P1 — support parameterized user commands, collisions, revocation and atomic hot reload #165

New issue

Closed

opened 2026-08-22 21:54:10 +00:00 by lost-rob0t · 0 comments

lost-rob0t commented

2026-08-22 21:54:10 +00:00

(Migrated from github.com)

Parent epic: #151
Depends on: #164

Goal

Make authored commands first-class parameterized semantic intents and harden their lifecycle under conflicts/provider changes.

Scope

user-defined typed slots reuse #150 validators/dialogue;
trigger templates/aliases are bounded and deterministically parsed;
missing slots ask through the normal clarification state machine;
built-in and user-command collision policy explicit and tested;
command-to-command ambiguity produces clarification/conflict, not first-match accidents;
capability/provider permission revocation takes effect without rewriting stored semantic intent;
disabled/unavailable provider yields typed unavailable/denied result;
atomic hot reload while concurrent turns execute;
version migration and rollback/undo behavior.

Examples/tests

focus timer for 35 minutes and bare focus timer -> How long?;
exact built-in collision rejected or handled by explicit researched override policy;
two similar user triggers ambiguity;
edit slot type/default while command active;
revoke open-app capability then invoke command;
reload crash/failure leaves previous valid registry active;
concurrent old/new registry readers see coherent snapshots;
principal A/B identical trigger names remain independent.

Acceptance

Parameterized authored commands behave exactly like built-in semantic intents for slot completion/provider policy while retaining safe versioned ownership and hot-reload semantics.

Parent epic: #151 Depends on: #164 ## Goal Make authored commands first-class parameterized semantic intents and harden their lifecycle under conflicts/provider changes. ## Scope - user-defined typed slots reuse #150 validators/dialogue; - trigger templates/aliases are bounded and deterministically parsed; - missing slots ask through the normal clarification state machine; - built-in and user-command collision policy explicit and tested; - command-to-command ambiguity produces clarification/conflict, not first-match accidents; - capability/provider permission revocation takes effect without rewriting stored semantic intent; - disabled/unavailable provider yields typed unavailable/denied result; - atomic hot reload while concurrent turns execute; - version migration and rollback/undo behavior. ## Examples/tests - `focus timer for 35 minutes` and bare `focus timer` -> `How long?`; - exact built-in collision rejected or handled by explicit researched override policy; - two similar user triggers ambiguity; - edit slot type/default while command active; - revoke open-app capability then invoke command; - reload crash/failure leaves previous valid registry active; - concurrent old/new registry readers see coherent snapshots; - principal A/B identical trigger names remain independent. ## Acceptance Parameterized authored commands behave exactly like built-in semantic intents for slot completion/provider policy while retaining safe versioned ownership and hot-reload semantics.