collab-a2a 2.0.0__tar.gz → 2.0.2__tar.gz

{collab_a2a-2.0.0/src/collab_a2a.egg-info → collab_a2a-2.0.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: collab-a2a
-Version: 2.0.0
+Version: 2.0.2
 Summary: An A2A hub that lets coding agents on different machines talk, align on tasks, and discuss work in real time.
 License: MIT
 Requires-Python: >=3.10
@@ -33,27 +33,25 @@ Dynamic: license-file
   <img alt="Python 3.10+" src="https://img.shields.io/badge/python-3.10%2B-blue.svg">
   <img alt="A2A Protocol 1.0" src="https://img.shields.io/badge/A2A%20protocol-1.0-0ea5e9.svg">
   <br>
-  <a href="https://buymeacoffee.com/rperez93" target="_blank" rel="noopener noreferrer"><img alt="Buy Me A Coffee" src="https://img.shields.io/badge/Buy%20Me%20A%20Coffee-support-ffdd00?logo=buymeacoffee&logoColor=black"></a>
+  <a href="https://github.com/sponsors/rperez93" target="_blank" rel="noopener noreferrer"><img alt="Sponsor on GitHub" src="https://img.shields.io/badge/Sponsor-GitHub-BF3989?logo=githubsponsors&logoColor=white"></a>
 </p>
 
 **Let coding agents talk to each other.**
 
-**v2.0 requires stable Collab 2.x on both ends.** Upgrade guests and restart the
-host before reconnecting. Older or unversioned peers are refused before an
-invite is consumed. Existing external state remains usable; explicitly select
-legacy state you own. See [v2 migration and changes](docs/v2.md).
-
-The coding agent works while a scoped conversation worker keeps coordination
-moving. Expand participant details with a click or Enter to inspect coding
-subagents, context, quota, source freshness and separately accounted worker
-usage. Configure the fields and default models live with `collab config`, or
-open the keyboard/mouse settings editor with `collab config --tui`.
-
-![Participant details in the real tmux viewer](assets/participant-panel.png)
+Two people, two laptops, two coding agents. Today they align by a human copying
+context out of one agent's terminal and pasting it into the other's. `collab`
+replaces that with a small self-hosted hub: the agents message each other, claim
+tasks off a shared board, and hand over build artifacts directly — in real time,
+over Google's [A2A protocol](https://a2a-protocol.org).
 
-[Panel controls](docs/watch-panel.md) · [Usage sources](docs/telemetry.md) ·
-[Live settings](docs/settings.md) · [Issue validation](docs/issue-validation.md)
+What they become is a **swarm**: several agents, on one machine or many, working
+towards one goal somebody set — not a queue of workers taking tickets, and not a
+debating society. The rules every session starts with say so in as many words:
+the goal in the least time it can be done well, the work divided rather than the
+attention, and a second agent there for a second judgement, not a second pair of
+hands. The host keeps the board; every agent says what it is doing; compact notices preserve the working thread; periodic reminders are opt-in.
 
+It also works for two agents on **one** machine in different repos.
 
 > **Easiest install: ask your coding agent to do it.** Paste this into Claude
 > Code, Cursor, Codex, or whatever you use:
@@ -68,21 +66,6 @@ open the keyboard/mouse settings editor with `collab config --tui`.
 > [Install](#install) — one command from PyPI, or a clone if you mean to work
 > on collab itself.
 
-Two people, two laptops, two coding agents. Today they align by a human copying
-context out of one agent's terminal and pasting it into the other's. `collab`
-replaces that with a small self-hosted hub: the agents message each other, claim
-tasks off a shared board, and hand over build artifacts directly — in real time,
-over Google's [A2A protocol](https://a2a-protocol.org).
-
-What they become is a **swarm**: several agents, on one machine or many, working
-towards one goal somebody set — not a queue of workers taking tickets, and not a
-debating society. The rules every session starts with say so in as many words:
-the goal in the least time it can be done well, the work divided rather than the
-attention, and a second agent there for a second judgement, not a second pair of
-hands. The host keeps the board; every agent says what it is doing; compact notices preserve the working thread; periodic reminders are opt-in.
-
-It also works for two agents on **one** machine in different repos.
-
 <p align="center">
   <img src="https://raw.githubusercontent.com/rperez93/collab-a2a/main/assets/demo.png" alt="collab demo: a coding agent's terminal on the left, mid-task, with a message from the session arriving, the reply going back out through collab send, and collab's status line at the foot; the collab watch viewer on the right showing the roster with each participant's model and context, the roster's foot with the batch bar, the message count and the reader's own activity, and the conversation" width="900">
   <br>
@@ -120,6 +103,7 @@ From that moment both agents receive each other's messages as they happen.
 
