Harbor as of 2026-10-05: the coordination hub for the Open Sea's decomp team

Messages, threads, acknowledgements, claims, bands, tasks, review items and
stop flags for a team of AI agents; the chat page and board; host hooks and
MCP server; autopilot; tools/watch.py. Put under version control after the
weekend's changes so they have history and can be rolled back. Local only.

harbor/texts.py

@@ -0,0 +1,46 @@
+"""Text shared by the installer, the MCP server and restart prompts: how hosts run Harbor and the team protocol."""
+import shutil
+import sys
+
+from .projects import PACKAGE_ROOT
+
+
+def launcher(python=None):
+    """The command prefix hosts run: forward slashes and no quotes, so cmd, PowerShell and bash all accept it."""
+    run = (PACKAGE_ROOT / "run.py").as_posix()
+    if python:
+        return python.split() + ["-B", run]
+    if shutil.which("py"):
+        return ["py", f"-{sys.version_info.major}.{sys.version_info.minor}", "-B", run]
+    return [sys.executable.replace("\\", "/"), "-B", run]
+
+
+def protocol(agent, project_name=None, cmd=None):
+    """The rules every agent follows, in the second person, as a markdown section."""
+    run = " ".join(cmd or launcher())
+    scope = (f"in the Harbor project `{project_name}`" if project_name else
+             "inside a Harbor project (a `.harbor.toml` file exists in your working folder or a parent folder)")
+    return f"""## Team coordination (Harbor)
+
+When you work {scope}, you are **{agent}** on a team coordinated through Harbor (MCP tools named `brief`,
+`send`, ..., or the command `{run} --as {agent} <command>`):
+- Start every session with `brief`. If it reports a STOP for you, do not start or continue work; only the human
+  can resume. If the brief shows a handoff you wrote earlier, continue from it.
+- Take work with `next` (the highest-priority task you may do) instead of waiting to be told. Mark it done, or
+  `submit` the commit to the review queue with the task number; landing closes the task.
+- Keep `doing` current: one line the whole team sees. Do not post progress reports to everyone.
+- Talk in threads named after the subject (fn:0x..., commit:..., task:N, item:N) and answer with `reply`. Write
+  plain sentences; never run words and numbers together. A correction uses `--corrects <id>`.
+- Acknowledge with `ack` (and a short note) instead of sending "received". Ask for an acknowledgement, or set a
+  due time, only when you need the answer.
+- Claim before you work on something that is not a task (`claim fn:0x...`); release when done.
+- Before you start on a function or file, read `context <scope>`: notes, earlier threads and review items about it.
+  Record what you learn that others will need (an ABI detail, a trap, a decision) with `note <scope> "..."`.
+- Before asking the team about a function, `who <scope>` tells you who has worked on it. When a thread grows long,
+  write a short `summarize <thread> "..."` (decided, open, who does what next).
+- When you need the human to decide, use `ask` with short options ("Yes", "No", "Later"); they answer with one tap.
+  Never post a bare question to everyone and wait.
+- When you wait on someone, use `wait_for_messages` (`wait`) instead of sleeping or polling.
+- Before your session ends or your allowance runs out, write a `handoff`: what you finished, what is in flight,
+  what comes next. It is shown to you (and the team) at the start of the next session.
+"""