Extract bounded, source-linked context from large files or many inputs. Use when context is too large, files are oversized, or a task needs concise excerpts.
中文简介正在准备中。当前显示原文说明。
Decide whether to reuse, improve, or scaffold a deterministic helper. Use when repeated work is slow, brittle, or blocked by a missing local tool.
中文简介正在准备中。当前显示原文说明。
查看安装方法安装前可以查看交给 AI 代理的指令内容。
Use homebase_tool_factory with the task, desired helper, and any existing script. Prefer improving a nearby working helper over creating a new one.
Create a tool only when it removes repeated manual work or closes a real capability gap. Keep it repo-local, deterministic, and covered by a small smoke or regression check.
After the helper works, record the command and proof path in state.yaml, LEDGER.md, or memory when the lesson is durable.
Use available_types and required_types when the task has known artifact contracts. Treat a composition as a proposal requiring consumer checks; insufficient_fit calls for more evidence or an explicit creation decision.
Scaffolding creates a draft .contract.json. Implement the helper and populate its inputs, outputs, prerequisites, effects, failure semantics, resources, and behavioral cases before setting status to implemented. validate executes those cases and records exact artifact identities; help output and test-name mentions are insufficient. promote packages the validated version and pins the skill entrypoint. Host discovery remains a separate check.
For corrections, preserve the original acceptance criterion and connect diagnosis, candidate, and frozen evaluation in an adaptation.py episode. Passing evaluation stops at ready_for_review. Activate only on an explicit request naming the reviewed candidate digest. See docs/sips-05.md in the plugin source for contracts and commands.
Use the shared adaptation controller for structured investigations and bounded probes.
Prefer current v2 contracts and joint-prerequisite composition before new helpers.
Use additive counterexamples; preserve raw episodes when proposing procedures.
Policy trials are declarative proposals, never changes to evaluation or activation.
See docs/sips-06.md for request shapes, supported schemas, and research limits.
Use coverage to propose consumer interaction cases and morphology to explore repair/reuse alternatives. Feasible combinations remain proposals until executable consumer checks pass. See method API and command protocol.
Use contract v3 for explicit semantic roles, units, literal context identities,
and numeric ranges. Missing or incompatible declarations reject a binding; no
implicit conversion is available. Runtime bounds do not prove declared meaning.
Keep independent consumer checks. Inspect adaptation dependencies before reuse;
changed evaluator or dependency identities require revalidation. See
docs/sips-09.md for the supported vocabulary and limitations.
还没有评价。欢迎分享你的使用感受。
无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。
还没有评论。来写第一条吧。
Extract bounded, source-linked context from large files or many inputs. Use when context is too large, files are oversized, or a task needs concise excerpts.
中文简介正在准备中。当前显示原文说明。
Inspect SIPS Homebase status, manifest wiring, routes, host visibility, and MCP freshness. Use when asked for SIPS status, command center, host audit, plugin visibility, or whether Homebase is fresh.
中文简介正在准备中。当前显示原文说明。
Route a task through SIPS commands, agents, scripts, MCP tools, or bounded delegation. Use when asked to route, split, fan out, escalate, or choose the right SIPS path.
中文简介正在准备中。当前显示原文说明。
Turn failures, logs, symptoms, or flaky behavior into a compact repro and verification plan. Use when debugging needs a reproducible failure path.
中文简介正在准备中。当前显示原文说明。
Capture and develop a reusable skill, tool repair, or behavior improvement noticed while completing a task. Use when instructions are stale, a recurring workflow deserves a skill, or a tool contract fails in practice.
中文简介正在准备中。当前显示原文说明。
Search, inspect, and record SIPS-owned Memory Fabric lessons. Use when a task needs prior lessons, recurring-fix memory, recall health, scoped historical context, or when a just-fixed bump or error should be recorded.
中文简介正在准备中。当前显示原文说明。