菜单
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.
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 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.
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.
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-knowledge |
| description | 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. |
| user-invocable | false |
| allowed-tools | Read, Grep |
Wiki / with_contextDerived from docs/specs/craw-code/authoring/knowledge.md. Golden:
demo/craw-code-golden/knowledge.py.
Knowledge is attached by composition, not as a directory file: build a Wiki, freeze it,
and summon it into a Definition with with_context. The result is a new frozen Definition
that carries a pinned reference to the knowledge — the body is not copied inline.
from crawfish.wiki import Wiki, TrustTier
from crawfish.derive import with_context, SummonMode
wiki = Wiki().with_page(
"triage-rubric",
"P0 = outage; P1 = broken core flow; P2 = degraded; P3 = cosmetic.",
trust=TrustTier.TRUSTED, # first-party curated; STILL summoned tainted
)
specialized = with_context(base_definition, wiki, mode=SummonMode.READONLY)
wiki.consult() materialises each page as a tainted context entry: knowledge arrives as
data the agent reads, never an instruction surface, and can never reach an instruction slot
or a static-only sink. Even TrustTier.TRUSTED content is summoned tainted — the trust tier
only ever raises suspicion. Taint can be dropped only via an audited declassify,
unreachable from a fluid path.
Spine rule (knowledge-is-tainted): Summoned knowledge is tainted data, never an instruction surface.
with_context stores a SummonRef — {id, version, mode} — where version is the Wiki's
content sha snapshotted at compose time. export().checksum moves iff the pinned summon
version moves, and the summoned body never enters the export checksum. A frozen (eval-mode)
Wiki is required for a stable summon; readonly() seals an unfrozen Wiki first.
Spine rule (pinned-by-hash): Knowledge is summoned by pinned content hash; the body never enters the export checksum.