fix: register extensions for the active integration only

extension add registered commands for every detected agent, and
integration upgrade back-filled enabled extensions for non-active
integrations. Maintainer direction on #2948: treat the project as
single-active. Only the active integration gets extension artifacts;
use/switch rescaffold the target when the user selects it.

- extension add now routes through the all-agents pass restricted to
  the active integration (only_agent), keeping detection and
  missing-skills-dir recovery safeguards. Projects without recorded
  init-options fall back to detection-based registration.
- integration upgrade re-registers extensions only when upgrading the
  active integration, reversing the #2886 back-fill for non-active
  targets at maintainer request.

Fixes #2948

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
This commit is contained in:
marcelsafin
2026-07-10 20:23:09 +02:00
co-authored by Copilot
parent 1736f0746b
commit af9205fc2c
6 changed files with 185 additions and 66 deletions
+7 -12
View File
@@ -376,19 +376,14 @@ def _register_extensions_for_agent(
"""Register all enabled extensions' commands/skills for ``agent_key``.
``use`` / ``switch`` re-register enabled extensions for the agent they
activate; ``upgrade`` backfills them for the refreshed agent. Plain
``install`` deliberately does not call this helper so adding a secondary
integration has no extension side effects until it is selected or upgraded.
See issue #2886.
activate (rescaffold); ``upgrade`` does so only for the *active*
integration. Plain ``install`` and upgrade of a non-active integration
deliberately skip this helper so a secondary integration has no extension
side effects until it is selected. See issues #2886 and #2948.
Known limitation: extension *skill* rendering is scoped to the active
agent (init-options track a single ``ai`` / ``ai_skills`` pair). A
skills-mode agent registered while it is *not* the active agent (e.g.
Copilot ``--skills`` registered while non-active) therefore
receives command files rather than skills here — matching ``extension
add``'s multi-agent behavior. ``use`` / ``switch`` avoid this because they
make the target the active agent first. Per-agent skills parity is tracked in
#2948.
Callers always pass the active agent (use/switch activate the target
before registering), so extension *skill* rendering — which is scoped to
the active ``ai`` / ``ai_skills`` init-options — matches ``agent_key``.
Best-effort: never aborts the surrounding integration operation. Callers
invoke it *after* the use/upgrade/switch transaction has committed so a
@@ -491,17 +491,19 @@ def integration_upgrade(
if stale_removed:
console.print(f" Removed {len(stale_removed)} stale file(s) from previous install")
# Re-register enabled extensions for the upgraded agent so its extension
# commands are (re)created — including agents installed before this
# back-fill existed. Mirrors switch for command registration; see #2886.
# Done after the upgrade has fully settled (Phase 2 included) and outside
# the try/except above so this best-effort step cannot affect upgrade
# success.
_register_extensions_for_agent(
project_root,
key,
continuing="The integration was upgraded, but installed extensions may need re-registration.",
)
# Re-register enabled extensions only when upgrading the *active*
# integration, so its extension commands are (re)created after the
# upgrade settled (Phase 2 included). Done outside the try/except above
# so this best-effort step cannot affect upgrade success. Non-active
# integrations are rescaffolded by `use` / `switch` instead — the #2886
# back-fill for non-active agents was removed at maintainer request
# (#2948).
if key == installed_key:
_register_extensions_for_agent(
project_root,
key,
continuing="The integration was upgraded, but installed extensions may need re-registration.",
)
name = (integration.config or {}).get("name", key)
console.print(f"\n[green]✓[/green] Integration '{name}' upgraded successfully")