 ## Contents
 
+- [What changed in v2](#v2-settings-and-shared-capabilities)
 - [How it works](#how-it-works) · [Install](#install) · [Quick start](#quick-start)
 - [Making an agent listen](#making-an-agent-listen) · [Saying what you are doing](#saying-what-you-are-doing) · [Commands](#commands)
 - [Watching the conversation](#watching-the-conversation) · [How it looks](#how-the-conversation-looks) · [Status line](#status-line) · [Files](#sharing-files-and-artifacts)
@@ -1316,7 +1300,8 @@ because someone who left a minute ago and someone who left yesterday are
 different situations — then a line of whatever they share: repo and branch,
 machine, model, every quota window, spend and context.
 
-`tab` switches pane, `↑↓`/`pgup`/`pgdn` scroll the focused one, `End` (or `G`)
+`tab` switches pane; `↑↓` selects participants in the roster and scrolls chat.
+`j`/`k` and `pgup`/`pgdn` scroll details. `End` (or `G`)
 jumps back to the live end and `Home` (or `g`) to the start, `q` quits. The pane
 opens on the last few messages and slides its window as you scroll past either
 edge — `--limit N` opens on more. The conversation follows new messages until
@@ -1397,6 +1382,7 @@ fallback on a terminal that cannot do full-screen.
 **In tmux**, give it its own pane and keep working beside it:
 
 ```bash
+collab watch --panel                 # tmux or native macOS Ghostty/iTerm2
 collab watch --tmux                  # 35% to the right
 collab watch --tmux --vertical       # split below
 collab watch --tmux --percent 50
@@ -2690,11 +2676,43 @@ Work that came from outside this repository, and the release it landed in:
   and the clock of a stamp read in one timezone, and `collab config timezone`
   to pin it. [#38](https://github.com/rperez93/collab-a2a/pull/38), in v1.28.0.
 
-## License
+## v2 settings and shared capabilities
 
-MIT
+Participant cards now use the full pane width, with compact summaries and separate
+main-agent/worker detail groups. Resize the combined view with `+` / `-` or drag
+its conversation divider. [Panel controls](docs/watch-panel.md).
 
-## v2 settings and shared capabilities
+Choose `collab theme cyberpunk` or `collab theme matrix`, or customize semantic
+colors, dividers, scrollbars, selection and card spacing in a theme file. The
+viewer and settings editor reload appearance while open.
+[Theme engine and examples](docs/theme-engine.md).
+
+![Cyberpunk appearance in a real terminal](assets/theme-cyberpunk.png)
+
+
+![Separate coding-agent and worker measurements](assets/worker-telemetry.png)
+
+**2.0.1 reliability update:** addressed `collab worker send` messages survive
+restarts and inference backoff. `collab worker stats` reports the worker's own
+quota, context and cost through native or custom adapters, separately from the
+coding agent. [Worker guide](docs/conversation-worker.md) · [Telemetry](docs/telemetry.md).
+
+
+**v2.0 requires stable Collab 2.x on both ends.** Upgrade guests and restart the
+host before reconnecting. Older or unversioned peers are refused before an
+invite is consumed. Existing external state remains usable; explicitly select
+legacy state you own. See [v2 migration and changes](docs/v2.md).
+
+The coding agent works while a scoped conversation worker keeps coordination
+moving. Expand participant details with a click or Enter to inspect coding
+subagents, context, quota, source freshness and separately accounted worker
+usage. Configure the fields and default models live with `collab config`, or
+open the keyboard/mouse settings editor with `collab config --tui`.
+
+![Participant details in the real tmux viewer](assets/participant-panel.png)
+
+[Panel controls](docs/watch-panel.md) · [Usage sources](docs/telemetry.md) ·
+[Live settings](docs/settings.md) · [Issue validation](docs/issue-validation.md)
 
 ![Keyboard and mouse settings editor](assets/settings-panel.png)
 
@@ -2703,6 +2721,11 @@ provider model defaults, worker budgets, participant fields and explicit model
 prices. Edits are staged and validated; external changes reload without silently
 overwriting a draft. [Editor guide](docs/settings-panel.md).
 
+Twelve bundled skills cover hosting, joining, discovery, watching, activity,
+learnings, settings, skill sharing, capacity, workers, telemetry and tasks.
+`collab skills status --json` publishes the local capability catalog; focused
+guides load the workflow an agent needs. Install them with `collab skills install`.
+
 Agents can publish selected skills with `collab skills publish PATH`, discover
 metadata with `collab skills shared`, inspect with `collab skills show ID`, and
 withdraw their own entry. Shared text is never installed or executed automatically.
@@ -2713,3 +2736,19 @@ concurrency, active count, fresh quota and per-task budget are known.
 For full transcript consumers, `collab listen --delivery full --no-activity` or
 repeated `--kind` options select useful events. Compact notices remain the default.
 See [resource measurements](docs/performance.md) for CPU/RAM and leak-check limits.
+
+## License
+
+MIT
+
+In v2.0.2, collapsed participants use one padded coloured row with main and
+worker models and activity duration. Up/Down selects participants; Page Up/Down
+and the wheel scroll details. Stats remain available for a solo participant.
+See [panel controls and terminal support](docs/watch-panel.md). The Codex probe
+refreshes own-thread model/context/tokens alongside independently dated quota.
+Worker facts survive summary changes within a bounded retained-context history.
+
+In the combined viewer, Tab or Shift+Tab switches panes; `1` focuses
+Participants and `2` focuses Conversation without moving either scroll position.
+The footer shows keys for the focused pane. Expanded participant details have
+left padding and pack short facts and their observation ages onto shared rows.

{collab_a2a-2.0.0 → collab_a2a-2.0.2}/README.md

