[pauleisman-a11y]

Summary

Adds a GoldenDirectiveTermination termination condition to
autogen_agentchat.conditions, implementing the Golden Directive
relationship score as a run-stopping condition:

Y = 2(A x E) / (A + E)

A (agent health) and E (environment health) are supplied by
callables, following the pattern of FunctionalTermination, so the
caller owns how A and E are measured. The condition terminates the
run when the current Y falls below a caller-set min_score, and
exposes the last computed score as last_score.

The Golden Directive (https://thegoldendirective.pages.dev/) is a
candidate training reward proposed by Paul Eisman: A is an agent's
net contribution-health score, floored at 0, designed to be grounded
in resource ledgers; E is the health of everything not-A, against a
healthy baseline. If either A or E reaches 0, Y reaches 0.

Scope, stated plainly:

• This condition only computes Y from supplied values and stops a
  run below a threshold. It does not measure A or E.
• It does not implement the spec's full-versus-cropped reporting
  rule, its reversal (E = A Y / (2A - Y)), or its proposed
  fence-patch guardrail.
• The Directive itself is a proposal, not a deployed or validated
  training method; its results so far are pilot-scale toy
  gridworld runs, reported on the site.

Tests

Added to tests/test_termination_condition.py:

• Canonical numeric checks from the spec: 8,8 -> 8; 8,7 -> 7.4667
  (about 7.5); 2,3 -> 2.4; A=0 or E=0 -> 0; 1,10 -> 1.8182;
  10,9 -> 9.4737 (A rising enough can still raise Y while E falls);
  negative inputs floored at 0.
• Behaviour: no termination while Y is at or above the threshold,
  termination with a StopMessage when E drops to 0, reset clears
  the state, async health providers are awaited.

Local run: the full test_termination_condition.py file passes,
15 tests (13 existing + 2 new).

Files changed

• python/packages/autogen-agentchat/src/autogen_agentchat/conditions/_terminations.py (+119)
• python/packages/autogen-agentchat/src/autogen_agentchat/conditions/__init__.py (+2)
• python/packages/autogen-agentchat/tests/test_termination_condition.py (+58)

[pauleisman-a11y]

@microsoft-github-policy-service agree

[pauleisman-a11y]

@microsoft-github-policy-service agree

[pauleisman-a11y]

@microsoft-github-policy-service agree