Born Ready

one lifecycle hook that ends mount-order guessing

Amplifier Core · shipped v1.4.1

The composition gap

Modules mount in ordered phases — but each one wakes up before the coordinator is whole

Amplifier builds every session from independent modules mounted in a fixed order. A module's mount() runs while the coordinator is only partially composed, so cross-module wiring never had a safe place to happen.

This is a real, shipped change — here's the proof.

1

orchestrator

2

context

3

provider

4

tool

5

hook

Proof · it shipped

on_session_ready() landed in microsoft/amplifier-core PR #63

Merged 2026-04-25 by Diego Colombo. The PR bumped the kernel from 1.3.3 to 1.4.0 and reached consumers via v1.4.1 — a change of +1598 / −25 across 24 files.

So why did a hook like this need to exist?

#63

PR in amplifier-core

1.4.1

shipped to consumers

+1598

lines added (−25)

24

files changed

The BUT · defensive init

Untrusted ordering forced dual-path — and even triple-path — init

Because mount() sees only a partial coordinator, module authors checked for a capability in both mount() (not yet available) and in a request handler (too late to configure) — and stacked triple-path event discovery to survive the ordering they couldn't trust.

The fix is a single guaranteed moment after composition.

Dual-path init — check for a capability in both mount() and a request handler
Triple-path event discovery — three fallback mechanisms because no single path is reliable at mount time

The mechanism

A new Phase 6 fires on_session_ready(coordinator) exactly once

It runs after every module across every phase has finished mount(), in initialize_session — after the hooks loop and before session:fork. Every module gets one moment when the coordinator is fully composed.

And adopting it carries almost no risk.

async def on_session_ready(coordinator):
    # the coordinator is now
    # fully composed — wire
    # cross-module deps here
    ...

# Phase 6 dispatch, then:
# emit session:fork

Safe to adopt

Callbacks run in mount order and are non-fatal per module

Phase 6 walks the queue in mount registration order. A failing callback is caught and logged (WARNING, with exc_info) and emits a module:on_session_ready_failed event — while the remaining callbacks still run.

The contract behind it is enforced, not hoped for.

Mount registration order — callbacks fire in the order modules mounted
Failure isolation — one failure is logged, the rest still run
Observability — emits module:on_session_ready_failed with {module_id, error}

The turn · enforced contract

The contract is validated across all five module-type validators

check_on_session_ready() validates that the hook is async and takes a coordinator argument, wired into the hook, tool, orchestrator, provider and context validators — backed by 55 new/updated tests with 0 regressions.

Which is exactly what lets authors drop the old anti-pattern.

5

module-type validators

55

new / updated tests

0

regressions

The payoff

One guaranteed instant when the coordinator is whole

Every module now gets a fully-composed moment to wire its dependencies — which is exactly what lets it drop the defensive dual-path anti-pattern documented in CONTRACTS.md. Guessing mount order is replaced by a reliable post-composition hook.

The takeaway

Give a composed system one post-composition hook, and order-guessing disappears

The pattern generalizes beyond Amplifier: when independent parts compose in phases, one explicit "everything is ready now" moment retires the defensive, order-guessing init that fragile ordering otherwise forces.

Sources

Research Methodology

Data as of: April 2026

Feature status: Shipped and released — merged to microsoft/amplifier-core main 2026-04-25 (PR #63); tags v1.4.0 and v1.4.1 exist; present in installed amplifier_core 1.6.0. Scope is Python-only.

Correct repository: microsoft/amplifier-core (the kernel). The initially-suggested michaeljabbour/amplifier-collab-core is a module, not the kernel, and returned 0 matches for on_session_ready.

Research performed:

PR discovery: gh pr list --repo microsoft/amplifier-core --state all --search on_session_ready --json number,title,mergedAt,author
PR detail & diff: gh pr view 63 --repo microsoft/amplifier-core --json additions,deletions,files,commits,body,mergedAt,author
Ship version: gh pr view 68 --repo microsoft/amplifier-core --json body
Dispatch code: sed -n '244,272p' amplifier_core/_session_init.py (installed v1.6.0)
Events: gh api repos/microsoft/amplifier-core/contents/crates/amplifier-core/src/events.rs
Contract: gh api repos/microsoft/amplifier-core/contents/CONTRACTS.md | base64 -d
Repo confirmation: git remote -v; grep -n on_session_ready

Primary contributors: Diego Colombo (colombod) — author of PR #63 (feature + contract). Brian Krabach (bkrabach) — author of follow-up docs PR #68.

Gaps / excluded: Two speculative topic claims (a "7% observability graph" loss and Digital Twin Universe end-to-end verification) were independently re-checked, found unsupported, and deliberately omitted from this deck.