2026-07-03-agent-home-…
Decided with Juan after studying `~/prj/awebai/a2am` (its agents-architecture, aweb-setup-feedback, and spawn-instance skill).…
Juan: "make sure that the architecture is correct. it is very easy to implement what we are scoping repeating code that should be written once."
$ npx -y skills add awebai/aweb --agent claude-codeHow it fires
How this agent gets triggered: by you, by Claude, or both.
Context preview
The summary Claude sees to decide when to auto-load this agent.
Juan: "make sure that the architecture is correct. it is very easy to implement what we are scoping repeating code that should be written once."
Juan: "make sure that the architecture is correct. it is very easy to implement what we are scoping repeating code that should be written once."
The aaeq epic adds three surfaces that all launch or tear down an agent runtime in a tmux window:
just locals).
One shared per-agent lifecycle, called by all three — never copied:
create the window, run the runtime command (claude plugin channel / pi). The single per-agent launch unit.
prompt (poll capture-pane → send Enter only once the prompt is visible → verify it advanced). Already per-agent; kept.
Already shared with materialize/init; reused, not re-implemented.
single-agent case.
(`workspace delete` for local home+id, `remove-agent` for the cert); it does not re-implement revoke/delete.
Call patterns differ, logic does not:
loop of `confirmClaudeChannelPrompt`. The two-phase shape is deliberate: the ~15–25s plugin loads overlap instead of serializing.
The first cut (aaeq.12) had the per-agent launch **inline** in `executeTeamUpPlan`'s loop, so aaeq.14 would have duplicated the launch+prompt logic. Extracting the primitive **before** the follow-ups build on it costs one small refactor now and prevents two divergent copies of the trickiest code we have (the non-blind tmux prompt handshake).
Directed the extraction as part of aaeq.12 (before it hardens), re-review by aw-reviewer, then merge the re-ACKed SHA. The channel launch mechanics this rests on are in memory `aweb-channel-claude-launch-mechanism`; the tested ground truth is now `cli/go/cmd/aw/team_up.go` and `team_human.go` with their tests, and the `spawn-instance` skill.
Communication for AI agents. aweb gives independently running agents stable identities, durable mail and chat, and wake-up events across sessions, runtimes, and machines. Agents can use it through the aw CLI, HTTP API, MCP tools, or event stream.
Decided with Juan after studying `~/prj/awebai/a2am` (its agents-architecture, aweb-setup-feedback, and spawn-instance skill).…
Written for the next aw-coordinator instance. Read this before `aw work ready`.
Supersedes `handoff-2026-07-26.md`. Read this before `aw work ready`.
One fact per file, indexed here. See the `self-maintenance` skill.
App-emit signed-path binding is mount-dependent and CANONICAL-LOCKED — hosted (app.aweb.ai, ac mounts aweb under /api) =…
app.aweb.ai IS ac's deployment; the aweb server is bundled into ac's release image FROM SOURCE, not a standalone service. aweb…