flag (#2026) left every
+  # claude-coordinator PR unmergeable overnight because the `windows` **check**
+  # still reported `failure` to the Checks API even though the workflow run's
+  # conclusion was softened. If this job is ever made per-PR again, that
+  # interaction is the thing to re-check first.
   #
   # Python 3.12 only, deliberately not the 3.12/3.13 matrix `test` runs above
   # — the matrix axis there is Python versions on Linux; this job's axis is
@@ -102,18 +111,15 @@ jobs:
       - uses: actions/setup-python@v5
         with:
           python-version: "3.12"
-      # Both steps carry the flag, not just pytest: a Windows-only *install*
-      # failure would otherwise redden the job and re-block the whole repo
-      # exactly as a test failure did. Until #1156, this job's contract is
-      # "always report, never block" — and a failed install is every bit as
-      # visible in the log as a failed test.
+      # #2000: no `continue-on-error` on either step any more — a Windows-only
+      # install failure (a base dep that does not resolve on win32, say) is
+      # exactly as much of a real defect as a test failure, and this job now
+      # exists to fail on it. See the header for why re-arming it is safe.
       - run: pip install -e ".[dev]"
-        continue-on-error: true
       # #2180: same split as the `test` job above — the sealed acceptance
       # suite is gated separately through the #2164 CI wrapper, not swept
       # into this job's ordinary pytest run.
       - run: pytest --ignore=tests/acceptance
-        continue-on-error: true
 
   # #1484: run the full suite with `gh` made unresolvable on PATH.
   #

{code_coordinator-0.5.109 → code_coordinator-0.5.111}/CLAUDE.md

@@ -1,4 +1,4 @@
-# claude-coordinator
+# code-coordinator
 
 CLI tool + per-machine agent server that coordinates Claude Code workers across multiple machines and repos over Tailscale.
 

{code_coordinator-0.5.109 → code_coordinator-0.5.111}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-coordinator
-Version: 0.5.109
+Version: 0.5.111
 Summary: Coordinate Claude Code workers across multiple machines and repos.
 Author-email: John Donaghy <johnfdonaghy@gmail.com>
 License-Expression: LicenseRef-FSL-1.1-MIT
@@ -38,11 +38,11 @@ Requires-Dist: setuptools-scm>=8; extra == "dev"
 Requires-Dist: wheel; extra == "dev"
 Dynamic: license-file
 
-# claude-coordinator
+# code-coordinator
 
 Coordinate a fleet of Claude Code workers — and human-attended interactive `claude` sessions — from a single board.
 
-Claude Code is great at one task at a time. Real projects have dozens of issues, spread across repos, that each want their own session, their own scoping, and their own review. `claude-coordinator` runs many workers in parallel — on one machine or across several over Tailscale — behind a coordinator that picks the model, avoids file conflicts, routes by capability, and drives the full **Work → Test → Review → Merge** pipeline. Each stage can run as a cheap `claude -p` worker *or* as a human-attended interactive session you launch and steer from the board.
+Claude Code is great at one task at a time. Real projects have dozens of issues, spread across repos, that each want their own session, their own scoping, and their own review. `code-coordinator` runs many workers in parallel — on one machine or across several over Tailscale — behind a coordinator that picks the model, avoids file conflicts, routes by capability, and drives the full **Work → Test → Review → Merge** pipeline. Each stage can run as a cheap `claude -p` worker *or* as a human-attended interactive session you launch and steer from the board.
 
 ## The problem
 
@@ -110,7 +110,7 @@ Two rules shape the flow:
 ## Quick demo
 
 ```bash
-pip install claude-coordinator     # from PyPI
+pip install code-coordinator     # from PyPI
 coord init                         # interactive setup: detects repos, writes coordinator.yml
 coord agent &                      # start the agent server (port 7433) — see Quick Start for the systemd setup
 
@@ -139,7 +139,7 @@ Prefer to drive it by hand? Launch any stage as an interactive session from `coo
 ### 1. Install
 
 ```bash
-pip install claude-coordinator
+pip install code-coordinator
 ```
 
 The `coord` CLI is now on your PATH. The same package provides the agent server (`coord agent`), so the coordinator side and the worker side share one install.
@@ -168,7 +168,7 @@ coord agent &     # port 7433; auto-detects the machine from hostname
 For anything beyond a trial, use the installer (systemd user service, auto-restart, survives reboots, separate worker logs):
 
