Files
roboco/docs/rag/workflows/a2a-collaboration.md
T
879afc14a4 Board Program LEARN context, ruff 0.16, and verb-rejection observability (#700)
* fix(board): LEARN decisions name the item, not its per-cycle index

A cycle's reject reasons are rendered into the NEXT cycle's exploration
prompt, but the ref recorded alongside each reason was the item's stored
id (item-0/item-1) — a per-cycle index that means something different
every cycle and appears nowhere the explorer can resolve. The reason
survived the loop; what it was about did not.

Record the item's title instead, via a shared learn_ref() helper (falls
back to the id when title-less, and reads target_task_title for Scales,
whose items name the live task they mutate).

* chore(lint): satisfy ruff 0.16 — keyword-only signatures and markdown formatting

The dev toolchain resolved ruff 0.16.0, which stabilises PLR0917 (too many
positional arguments) and formats python code blocks inside markdown. Both
fired repo-wide and neither had anything to do with the code they flagged.

- 36 signatures gain a `*` so their tail arguments are keyword-only, and
  the 104 call sites that passed them positionally are converted. mypy was
  the safety net for the static ones; the full suite caught nine more that
  only bind at runtime (the MCP tool functions, whose real callers already
  pass named JSON arguments).
- 28 markdown files reformatted by 0.16's code-block formatter.
- One RUF036 (`None` mid-union) autofixed in the GitLab provider.

* fix(gateway): log the reason when a verb rejects

A rejected envelope rides an HTTP 200, its body is never logged, and there
is no trace table — so in the access log a verb an agent could not satisfy
looks identical to one that worked. On 2026-07-25 four Board Programs
(Periscope, Sentinel, Scales, Barfly) each POSTed their propose verb three
or four times, persisted nothing, and left their exploration tasks PENDING;
the reason was unrecoverable afterwards, from the logs or from the agents'
own transcripts.

Log error/message/remediate/missing plus the calling agent at
envelope_to_response — the one chokepoint every v1 flow and do route
returns through. Success envelopes stay silent.

---------

Co-authored-by: Renn F <rennf93@users.noreply.github.com>
2026-07-26 15:07:28 +02:00

2.4 KiB

A2A Collaboration Workflow

Overview

Agents collaborate directly through the dm content tool on the roboco-do MCP server, with read_a2a to read what you were sent. There is no roboco_agent_* or roboco_a2a_* tool — A2A is just dm + read_a2a.

Key: A2A is about existing tasks, NOT task creation. Pass the task_id you're collaborating on so the message is linked to it.

Flow

1. Reach out → dm(recipient, text, task_id)      for a direct message
2. Receive   → read_a2a()                        to read incoming A2A message bodies
             → notify_list() / notify_get(id)    to read your notify inbox

Direct Messages (same cell only)

# Direct A2A inside your cell (same team — no policy gate)
dm(
    recipient="be-qa",
    text="Quick sanity check on the rate-limit boundary before I open the PR?",
    task_id="<task>",
)

Cross-cell dm is denied by policy. If you need something from another cell, route it through your Cell PM via escalate_up(task_id, reason) — the PM coordinates across cells.

The CEO

CEO-initiated conversations may arrive and are replied to in-thread like any other unread A2A.

Receiving Messages — read_a2a

Your claim briefing surfaces incoming A2A under unread_a2a — each entry shows the sender and a preview of their latest message. To read the full bodies (and clear them):

read_a2a()  # -> {"messages": [{from_agent, content, created_at}, ...]}

read_a2a() returns only INCOMING messages (never your own sends) and marks them read. It also clears i_am_idle()'s unread-A2A soft-block.

Task Creation Rules

Only PMs create tasks (via the delegate verb). Regular agents cannot create work from a dm.

If a conversation surfaces work that needs a new task:

  1. Escalate to your Cell PM: escalate_up(task_id, reason="Needs a subtask for X")
  2. The PM decides whether to delegate a subtask

Permissions

Most roles can dm (same-cell) and read incoming messages with read_a2a, plus check their notify inbox with notify_list / notify_get.

The Auditor is a silent observer of peers: it never initiates dm to another agent and has no notify. It does carry dm/read_a2a so it can read and reply in-thread when the CEO opens a DM with it — that reply path is stateful (not gated by the peer-initiation rule).

Only PMs and the Board can send ack-required notify signals; regular agents use dm only.