adk-skill: Instructions file for Codex

Working on ADK Skill

This repository distributes a Markdown skill for Google's Agent Development Kit. adk-skill/ is the installable package; project documentation and validation/ are development resources. Read README.md for setup, supported SDK versions and verification commands, CHANGELOG.md for release notes, and tasks.md for the active update status.

Editing

• Keep the complete adk-skill/SKILL.md below 500 lines. Its frontmatter name must match the directory; description and compatibility limits are 1024 and 500 characters.
• Keep essential decisions in the entrypoint and detailed guidance in its nine references/*.md files. Link references directly from the entrypoint; do not introduce chained local reference links. These are repository gates.
• Keep Python examples in the entrypoint; native SDK examples belong in references/multi-language.md. Identify application fragments and provide imports for complete examples.
• Check APIs against the official tag/package version being documented. Record the version and verification date; distinguish stable, preview and experimental APIs. Do not infer cross-language parity from version numbers.
• Update existing documentation in place. Maintain human-facing instructions in README.md, release notes in CHANGELOG.md, and progress in tasks.md; avoid duplicate reports inside the installable skill.
• Keep the historical effectiveness benchmark labeled as historical until it is rerun.

Validation

• Follow the offline test commands in README.md. Tests execute Markdown examples with the target ADK; python -m unittest discover -s validation -v must not call models, cloud services or use real credentials.
• Run the schema/quality and static security checks pinned in .github/workflows/release.yml, plus its structural checks and git diff --check.
• Review scanner findings individually before changing .skillspector-baseline.yaml; preserve explicit reasons and avoid wildcard suppressions.
• Compile/type-check native examples when changed. Report toolchain versions, failed attempts and checks that need external services. Python CI does not validate Android, KSP generation, Live audio or production deployment.
• The artifact is Markdown; there is no application build. validation/ contains the maintained regression suite. Existing ignored tests/, doc/ and reports/ may contain local fixtures or history; do not treat them as release inputs or inspect credential files.