AGENT SKILL

brainstorming

Use when defining ambiguous or high-complexity new features, product behavior, UI/component design, architecture choices, contract changes, or when grilling/pressure-testing a plan or design. Routine small requests stay on the fast path.

使用场景视觉设计
License 许可
Apache-2.0
最近更新
2026-11-14
Stars
★ 1.6k
Forks
491

01 前言 Front Matter

name
brainstorming
description
Use when defining ambiguous or high-complexity new features, product behavior, UI/component design, architecture choices, contract changes, or when grilling/pressure-testing a plan or design. Routine small requests stay on the fast path.

02 资源文件 3

03 SKILL.md 正文

<EXPLICIT-MODE-GATE> If activation mode is explicit (~/.config/aegis/config.toml has activation_mode = "explicit", or AEGIS_ACTIVATION_MODE=explicit is visible in the environment) and the current user request did not explicitly invoke Aegis or this skill by name, exit back to the fast path: answer concisely without this workflow's checklist, ceremony, or document requirements. If the user explicitly named Aegis or this skill, proceed normally. </EXPLICIT-MODE-GATE>

Execute

Choose the route before reading references:

  1. Tiny wording edits, simple Q&A/status, local utility or mechanical changes stay on the fast path. Bugs/diagnosis use systematic-debugging; PR, diff, or current-code review uses requesting-code-review; explicit goal framing uses goal-framing. File count alone does not decide design complexity.
  2. Direct interview requests use Grilling Mode below; soft challenge intent requires its one-line mode confirmation. Ordinary evaluation is not grilling.
  3. Ambiguous or medium/high-complexity feature, product, UI, architecture or contract design follows Default Design Path. Escalate depth from evidence: contracts, owners, persistence, migration, security, consumers and blast radius.

<HARD-GATE> Do NOT write code, scaffold projects, or invoke implementation skills until the needed design/spec approval is satisfied. Existing approval satisfies unchanged decisions; clarify only decision-changing gaps. Grilling never authorizes implementation or normal design artifacts while the interview is active. </HARD-GATE>

Route Fixtures

These rows are calibration expectations for method behavior, not a runtime regex router. The Agent selects the route from evidence; route selection is not a user question.

Scenario Route
想法还没想清楚,先梳理功能设计 normal brainstorming, compact output first
讨论公共 API 契约和兼容边界 normal brainstorming, design sections before implementation
盘问/拷问/审问这个方案,不要顺着我 Grilling Mode
修复登录按钮的空指针 systematic-debugging
review 当前 PR / diff / 当前代码 requesting-code-review
给我一个有目标的方案 goal-framing when goal intent is explicit; otherwise normal brainstorming
把按钮文案从保存改成提交 fast-path; no design ceremony

Grilling Mode

Grilling Entry Signals

  • Direct: grill me, grill this plan, 审问我, 盘问我, 拷问我, or an explicit request to interrogate/pressure-test a plan enters this mode.
  • Soft: challenge assumptions, find holes, red-team, or "别顺着我" without explicit interview intent: ask only Grill or normal brainstorming? and enter only after confirmation.
  • Not grilling: literal phrase explanations stay explanatory; PR, diff, or current-code review routes to requesting-code-review.

Grilling Mode overrides the normal brainstorming execution flow. Suspend the Default Design Path, compact output contract, and documentation or planning requirements until the user exits the interview; retain the no-implementation gate. Only after entry, read grilling-mode-reference.md for interview pace, opening, questions and exit procedure. Do not read it for ordinary design, soft-intent confirmation, or literal phrase explanations. Stop the interview when the user says stop, defer, or that questions are sufficient. Its summary does not authorize planning or implementation; return to the normal design approval gate if the user asks to proceed.

Default Design Path

Track the applicable steps in order; conditional steps do not create tasks or artifacts until their trigger is met. Keep one workflow, not separate checklist, process and after-design passes.