 ```bash
-curl -sSL https://raw.githubusercontent.com/JDonaghy/claude-coordinator/main/install-agent.sh | bash
+curl -sSL https://raw.githubusercontent.com/JDonaghy/code-coordinator/main/install-agent.sh | bash
 ```
 
 The same script sets up remote worker machines — see [Worker node setup](#worker-node-setup).
@@ -549,7 +549,7 @@ The tool encodes a pattern from real multi-agent sessions: **separate the tech l
 
 1. On the new machine, run the installer (venv + systemd service in one shot):
    ```bash
-   curl -sSL https://raw.githubusercontent.com/JDonaghy/claude-coordinator/main/install-agent.sh | bash -s -- --machine <name>
+   curl -sSL https://raw.githubusercontent.com/JDonaghy/code-coordinator/main/install-agent.sh | bash -s -- --machine <name>
    ```
    No git clone needed — `install-agent.sh` pulls from PyPI.
 2. Add the machine to `coordinator.yml` under `machines:` with its Tailscale hostname and capabilities.
@@ -562,9 +562,9 @@ For Tailscale setup, see [tailscale.com/kb](https://tailscale.com/kb/). The agen
 To add a worker machine (no repo checkout needed):
 
 ```bash
-curl -sSL https://raw.githubusercontent.com/JDonaghy/claude-coordinator/main/install-agent.sh | bash
+curl -sSL https://raw.githubusercontent.com/JDonaghy/code-coordinator/main/install-agent.sh | bash
 # or with options:
-curl -sSL https://raw.githubusercontent.com/JDonaghy/claude-coordinator/main/install-agent.sh | bash -s -- --machine myserver --port 7433
+curl -sSL https://raw.githubusercontent.com/JDonaghy/code-coordinator/main/install-agent.sh | bash -s -- --machine myserver --port 7433
 ```
 
 This installs coord, sets up a systemd service with auto-restart, and starts the agent. Then add the machine to `coordinator.yml` and run `coord status` to verify connectivity.

{code_coordinator-0.5.109 → code_coordinator-0.5.111}/README.md

@@ -1,8 +1,8 @@
-# claude-coordinator
+# code-coordinator
 
 Coordinate a fleet of Claude Code workers — and human-attended interactive `claude` sessions — from a single board.
 
-Claude Code is great at one task at a time. Real projects have dozens of issues, spread across repos, that each want their own session, their own scoping, and their own review. `claude-coordinator` runs many workers in parallel — on one machine or across several over Tailscale — behind a coordinator that picks the model, avoids file conflicts, routes by capability, and drives the full **Work → Test → Review → Merge** pipeline. Each stage can run as a cheap `claude -p` worker *or* as a human-attended interactive session you launch and steer from the board.
+Claude Code is great at one task at a time. Real projects have dozens of issues, spread across repos, that each want their own session, their own scoping, and their own review. `code-coordinator` runs many workers in parallel — on one machine or across several over Tailscale — behind a coordinator that picks the model, avoids file conflicts, routes by capability, and drives the full **Work → Test → Review → Merge** pipeline. Each stage can run as a cheap `claude -p` worker *or* as a human-attended interactive session you launch and steer from the board.
 
 ## The problem
 
@@ -70,7 +70,7 @@ Two rules shape the flow:
 ## Quick demo
 
 ```bash
-pip install claude-coordinator     # from PyPI
+pip install code-coordinator     # from PyPI
 coord init                         # interactive setup: detects repos, writes coordinator.yml
 coord agent &                      # start the agent server (port 7433) — see Quick Start for the systemd setup
 
@@ -99,7 +99,7 @@ Prefer to drive it by hand? Launch any stage as an interactive session from `coo
 ### 1. Install
 
 ```bash
-pip install claude-coordinator
+pip install code-coordinator
 ```
 
 The `coord` CLI is now on your PATH. The same package provides the agent server (`coord agent`), so the coordinator side and the worker side share one install.
@@ -128,7 +128,7 @@ coord agent &     # port 7433; auto-detects the machine from hostname
 For anything beyond a trial, use the installer (systemd user service, auto-restart, survives reboots, separate worker logs):
 
 ```bash
-curl -sSL https://raw.githubusercontent.com/JDonaghy/claude-coordinator/main/install-agent.sh | bash
+curl -sSL https://raw.githubusercontent.com/JDonaghy/code-coordinator/main/install-agent.sh | bash
 ```
 