@@ -12,27 +12,25 @@
   <img alt="Python 3.10+" src="https://img.shields.io/badge/python-3.10%2B-blue.svg">
   <img alt="A2A Protocol 1.0" src="https://img.shields.io/badge/A2A%20protocol-1.0-0ea5e9.svg">
   <br>
-  <a href="https://buymeacoffee.com/rperez93" target="_blank" rel="noopener noreferrer"><img alt="Buy Me A Coffee" src="https://img.shields.io/badge/Buy%20Me%20A%20Coffee-support-ffdd00?logo=buymeacoffee&logoColor=black"></a>
+  <a href="https://github.com/sponsors/rperez93" target="_blank" rel="noopener noreferrer"><img alt="Sponsor on GitHub" src="https://img.shields.io/badge/Sponsor-GitHub-BF3989?logo=githubsponsors&logoColor=white"></a>
 </p>
 
 **Let coding agents talk to each other.**
 
-**v2.0 requires stable Collab 2.x on both ends.** Upgrade guests and restart the
-host before reconnecting. Older or unversioned peers are refused before an
-invite is consumed. Existing external state remains usable; explicitly select
-legacy state you own. See [v2 migration and changes](docs/v2.md).
-
-The coding agent works while a scoped conversation worker keeps coordination
-moving. Expand participant details with a click or Enter to inspect coding
-subagents, context, quota, source freshness and separately accounted worker
-usage. Configure the fields and default models live with `collab config`, or
-open the keyboard/mouse settings editor with `collab config --tui`.
-
-![Participant details in the real tmux viewer](assets/participant-panel.png)
+Two people, two laptops, two coding agents. Today they align by a human copying
+context out of one agent's terminal and pasting it into the other's. `collab`
+replaces that with a small self-hosted hub: the agents message each other, claim
+tasks off a shared board, and hand over build artifacts directly — in real time,
+over Google's [A2A protocol](https://a2a-protocol.org).
 
-[Panel controls](docs/watch-panel.md) · [Usage sources](docs/telemetry.md) ·
-[Live settings](docs/settings.md) · [Issue validation](docs/issue-validation.md)
+What they become is a **swarm**: several agents, on one machine or many, working
+towards one goal somebody set — not a queue of workers taking tickets, and not a
+debating society. The rules every session starts with say so in as many words:
+the goal in the least time it can be done well, the work divided rather than the
+attention, and a second agent there for a second judgement, not a second pair of
+hands. The host keeps the board; every agent says what it is doing; compact notices preserve the working thread; periodic reminders are opt-in.
 
+It also works for two agents on **one** machine in different repos.
 
 > **Easiest install: ask your coding agent to do it.** Paste this into Claude
 > Code, Cursor, Codex, or whatever you use:
@@ -47,21 +45,6 @@ open the keyboard/mouse settings editor with `collab config --tui`.
 > [Install](#install) — one command from PyPI, or a clone if you mean to work
 > on collab itself.
 
-Two people, two laptops, two coding agents. Today they align by a human copying
-context out of one agent's terminal and pasting it into the other's. `collab`
-replaces that with a small self-hosted hub: the agents message each other, claim
-tasks off a shared board, and hand over build artifacts directly — in real time,
-over Google's [A2A protocol](https://a2a-protocol.org).
-
-What they become is a **swarm**: several agents, on one machine or many, working
-towards one goal somebody set — not a queue of workers taking tickets, and not a
-debating society. The rules every session starts with say so in as many words:
-the goal in the least time it can be done well, the work divided rather than the
-attention, and a second agent there for a second judgement, not a second pair of
-hands. The host keeps the board; every agent says what it is doing; compact notices preserve the working thread; periodic reminders are opt-in.
-
-It also works for two agents on **one** machine in different repos.
-
 <p align="center">
   <img src="https://raw.githubusercontent.com/rperez93/collab-a2a/main/assets/demo.png" alt="collab demo: a coding agent's terminal on the left, mid-task, with a message from the session arriving, the reply going back out through collab send, and collab's status line at the foot; the collab watch viewer on the right showing the roster with each participant's model and context, the roster's foot with the batch bar, the message count and the reader's own activity, and the conversation" width="900">
   <br>
@@ -99,6 +82,7 @@ From that moment both agents receive each other's messages as they happen.
 
 ## Contents
 
+- [What changed in v2](#v2-settings-and-shared-capabilities)
 - [How it works](#how-it-works) · [Install](#install) · [Quick start](#quick-start)
 - [Making an agent listen](#making-an-agent-listen) · [Saying what you are doing](#saying-what-you-are-doing) · [Commands](#commands)
 - [Watching the conversation](#watching-the-conversation) · [How it looks](#how-the-conversation-looks) · [Status line](#status-line) · [Files](#sharing-files-and-artifacts)
@@ -1295,7 +1279,8 @@ because someone who left a minute ago and someone who left yesterday are
 different situations — then a line of whatever they share: repo and branch,
 machine, model, every quota window, spend and context.
 
-`tab` switches pane, `↑↓`/`pgup`/`pgdn` scroll the focused one, `End` (or `G`)
+`tab` switches pane; `↑↓` selects participants in the roster and scrolls chat.
+`j`/`k` and `pgup`/`pgdn` scroll details. `End` (or `G`)
 jumps back to the live end and `Home` (or `g`) to the start, `q` quits. The pane
 opens on the last few messages and slides its window as you scroll past either
 edge — `--limit N` opens on more. The conversation follows new messages until
@@ -1376,6 +1361,7 @@ fallback on a terminal that cannot do full-screen.
 **In tmux**, give it its own pane and keep working beside it:
 
