adk-code
Use this sub-skill when writing or revising Python ADK agent code in an agents-cli scaffolded project, including agent objects, tools, callbacks, state, A2A surfaces, event triggers, and recipe selection.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- adk-code
- license
- Apache 2.0
- metadata
- { "author": "Google", "license": "Apache-2.0", "version": "1.3.1", "requires": { "bins": [ "agents-cli" ], "install": "uv tool install google-agents-cli" }, "disco-role": "operating" }
- description
- Use this sub-skill when writing or revising Python ADK agent code in an agents-cli scaffolded project, including agent objects, tools, callbacks, state, A2A surfaces, event triggers, and recipe selection.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/adk-code-guide.md3.2 KB
- 参考文档references/adk-python.md35.7 KB
- 参考文档references/adk-workflows.md16.9 KB
- 参考文档references/samples.md8.2 KB
03 SKILL.md 正文
ADK Code Patterns
Use this sub-skill inside the google-agents-cli repo skill. It is a router plus operating checklist; move into the bundled references for full command flags, schemas, and examples.
When to Use
- The user wants to implement or modify agent Python code after scaffolding.
- The task asks about tools, callbacks, state, memory, A2A, event triggers, or recipes.
- You need to avoid hand-writing generated A2A/FastAPI surfaces.
Workflow
- Inspect the scaffolded app layout before editing.
- Use recipes for retrieval, sandboxing, memory, OAuth, guardrails, and scheduling instead of inventing flags.
- Edit agent instructions/tools/state deliberately and preserve generated serving glue.
- Verify locally with
agents-cli runand then with eval workflows.
Read These References
references/adk-code-guide.md— read for adk code guide details.references/adk-python.md— read for adk python details.references/adk-workflows.md— read for adk workflows details.references/samples.md— read for samples details.
Verification and Safety
Safe checks: inspect scaffolded files and run local agents-cli run only when project dependencies and model credentials are ready.
Boundaries
- Does not scaffold or deploy projects.
- Does not use pytest as the primary oracle for LLM response quality; use eval.
Related Sub-Skills
../workflow/SKILL.md— lifecycle routing and approval gates.../scaffold/SKILL.md— project creation/enhancement.../adk-code/SKILL.md— ADK Python implementation patterns.../eval/SKILL.md— evaluation loops and metrics.../deploy/SKILL.md— deployment and infrastructure.../publish/SKILL.md— Gemini Enterprise registration.../observability/SKILL.md— logging, tracing, and analytics.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).