feat(skills): add SkillDiscoveryMode to skip the discovery turn

[thomaslprr]

Link to Issue or Description of Change

1. Link to an existing issue (if applicable):

• Closes: feat(skills): add include_list_skills to inject the L1 catalog without list_skills #7092

Problem:
There is no public way to inject the L1 skill catalog into the system instruction. Hiding list_skills required mutating _tools.

Solution:
Add SkillToolset(discovery_mode=SkillDiscoveryMode.EAGER|LAZY).

• LAZY (default): expose list_skills.
• EAGER: hide list_skills and inject <available_skills> XML so the model can call load_skill directly.
  Non-enum values raise TypeError.

Testing Plan

Unit Tests:

• I have added or updated unit tests for my change.
• All unit tests pass locally.

.venv/bin/python -m pytest tests/unittests/tools/test_skill_toolset.py -q
# 159 passed

Manual End-to-End (E2E) Tests:

• Local sample based on contributing/samples/environment_and_skills/skills_agent with discovery_mode=SkillDiscoveryMode.EAGER:
  • no list_skills tool
  • <available_skills> present in the system instruction (greeting-skill, weather-skill)
  • agent can call load_skill directly
• Default LAZY still exposes list_skills.
• With a registry, search_skills remains available in EAGER.

Checklist

• I have read the CONTRIBUTING.md document.
• I have performed a self-review of my own code.
• I have added tests that prove my fix is effective or that my feature works.
• New and existing unit tests pass locally with my changes.
• I have manually tested my changes end-to-end.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[wukath]

Hi @thomaslprr, thanks for the PR! We like the idea of skipping the discovery turn, but we reshaped it slightly to use an enum (discovery_mode=SkillDiscoveryMode.EAGER vs LAZY) since omitting list_skills has the major side effect of injecting the catalog into the system prompt. We kept your core logic and all your tests -- would this approach work for you?

[thomaslprr]

Hi @thomaslprr, thanks for the PR! We like the idea of skipping the discovery turn, but we reshaped it slightly to use an enum (discovery_mode=SkillDiscoveryMode.EAGER vs LAZY) since omitting list_skills has the major side effect of injecting the catalog into the system prompt. We kept your core logic and all your tests -- would this approach work for you?

Hi @wukath, yes, that works for me - discovery_mode=SkillDiscoveryMode.EAGER|LAZY is clearer than a boolean given the system-prompt side effect. Happy for you to land it that way. Thanks!

Would you like me to update this PR to the enum API, or will you land the reshaped version on your side? I already have the changes ready locally if you'd like me to push them.

[thomaslprr]

@wukath, I pushed the enum reshape as discussed:

• discovery_mode=SkillDiscoveryMode.LAZY (default): keep list_skills
• discovery_mode=SkillDiscoveryMode.EAGER: hide list_skills and inject <available_skills>
• non-enum values raise TypeError

Core logic and tests are updated (159 passed locally). I also set up a local ADK sample based on skills_agent with discovery_mode=EAGER and confirmed the agent run path: no list_skills, catalog injected, load_skill available. Ready for another look.

Thanks!

[adk-bot]

Thank you @thomaslprr for your contribution! 🎉

Your changes have been successfully imported and merged via Copybara in commit 3d3c1d9.

Closing this PR as the changes are now in the main branch.