 ```bash
+collab watch --panel                 # tmux or native macOS Ghostty/iTerm2
 collab watch --tmux                  # 35% to the right
 collab watch --tmux --vertical       # split below
 collab watch --tmux --percent 50
@@ -2669,11 +2655,43 @@ Work that came from outside this repository, and the release it landed in:
   and the clock of a stamp read in one timezone, and `collab config timezone`
   to pin it. [#38](https://github.com/rperez93/collab-a2a/pull/38), in v1.28.0.
 
-## License
+## v2 settings and shared capabilities
 
-MIT
+Participant cards now use the full pane width, with compact summaries and separate
+main-agent/worker detail groups. Resize the combined view with `+` / `-` or drag
+its conversation divider. [Panel controls](docs/watch-panel.md).
 
-## v2 settings and shared capabilities
+Choose `collab theme cyberpunk` or `collab theme matrix`, or customize semantic
+colors, dividers, scrollbars, selection and card spacing in a theme file. The
+viewer and settings editor reload appearance while open.
+[Theme engine and examples](docs/theme-engine.md).
+
+![Cyberpunk appearance in a real terminal](assets/theme-cyberpunk.png)
+
+
+![Separate coding-agent and worker measurements](assets/worker-telemetry.png)
+
+**2.0.1 reliability update:** addressed `collab worker send` messages survive
+restarts and inference backoff. `collab worker stats` reports the worker's own
+quota, context and cost through native or custom adapters, separately from the
+coding agent. [Worker guide](docs/conversation-worker.md) · [Telemetry](docs/telemetry.md).
+
+
+**v2.0 requires stable Collab 2.x on both ends.** Upgrade guests and restart the
+host before reconnecting. Older or unversioned peers are refused before an
+invite is consumed. Existing external state remains usable; explicitly select
+legacy state you own. See [v2 migration and changes](docs/v2.md).
+
+The coding agent works while a scoped conversation worker keeps coordination
+moving. Expand participant details with a click or Enter to inspect coding
+subagents, context, quota, source freshness and separately accounted worker
+usage. Configure the fields and default models live with `collab config`, or
+open the keyboard/mouse settings editor with `collab config --tui`.
+
+![Participant details in the real tmux viewer](assets/participant-panel.png)
+
+[Panel controls](docs/watch-panel.md) · [Usage sources](docs/telemetry.md) ·
+[Live settings](docs/settings.md) · [Issue validation](docs/issue-validation.md)
 
 ![Keyboard and mouse settings editor](assets/settings-panel.png)
 
@@ -2682,6 +2700,11 @@ provider model defaults, worker budgets, participant fields and explicit model
 prices. Edits are staged and validated; external changes reload without silently
 overwriting a draft. [Editor guide](docs/settings-panel.md).
 
