Systems Lab

Agent skill

python-coder

Implement and refactor production Python, FastAPI, Pydantic, async, and migrations.

activeSelf-containedActs undeclared1,006 words

From sno-ai/sno-station-skills · 50 skill entries · 861 · pushed 2026-10-30

What it does when it runs

Implement and refactor production Python, FastAPI, Pydantic, async, and migrations.

Automated analysis of the skill and the 2 files bundled beside it. A skill’s own description is written to be selected by an agent, so it describes the job and not the dependencies.

Keys and connectors you must supply
None found.
Hosts it reaches
No third-party host appears in the skill or its bundled files.
Tool permissions it declares
No allowed-tools in the frontmatter. It does act, so it runs under whatever permissions your session already grants.
Actions present in the files
shell

Ask about python-coder

Opens your assistant with this page's recorded context. Check the linked sources before relying on it.

Is this safe to install?ClaudeChatGPT
Adapt it to my stackClaudeChatGPT
What else do I need for it to workClaudeChatGPT
Rather ask a human? Talk to Cheetah
git clone --depth 1 --filter=blob:none --sparse https://github.com/sno-ai/sno-station-skills.git /tmp/sno-station-skills
git -C /tmp/sno-station-skills sparse-checkout set "plugins/sno-station-standalone/skills/python-coder"
mkdir -p ~/.claude/skills/python-coder
cp -R "/tmp/sno-station-skills/plugins/sno-station-standalone/skills/python-coder/." ~/.claude/skills/python-coder/

Picked up without a restart. A project skill of the same name is shadowed by your personal one. For one repository only, swap ~/.claude/skills for .claude/skills. Claude Code docs ↗

Or take the whole library

This repo ships a .claude-plugin manifest, so Claude Code can install all 50 skills at once. Plugin skills are invoked as /<plugin>:<skill>, so they never collide with your own.

/plugin marketplace add sno-ai/sno-station-skills
/plugin

The folder is the same in every client that implements the format — 46 of them — so if yours is not above, only the destination changes.

Reproduced in full from sno-ai/sno-station-skills/blob/1c92b0ff8931aea0f29fc3cdca50a0f3e772c0dc/plugins/sno-station-standalone/skills/python-coder/SKILL.md, which is licensed APACHE-2.0 (repository). 1,006 words, 5 headings.

Python Coder

Every development task has a finite budget including setup, tests, and reporting. Before running tests, estimate duration from available timings and identify the changed behavior or directly affected consumer. If timing is unknown, state an estimate; do not build a timing system or run a preliminary suite. Full evaluation, unit, integration, or repository suites require the owner's explicit request, never a skill, PRD, review, or default command. Remove checks that neither prove the requested main function nor expose its defects. Repeat only after a relevant change or failure. Every new security check requires the owner's explicit approval for that specific check; remove auxiliary security, assertion, hash, or bookkeeping blockers of useful work.

Write production Python that fits the current repository, stays small, and is easy to verify. Local repo instructions add project-specific detail; they do not weaken these rules.

Runtime Source of Truth

Honor the Python version declared by the user or the repository (pyproject.toml requires-python, .python-version, setup.cfg, tox.ini, CI, package metadata). Never emit syntax, stdlib APIs, or typing features newer than the target. For 3.14+ targets, prefer 3.14 idioms where they improve clarity or correctness: lazy annotations (PEP 649), InterpreterPoolExecutor (PEP 734), t-strings (PEP 750), free-threading awareness (PEP 779).

Boundaries

  • No backward compatibility by default. No legacy paths, compatibility wrappers, migration shims, dual-format support, or fallback behavior unless explicitly requested. A published external contract is the only exception, and only when the user or repo instructions confirm it is published.
  • Read before writing. Relevant exports, immediate callers, shared utilities, nearby tests, and the package's typecheck/test setup — before adding code.
  • Official docs for third-party behavior. Before changing a dependency version, or using a third-party API, SDK, CLI, or service in a way neither the PRD nor existing code already shows, read its current official documentation; for a version change, read the release notes between the two versions. Name the page and version in the report. When a PRD governs the work and lacks or contradicts that fact, add it to the PRD's execution notes.
  • Surgical changes. Touch only what the task requires. No adjacent refactoring, no speculative abstractions or configurability, no error handling for scenarios that cannot happen.
  • Deterministic code for deterministic decisions. Routing, retries, status-code handling, schema validation, parsing, and transforms belong in code, never in a model call.
  • Conflicting local patterns: pick one source of truth (by authority, recency, test coverage), state the choice, flag the other for cleanup. Never blend conventions into a third style.