 The same script sets up remote worker machines — see [Worker node setup](#worker-node-setup).
@@ -509,7 +509,7 @@ The tool encodes a pattern from real multi-agent sessions: **separate the tech l
 
 1. On the new machine, run the installer (venv + systemd service in one shot):
    ```bash
-   curl -sSL https://raw.githubusercontent.com/JDonaghy/claude-coordinator/main/install-agent.sh | bash -s -- --machine <name>
+   curl -sSL https://raw.githubusercontent.com/JDonaghy/code-coordinator/main/install-agent.sh | bash -s -- --machine <name>
    ```
    No git clone needed — `install-agent.sh` pulls from PyPI.
 2. Add the machine to `coordinator.yml` under `machines:` with its Tailscale hostname and capabilities.
@@ -522,9 +522,9 @@ For Tailscale setup, see [tailscale.com/kb](https://tailscale.com/kb/). The agen
 To add a worker machine (no repo checkout needed):
 
 ```bash
-curl -sSL https://raw.githubusercontent.com/JDonaghy/claude-coordinator/main/install-agent.sh | bash
+curl -sSL https://raw.githubusercontent.com/JDonaghy/code-coordinator/main/install-agent.sh | bash
 # or with options:
-curl -sSL https://raw.githubusercontent.com/JDonaghy/claude-coordinator/main/install-agent.sh | bash -s -- --machine myserver --port 7433
+curl -sSL https://raw.githubusercontent.com/JDonaghy/code-coordinator/main/install-agent.sh | bash -s -- --machine myserver --port 7433
 ```
 
 This installs coord, sets up a systemd service with auto-restart, and starts the agent. Then add the machine to `coordinator.yml` and run `coord status` to verify connectivity.

{code_coordinator-0.5.109 → code_coordinator-0.5.111}/code_coordinator.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-coordinator
-Version: 0.5.109
+Version: 0.5.111
 Summary: Coordinate Claude Code workers across multiple machines and repos.
 Author-email: John Donaghy <johnfdonaghy@gmail.com>
 License-Expression: LicenseRef-FSL-1.1-MIT
@@ -38,11 +38,11 @@ Requires-Dist: setuptools-scm>=8; extra == "dev"
 Requires-Dist: wheel; extra == "dev"
 Dynamic: license-file
 
-# claude-coordinator
+# code-coordinator
 
 Coordinate a fleet of Claude Code workers — and human-attended interactive `claude` sessions — from a single board.
 
-Claude Code is great at one task at a time. Real projects have dozens of issues, spread across repos, that each want their own session, their own scoping, and their own review. `claude-coordinator` runs many workers in parallel — on one machine or across several over Tailscale — behind a coordinator that picks the model, avoids file conflicts, routes by capability, and drives the full **Work → Test → Review → Merge** pipeline. Each stage can run as a cheap `claude -p` worker *or* as a human-attended interactive session you launch and steer from the board.
+Claude Code is great at one task at a time. Real projects have dozens of issues, spread across repos, that each want their own session, their own scoping, and their own review. `code-coordinator` runs many workers in parallel — on one machine or across several over Tailscale — behind a coordinator that picks the model, avoids file conflicts, routes by capability, and drives the full **Work → Test → Review → Merge** pipeline. Each stage can run as a cheap `claude -p` worker *or* as a human-attended interactive session you launch and steer from the board.
 
 ## The problem
 
@@ -110,7 +110,7 @@ Two rules shape the flow:
 ## Quick demo
 
 ```bash
-pip install claude-coordinator     # from PyPI
+pip install code-coordinator     # from PyPI
 coord init                         # interactive setup: detects repos, writes coordinator.yml
 coord agent &                      # start the agent server (port 7433) — see Quick Start for the systemd setup
 
@@ -139,7 +139,7 @@ Prefer to drive it by hand? Launch any stage as an interactive session from `coo
 ### 1. Install
 
 ```bash
-pip install claude-coordinator
+pip install code-coordinator
 ```
 
 The `coord` CLI is now on your PATH. The same package provides the agent server (`coord agent`), so the coordinator side and the worker side share one install.
@@ -168,7 +168,7 @@ coord agent &     # port 7433; auto-detects the machine from hostname
 For anything beyond a trial, use the installer (systemd user service, auto-restart, survives reboots, separate worker logs):
 