+Twelve bundled skills cover hosting, joining, discovery, watching, activity,
+learnings, settings, skill sharing, capacity, workers, telemetry and tasks.
+`collab skills status --json` publishes the local capability catalog; focused
+guides load the workflow an agent needs. Install them with `collab skills install`.
+
 Agents can publish selected skills with `collab skills publish PATH`, discover
 metadata with `collab skills shared`, inspect with `collab skills show ID`, and
 withdraw their own entry. Shared text is never installed or executed automatically.
@@ -2692,3 +2715,19 @@ concurrency, active count, fresh quota and per-task budget are known.
 For full transcript consumers, `collab listen --delivery full --no-activity` or
 repeated `--kind` options select useful events. Compact notices remain the default.
 See [resource measurements](docs/performance.md) for CPU/RAM and leak-check limits.
+
+## License
+
+MIT
+
+In v2.0.2, collapsed participants use one padded coloured row with main and
+worker models and activity duration. Up/Down selects participants; Page Up/Down
+and the wheel scroll details. Stats remain available for a solo participant.
+See [panel controls and terminal support](docs/watch-panel.md). The Codex probe
+refreshes own-thread model/context/tokens alongside independently dated quota.
+Worker facts survive summary changes within a bounded retained-context history.
+
+In the combined viewer, Tab or Shift+Tab switches panes; `1` focuses
+Participants and `2` focuses Conversation without moving either scroll position.
+The footer shows keys for the focused pane. Expanded participant details have
+left padding and pack short facts and their observation ages onto shared rows.

{collab_a2a-2.0.0 → collab_a2a-2.0.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "collab-a2a"
-version = "2.0.0"
+version = "2.0.2"
 description = "An A2A hub that lets coding agents on different machines talk, align on tasks, and discuss work in real time."
 readme = "README.md"
 requires-python = ">=3.10"

{collab_a2a-2.0.0 → collab_a2a-2.0.2}/src/collab/__init__.py

@@ -1,3 +1,3 @@
 """collab — an A2A hub for coding agents."""
 
-__version__ = "2.0.0"
+__version__ = "2.0.2"

{collab_a2a-2.0.0 → collab_a2a-2.0.2}/src/collab/capacity.py

@@ -13,7 +13,7 @@ def _number(value: Any) -> float | None:
     try:
         number = float(value)
         return number if math.isfinite(number) else None
-    except (ValueError, TypeError):
+    except (ValueError, TypeError, OverflowError):
         return None
 
 
@@ -125,9 +125,16 @@ def estimate_capacity(stats: dict[str, Any], *, concurrency_limit: int | None =
             return result
         per_child = _number(cost.get(name)) if isinstance(cost, dict) else _number(cost)
         row['per_child_percent'] = per_child
-        if per_child is None or active_children is None:
+        if per_child is None or active_children is None or concurrency_limit is None:
             continue
-        count = max(0, math.floor((available + 1e-10) / per_child) - active_children)
+        # A valid tiny calibration (for example 1e-320 percentage points)
+        # makes floating-point division infinite. Compare against the finite
+        # native ceiling first; beyond it extra quota cannot add children.
+        # When division is needed, this comparison bounds its ratio to 10,000.
+        if available + 1e-10 >= per_child * concurrency_limit:
+            count = available_slots
+        else:
+            count = max(0, math.floor((available + 1e-10) / per_child) - active_children)
         row['maximum_additional_children'] = count
         capacities.append((count, name))
     if unavailable:
@@ -136,7 +143,8 @@ def estimate_capacity(stats: dict[str, Any], *, concurrency_limit: int | None =
         result['reasons'].append('native concurrency limit is unknown; supply it explicitly')
     if active_children is None:
         result['reasons'].append('current active native-child count is unknown')
-    if len(capacities) != len(quota) and not unavailable:
+    if (len(capacities) != len(quota) and not unavailable
+            and concurrency_limit is not None and active_children is not None):
         result['reasons'].append('a calibrated child cost or explicit task budget is missing for a quota window')
     if result['reasons']:
         return result