AGENT SKILL

strands-agents

Route Strands Agents monorepo work across the Python SDK, TypeScript SDK, docs site, MCP server, cross-SDK parity, testing, troubleshooting, and contribution workflows.

License 许可
Apache-2.0
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
strands-agents
license
Apache 2.0
metadata
{ "disco-role": "operating" }
description
Route Strands Agents monorepo work across the Python SDK, TypeScript SDK, docs site, MCP server, cross-SDK parity, testing, troubleshooting, and contribution workflows.
disable-model-invocation
true

02 资源文件 8

03 SKILL.md 正文

Strands Agents

Use this repo skill for tasks about the Strands Agents monorepo, the strands-agents Python package, the @strands-agents/sdk TypeScript package, the Astro/Starlight documentation site, or the strands-agents-mcp-server package.

Read first

  1. Read repo-overview.md to confirm repository layout, selected scope, and excluded infrastructure.
  2. Read cross-sdk-parity-and-contribution.md before changing public APIs, hook events, docs, tests, or contribution-facing behavior.
  3. Read troubleshooting.md when install, dependency, credential, browser, network, AWS-infra, or staleness problems appear.
  4. Check repo-provenance.md when deciding whether this skill should be refreshed against a newer checkout.
  5. Use repo-routing-metadata.json only as structured router metadata for managed repo-skill import flows.

Public install clues

  • Python users normally install strands-agents with any task-required optional extras and may also install strands-agents-tools for community tools.
  • TypeScript users install @strands-agents/sdk with Node.js 20+.
  • MCP server users can launch strands-agents-mcp-server through a Python package install or uvx.
  • In a checkout, use the package-specific sub-skill before installing broad dev dependencies or all optional extras.

Route by task

Task signals Go to
Python Agent, @tool, model providers, MCP client, memory, sessions, hooks, interventions, plugins, sandbox, telemetry, graph/swarm/A2A, Python tests python-sdk
TypeScript Agent, Zod/JSON tools, model providers, MCP tools/client, middleware, hooks, interventions, memory, sessions, storage, sandbox, telemetry, graph/swarm, examples, package exports typescript-sdk
Astro/Starlight docs pages, MDX, snippets, <Syntax>, tabs, sourceLinks, navigation, generated API docs, docs review/audit/planning docs-site
strands-agents-mcp-server, search_docs, fetch_doc, docs catalog indexing, cache hydration, section extraction, URL restrictions, MCP server tests mcp-server
Cross-SDK naming, literal parity, hook event parity, public/internal API, comments, logging, feature lifecycle, PR readiness cross-sdk-parity-and-contribution.md plus the relevant SDK sub-skill

Fast checks

  • Use python-import-smoke.sh when Python packages are installed and you need import/signature sanity for both Python packages.
  • Use ts-check.sh for root TypeScript workspace checks when npm dependencies are installed.
  • Use site-check.sh for docs-site checks when site npm dependencies are installed.
  • Sub-skills provide narrower helpers for Python SDK, TypeScript SDK, docs-site, and MCP server checks.

Operating rules

  • Work in the smallest relevant package and test slice first; do not run broad provider, browser, network, or AWS-infrastructure checks unless the task selects them.
  • Keep Python and TypeScript concepts in parity by meaning, not identical code. Identifiers match after language recasing; string literal casing follows the cross-SDK rules.
  • Public exports require deliberate API review. Python public surfaces use __all__; TypeScript public surfaces use package/barrel exports.
  • Do not call SDK tools "skills" in SDK-facing docs. Reserve "skills" for agent skill files.
  • When source files move, update docs sourceLinks in the same change.
  • Treat test-infra/ as opt-in real AWS infrastructure. Do not deploy it or set internal flags unless the task explicitly targets that stack.
  • Record any skipped credential, network, browser, Docker, AWS, or live-provider checks in the final handoff instead of implying they passed.

Scope boundaries

This skill is intentionally a reusable monorepo operating guide, not a release or CI automation manual. It covers package development, docs authoring, source-backed troubleshooting, and safe verification selection. Release automation, GitHub workflow internals, account-specific infrastructure, generated build output, caches, and one-off migration scripts are out of default scope unless a future task explicitly selects them.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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).

★ 351