 ```bash
-curl -sSL https://raw.githubusercontent.com/JDonaghy/claude-coordinator/main/install-agent.sh | bash
+curl -sSL https://raw.githubusercontent.com/JDonaghy/code-coordinator/main/install-agent.sh | bash
 ```
 
 The same script sets up remote worker machines — see [Worker node setup](#worker-node-setup).
@@ -549,7 +549,7 @@ The tool encodes a pattern from real multi-agent sessions: **separate the tech l
 
 1. On the new machine, run the installer (venv + systemd service in one shot):
    ```bash
-   curl -sSL https://raw.githubusercontent.com/JDonaghy/claude-coordinator/main/install-agent.sh | bash -s -- --machine <name>
+   curl -sSL https://raw.githubusercontent.com/JDonaghy/code-coordinator/main/install-agent.sh | bash -s -- --machine <name>
    ```
    No git clone needed — `install-agent.sh` pulls from PyPI.
 2. Add the machine to `coordinator.yml` under `machines:` with its Tailscale hostname and capabilities.
@@ -562,9 +562,9 @@ For Tailscale setup, see [tailscale.com/kb](https://tailscale.com/kb/). The agen
 To add a worker machine (no repo checkout needed):
 
 ```bash
-curl -sSL https://raw.githubusercontent.com/JDonaghy/claude-coordinator/main/install-agent.sh | bash
+curl -sSL https://raw.githubusercontent.com/JDonaghy/code-coordinator/main/install-agent.sh | bash
 # or with options:
-curl -sSL https://raw.githubusercontent.com/JDonaghy/claude-coordinator/main/install-agent.sh | bash -s -- --machine myserver --port 7433
+curl -sSL https://raw.githubusercontent.com/JDonaghy/code-coordinator/main/install-agent.sh | bash -s -- --machine myserver --port 7433
 ```
 
 This installs coord, sets up a systemd service with auto-restart, and starts the agent. Then add the machine to `coordinator.yml` and run `coord status` to verify connectivity.

{code_coordinator-0.5.109 → code_coordinator-0.5.111}/code_coordinator.egg-info/SOURCES.txt

@@ -193,10 +193,10 @@ coord/dashboard/webapp/dist/index.html
 coord/dashboard/webapp/dist/manifest.webmanifest
 coord/dashboard/webapp/dist/sw.js
 coord/dashboard/webapp/dist/workbox-e4022e15.js
-coord/dashboard/webapp/dist/assets/Gallery-Cq84itDE.js
+coord/dashboard/webapp/dist/assets/Gallery-BoKs3mpY.js
 coord/dashboard/webapp/dist/assets/Terminal-9CEnUXvW.css
-coord/dashboard/webapp/dist/assets/Terminal-DOXLcMue.js
-coord/dashboard/webapp/dist/assets/index-DHn-RIJ8.js
+coord/dashboard/webapp/dist/assets/Terminal-CEsP08iX.js
+coord/dashboard/webapp/dist/assets/index-CDHPlB6k.js
 coord/dashboard/webapp/dist/assets/index-Dq4kwTdw.css
 coord/dashboard/webapp/dist/assets/workbox-window.prod.es5-BqEJf4Xk.js
 coord/dashboard/webapp/dist/icons/icon-192.png
@@ -745,6 +745,7 @@ tests/test_usage_rollup.py
 tests/test_venv_live_install_2121.py
 tests/test_version_single_source.py
 tests/test_watch.py
+tests/test_windows_job_is_a_real_gate_2000.py
 tests/test_worker_events.py
 tests/test_worker_image_pin_sync.py
 tests/test_worker_safety.py

{code_coordinator-0.5.109 → code_coordinator-0.5.111}/code_coordinator.egg-info/scm_file_list.json

@@ -775,6 +775,7 @@
     "tests/test_venv_live_install_2121.py",
     "tests/test_version_single_source.py",
     "tests/test_watch.py",
+    "tests/test_windows_job_is_a_real_gate_2000.py",
     "tests/test_worker_events.py",
     "tests/test_worker_image_pin_sync.py",
     "tests/test_worker_safety.py",

{code_coordinator-0.5.109 → code_coordinator-0.5.111}/code_coordinator.egg-info/scm_version.json

@@ -1,7 +1,7 @@
 {
-  "tag": "0.5.109",
+  "tag": "0.5.111",
   "distance": 0,
-  "node": "g3cadef6459414c424ff96582a2bebcc8dd3ae08a",
+  "node": "g0bbfb660d44f1e03dcf6d5ded5b11f804d43aae9",
   "dirty": false,
   "branch": "HEAD",
   "node_date": "2026-08-16"

{code_coordinator-0.5.109 → code_coordinator-0.5.111}/coord/__init__.py

@@ -124,13 +124,13 @@ def _resolve_version(names: tuple[str, ...] = CANDIDATE_NAMES) -> str:
     version literal; a release is just ``git tag vX.Y.Z && git push origin
     vX.Y.Z``.
 
-    #2103: resolves through :func:`coord.dist_name.resolve_installed` rather
-    than a single hardcoded distribution name — ``__version__`` is what
-    ``/health``'s ``"version"`` field reports, which is the *only* field
-    ``coord agent update``'s polling loop keys off of to decide a machine
-    "came back". A bare ``claude-coordinator``-only lookup would degrade to
