dialectic-respond

Start by checking the local workspace cache inventory.

Use the shared helper:

• scripts/client/session_cache.py list

If the newest entry contains parent_agent_id, use it as expected-actor lineage context, not as proof by itself.

If you need to bind to that exact cached UUID before responding, use an advanced proof-owned rebind only with a matching current in-process token. Without proof, call /governance-start to mint fresh. Declare parent_agent_id only when this process is genuinely inheriting work from a finished predecessor.

Do not rely on weaker fingerprint/session fallback when dialectic authorization depends on the exact actor.

Then inspect the target session with:

• dialectic(action='get', session_id='<session_id>')

Use the returned required_role, required_agent_id, current_agent_can_submit, and recommended_action fields to decide the next step.

Workflow:

1. If phase is thesis: the paused agent should respond via dialectic(action='thesis', ...).
2. If phase is antithesis and the current bound agent is the assigned reviewer: respond via dialectic(action='antithesis', ...).
3. If phase is antithesis and the current bound agent should answer instead of the assigned reviewer: an authenticated operator caller can use dialectic(action='antithesis', ..., take_over_if_requested=true, takeover_reason='...'); anyone else gets TAKEOVER_FORBIDDEN. Otherwise wait for the assigned reviewer, or have the operator or the current reviewer use dialectic(action='reassign', session_id=..., new_reviewer_id=...).
4. If phase is synthesis: respond via dialectic(action='synthesis', ...).
5. If the session is already resolved/failed/escalated, report that instead of trying to write.

When reporting back:

• state the current phase
• state who is allowed to act next
• if you submitted a response, summarize the position you took
• if you had to take over reviewer ownership, say so explicitly