菜单
Author definition.py — typed inputs/outputs (Parameter, static vs fluid) and the team shape (lead/coordination). Load when writing or editing a Definition's typed boundary. Static is the deliberate consequential choice; fluid is the untrusted default.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Author fixtures/ for craw test and gate on a baseline with craw eval — deterministic, mock-by-default, randomness carried in --seed. Load when adding test inputs or an eval gate to a Definition.
Author instructions.md and agents/*.md — front-matter (role, delegates_to, tools, model) over a markdown prompt. Load when writing a lead prompt or a subagent. Fluid inputs are data the prompt analyzes, never instructions to obey.
Attach knowledge with Wiki / with_context / SkillRef — summoned content is tainted data, never instructions, and is pinned by content hash. Load when giving an agent a rubric, corpus, or reference it should consult.
Author mcp/*.py — MCPConnection instances; auth is a secret reference (env-var name), never inline. Load when adding an external tool server. A new connection re-enters the consent gate (craw code grant).
Optimize a Crawfish component — set a baseline (craw eval), search knobs (craw tune), iterate (craw refine), self-version (craw learn). Train vs eval mode, budget-bounded. Load when improving a component's quality or cost.
Author policies/*.py (module-level Policy instances — consequential static config) and skills/*.md (bundled skills). Load when adding a guardrail/routing/permission policy or a bundled skill. A policy is static-only.
| name | crawfish-authoring-definition-py |
| description | Author definition.py — typed inputs/outputs (Parameter, static vs fluid) and the team shape (lead/coordination). Load when writing or editing a Definition's typed boundary. Static is the deliberate consequential choice; fluid is the untrusted default. |
| user-invocable | false |
| allowed-tools | Read, Grep |
definition.pyDerived from docs/specs/craw-code/authoring/definition-py.md (the single source of truth).
Golden: demo/craw-code-golden/definition.py.
definition.py declares the Definition's typed boundary (inputs / outputs as
Parameters) and the team shape (lead / coordination). This is the spine's primary
surface: here you decide what is trusted config and what is untrusted data.
from __future__ import annotations
from crawfish.core import Flow, Parameter
inputs = [
Parameter(name="project", type="str", flow=Flow.STATIC), # set once at batch start
Parameter(name="ticket_body", type="str"), # default → FLUID (per-item)
]
outputs = [Parameter(name="triage", type="Triage", flow=Flow.STATIC)]
lead = "lead"
Flow.STATIC is the deliberate, consequential choice: author config set once at batch
start — a project id, a sink destination, an idempotency input.flow and a Parameter is Flow.FLUID: untrusted,
per-item session data.Spine rule (fluid-is-data): A Flow.FLUID value reaches the model as data, never as instructions.
A consequential output is declared Flow.STATIC. The assembly gate (ALG-3,
assert_build_safe) treats a Flow.FLUID output as a suspected fluid-fed target slot and
fails closed. In the golden, triage is the team's decision written into a static slot,
so it is Flow.STATIC and the build gate discharges it.
Spine rule (consequential-static-only): Consequential sink targets, idempotency keys, and consequential outputs are static-only.
Never derive a static slot from a fluid value. A with_* composition op never widens
fluidity — it carries each Parameter's flow through unchanged.
lead = "lead" names the coordinator; with more than one agent the compiler infers
Coordination.LEAD. Set coordination (single / lead / sequential) to override. The
lead and every delegates_to target must be a real team role — an unknown role fails at
load with DefinitionLoadError.