-    ``"0+unknown"`` forever on any machine that has ``code-coordinator``
-    installed instead, faking a fleet-wide "did not come back".
+    #2103/#2106: resolves through :func:`coord.dist_name.resolve_installed`
+    rather than a hardcoded distribution name literal — ``__version__`` is
+    what ``/health``'s ``"version"`` field reports, which is the *only*
+    field ``coord agent update``'s polling loop keys off of to decide a
+    machine "came back". A hardcoded literal here would need editing at the
+    next rename and would degrade to ``"0+unknown"`` on every machine in
+    between, faking a fleet-wide "did not come back".
 
     #2010: for a wheel install that metadata is always correct — it's a
     frozen snapshot of a build that just happened. For an *editable*

{code_coordinator-0.5.109 → code_coordinator-0.5.111}/coord/agent_app.py

@@ -36,14 +36,13 @@ def _agent_pkg_spec() -> str:
     `[server]` extra — a bare upgrade would, on a fresh venv, leave the
     agent without starlette/uvicorn and dead on the next restart.
 
-    #2103: resolved tolerantly against whichever of `code-coordinator` /
-    `claude-coordinator` is currently installed, rather than the old
-    hardcoded `claude-coordinator[server]` — installing the wrong name once
-    the fleet is mid-rename either 404s against PyPI or, worse, silently
-    reinstalls the stale package. Deliberately NOT caught here: if neither
-    name resolves, the caller (`_do_update`'s existing try/except) turns
-    that into an explicit `last_update.json` failure naming both names
-    tried, instead of guessing.
+    #2103/#2106: resolved via `coord.dist_name` rather than a hardcoded
+    `claude-coordinator[server]` — installing the wrong name either 404s
+    against PyPI or, worse, silently reinstalls a stale package.
+    Deliberately NOT caught here: if the name doesn't resolve, the caller
+    (`_do_update`'s existing try/except) turns that into an explicit
+    `last_update.json` failure naming what was expected, instead of
+    guessing.
     """
     return _dist_pkg_spec(extra="server")
 
@@ -82,18 +81,18 @@ def _installed_version() -> str | None:
     process hasn't restarted since the last update" apart from "the update
     never happened".
 
-    #2103: tries `code-coordinator` then falls back to `claude-coordinator`
-    (see ``coord.dist_name``) rather than hardcoding one name — installing
-    under the name this process doesn't query used to make a fully-updated
-    agent report ``None`` here, the exact false negative behind the
-    fleet's most-recurring `✗ did not come back`.
+    #2103/#2106: resolves via ``coord.dist_name`` (see that module) rather
+    than hardcoding the distribution name — installing under a name this
+    process doesn't query used to make a fully-updated agent report
+    ``None`` here, the exact false negative behind the fleet's
+    most-recurring `✗ did not come back`.
 
     This is a *report* site, not an *act* site (contrast `_agent_pkg_spec`,
     which deliberately lets `DistributionNotFoundError` propagate): callers
     already fall back to the literal string `"unknown"` when this returns
-    `None`, so "neither name resolved" isn't silently swallowed end to end
-    — but it's still logged here, naming both names tried, rather than
-    disappearing into a bare `None` with no trace of why.
+    `None`, so a miss isn't silently swallowed end to end — but it's still
+    logged here, naming what was expected, rather than disappearing into a
+    bare `None` with no trace of why.
     """
     try:
         return resolve_installed().version
@@ -414,12 +413,11 @@ def _detect_install_mode() -> tuple[bool, str | None]:
     ``pip install -e .``).  *project_path* is the on-disk source directory for
     editable installs, or *None* for regular (site-packages) installs.
 
-    #2103: ``pip show`` needs an exact distribution name, so this resolves
-    which of `code-coordinator` / `claude-coordinator` is actually installed
-    (see ``coord.dist_name``) first rather than hardcoding one — asking
-    `pip show` for a name nothing is installed under always reports "not
-    editable", which would misreport a real editable install once the
-    fleet's mid-rename.
+    #2103/#2106: ``pip show`` needs an exact distribution name, so this
+    resolves which name is actually installed (see ``coord.dist_name``)