Design Rules

Intent: types make illegal states unrepresentable, data structures replace branches, and every line is a liability that must earn its place.

Types

  • Zero Any. Use object + narrowing, generics, or TypedDict. cast() only at validated boundaries, with a comment saying why.
  • Literal[...] over bare str; TypedDict over dict[str, Any] — split required/optional via a required base class + total=False subclass so guaranteed keys stay guaranteed.
  • Protocol over ABC (structural typing, stub-friendly testing); class hierarchies max 1 level — favor composition.
  • @dataclass(frozen=True, slots=True) for value objects with behavior; TypedDict for data crossing boundaries (JSON, APIs, configs).
  • T | None over Optional[T]; type X = ... aliases (3.12+); typing.Self for fluent/builder methods; @overload when the return type depends on an argument; ClassVar[T] for class-level state.
  • Abstract types (Sequence/Mapping/Iterable from collections.abc) in parameters, concrete types in returns. Explicit return types on public functions.
  • from __future__ import annotations: omit on 3.14+ targets and remove it from files you edit there; on older targets follow the repo's existing annotation strategy — do not remove it opportunistically.
  • if TYPE_CHECKING: for heavy or cycle-prone imports.

Structure & errors

  • Functions under ~50 lines, max 3 indentation levels, early-return guard clauses, happy path at lowest indentation.
  • Exception hierarchies for expected errors: a base class plus narrow subclasses, so callers choose their catch specificity. Wrap only the line that can raise; never bare except:; no mutable default arguments; no magic strings/numbers (constants or Literal).
  • Resources (connections, files, locks) always via context managers or finally.
  • __all__ declares each module's public API. Pure logic in the core, side effects at the edges.
  • Replace manual counting/grouping/nested combinatoric loops with Counter, defaultdict, itertools.
  • Docstrings are contracts (preconditions, silenced errors, external failure modes), not narration; skip when nothing is non-obvious.

Async & parallelism

  • asyncio.TaskGroup over gather() — structured cancellation, no orphaned tasks.
  • CPU-bound work: InterpreterPoolExecutor over ProcessPoolExecutor (3.14+); keep threads for I/O.
  • Shared mutable state is always locked (threading.Lock / asyncio.Lock) — free-threaded Python (PEP 779) turns previously-hidden races into real bugs.

Boundaries & config

  • Keep fixed-format parsing and protocol literals in code; use existing translation resources only for text that requires localization.
  • Reuse the repository's configuration and boundary-validation approach; do not introduce a configuration framework, translation system, or startup gate for an ordinary change. Internal calls trust types.
  • SQL always via parameterized queries, HTML through the framework's escaping, log fields via extra={} — never hand-interpolated strings. T-strings (string.templatelib, 3.14+) do no escaping by themselves; use them only where the repo already has a vetted template processor for that exact sink.

Logging

  • Never print() in production code. logging.getLogger(__name__); %s formatting or extra={} — never f-strings in log calls; logger.exception for stack traces.

Verification

Before reporting done, every claim must trace to a command actually run this session:

  1. The repo's declared type checker passes with zero errors on changed files; no typing features newer than the declared runtime.
  2. Run the repo's declared check/fix command and report zero remaining errors.
  3. Search changed files for Any — confirm none; confirm annotation strategy matches the target runtime.
  4. Re-read the diff once: delete anything removable without losing correctness or clarity, and confirm every changed line traces to the request.

If a check fails, report the failure with its output — never report done with errors outstanding.

For advanced patterns (generic protocols, async generators, thread-safe singletons, custom decorators, sentinel objects), read references/advanced-patterns.md.

<!-- reminders:start --> <!-- reminders:end -->

Files bundled with it

These load only when the skill asks for them, so they cost nothing until it runs.

Need help setting it up?

This page tells you what python-coder does and what it needs. Cheetah builds the agent setup it runs inside: data, CRM, sequencing and the guardrails.

Book a call →

The directory stays free. There is nothing gated behind this.

Build my GTM engine

Tell us what you want to improve. Fedor reviews the workflow and replies by email when there is a useful fit.

Fedor reviews every request personally. If there is a useful fit, he will reply by email.