+    first rather than hardcoding one — asking `pip show` for a name
+    nothing is installed under always reports "not editable", which would
+    misreport a real editable install.
     """
     dist_name = resolve_installed_name()
     if dist_name is None:

{code_coordinator-0.5.109 → code_coordinator-0.5.111}/coord/agent_update.py

@@ -538,17 +538,15 @@ def _smoke_check(slot: Path, *, target_version: str | None) -> tuple[bool, str |
     mechanism :func:`_installed_version` uses) so a single subprocess call
     covers both "does it import" and "what version is this."
 
-    #2103: the version print resolves via ``coord.dist_name.resolve_installed``
-    (tries `code-coordinator` then falls back to `claude-coordinator`) rather
-    than a hardcoded ``m.version('claude-coordinator')`` — *slot* was just
-    installed from a pkg spec that itself resolved tolerantly (see
-    ``coord.agent_app._agent_pkg_spec``), so a hardcoded name here would
-    raise inside the new slot's own interpreter the moment that pkg spec
-    picked the other name, failing every smoke check fleet-wide the instant
-    the rename lands. Raising when *neither* name is installed is still the
-    right behavior here (not caught) — that means the fresh install is
-    genuinely broken, which is exactly what a failed smoke check should
-    report.
+    #2103/#2106: the version print resolves via
+    ``coord.dist_name.resolve_installed`` rather than a hardcoded
+    ``m.version('claude-coordinator')`` — *slot* was just installed from a
+    pkg spec that itself resolves via the same module (see
+    ``coord.agent_app._agent_pkg_spec``), so a name mismatch here would
+    raise inside the new slot's own interpreter for no reason. Raising when
+    the name isn't installed at all is still the right behavior here (not
+    caught) — that means the fresh install is genuinely broken, which is
+    exactly what a failed smoke check should report.
     """
     python = slot / "bin" / "python"
     coord_bin = slot / "bin" / "coord"

{code_coordinator-0.5.109 → code_coordinator-0.5.111}/coord/cli.py

@@ -255,21 +255,17 @@ def _warn_if_source_install_drift() -> None:
                 if staleness["days_behind"] is not None
                 else ""
             )
-            # #2103: suggest an upgrade of whichever distribution name is
-            # actually installed — `code-coordinator` since the #2104 rename,
-            # `claude-coordinator` on an agent that has not been updated past
-            # it yet. `_coord.__file__` is already confirmed above to be a
-            # real site-packages install, so one of the two names is
-            # guaranteed to resolve here; the literal fallback only guards a
-            # resolution race between that check and this one. It names the
-            # NEW distribution (#2104): the only way to reach it is "neither
-            # name resolved", and the one name guaranteed to still be
-            # publishable is the one every future release ships under —
-            # `claude-coordinator` is a tombstone that will never gain
-            # another version. `_dist_pkg_spec` is the module-level
-            # `coord.dist_name` import above (real module, independent of
-            # `_coord` here potentially being a test stand-in), so this
-            # doesn't depend on `coord.dist_name` already being cached.
+            # #2103/#2106: suggest an upgrade of the installed distribution
+            # name (`code-coordinator`), resolved via `coord.dist_name`
+            # rather than hardcoded. `_coord.__file__` is already confirmed
+            # above to be a real site-packages install, so resolution is
+            # guaranteed to succeed here; the literal fallback only guards a
+            # resolution race between that check and this one — the only
+            # way to reach it is "the distribution vanished between the two
+            # checks". `_dist_pkg_spec` is the module-level `coord.dist_name`
+            # import above (real module, independent of `_coord` here
+            # potentially being a test stand-in), so this doesn't depend on
+            # `coord.dist_name` already being cached.
             try:
                 install_target = _dist_pkg_spec(extra="server")
             except Exception:  # noqa: BLE001 — best-effort, never break the CLI