  1. Ground the request. Inspect current files, relevant authority docs and recent changes. Passively use active CONTEXT.md/CONTEXT-MAP.md terms; compose establishing-project-context only for a real semantic conflict. Decompose independent subsystems before designing them. Define the user's usable result, delivery and meaningful failure states; an unsupported action remains an unmet need, not a reason to silently narrow the requirement.
  2. Clarify decision-changing gaps. Follow the authority contract below. Ask one question at a time, preferably with choices and a recommendation. Reuse answers and approvals; do not ask the user for repository facts.
  3. Frame scope and alternatives. Maintain the working drafts below, propose 2-3 approaches with trade-offs and a recommendation, and say what is in, deferred or owned elsewhere. Apply only evidence-triggered checks from the conditional table before recommending the approach.
  4. Present the design. Scale sections to complexity: affected architecture, components/data flow, failure/recovery, consumers, compatibility and testing. Follow existing patterns; keep each unit cohesive and testable, and remove unnecessary features. Obtain required approval for each applicable design section before proceeding; return to clarification when evidence changes it.
  5. Persist only when needed. Apply the Doc Necessity Gate. For a written Design Spec, load the self-review section, fix placeholders, contradictions, ambiguity, scope and boundary defects, then obtain user review. A small Spec Brief may use concise review unless project authority requires a formal gate.
  6. Hand off only when ready. Check Design Ready And Design Complete below; preserve ADR signals for durable owner/contract/schema/dependency/compatibility/ runtime-boundary changes, including source refs, alternatives and baseline-sync questions. Do not create accepted architecture memory from unexecuted ideas. The terminal state is invoking writing-plans, not another implementation skill. For a scope ending at discussion or options, deliver that result; do not advance beyond the user's request.

Compact output contract

Use natural, proportional prose, not a mandatory stack of named cards. Aegis Visibility briefly explains the task-specific value of design first. Keep these working drafts current as scope changes; they are in-session views unless the Doc Necessity Gate requires persistence:

  • TaskIntentDraft: outcome, goal, success evidence, stop condition, non-goals, scope and risks;
  • BaselineReadSetHint: candidate authority refs and gaps;
  • BaselineUsageDraft: required, delivered-context (optional), acknowledged, cited and missing refs, plus the advisory decision;
  • ImpactStatementDraft: affected layers, owners, invariants, compatibility and non-goals.

Present the relevant intent, evidence, impact, options and decision needed at this stage. Conditional lenses below add only their applicable findings; mentioning a lens does not activate its template. Do not repeat satisfied checks or emit empty cards. Preserve decision-relevant meaning and evidence limits.

Doc Necessity Gate

Before writing any spec, plan, ADR or baseline artifact:

  1. If an existing spec/plan/ADR/baseline owns this surface, update it in place; never create a sibling owner.
  2. Write the smallest artifact only for a durable/irreversible surface (schema, public API, owner, dependency, migration, compatibility retirement), cross-session/ cross-person handoff, or an approval/authority requirement. Otherwise keep compact drafts in-session and write no document.
  3. Re-check at edit time and closeout; expand only if uncertainty or impact grows. Workspace creation and spec persistence follow project authority. Read the workspace/bootstrap reference only after selecting a durable artifact.

Role And Authority Contract

Agent-owned decisions

Resolve these directly without asking the user:

  • repository investigation strategy and evidence gathering order
  • file and function organization inside an accepted owner
  • testing commands and proportional verification mechanics
  • inline versus subagent execution when policy already allows it
  • reversible implementation structure that does not change product behavior, contract, authority, or durable boundaries

User-owned decisions

Ask the user only for:

  • product behavior or preference
  • irreversible, destructive, external, public, production, or sensitive impact
  • explicit product/contract commitments only the user can make
  • necessary information unavailable from repository, tools, and authority docs

Every user question must pass this test:

If the user chooses another answer, which design boundary, behavior, owner, acceptance criterion, or risk decision changes?

If none changes, do not ask. When a question passes this test, attach a recommended option and the reason for it, so the user decides between framed choices instead of researching. This classification clarifies which decisions are user-owned; it does not remove the approval points this workflow already defines.

Conditional Detailed Guidance

For design detail, read only the evidence-matched heading of expanded-design-guidance.md, never the whole file. Do not load design detail for route selection, first-turn clarification, route-away cases, or an active Grilling Mode interview. The main file owns routing, approval and handoff; references supply the selected procedure or template only.

Evidence trigger Exact section to read / action
Repository evidence is insufficient and a bounded probe can change the approach ## Design Probe
The software scenario class is known and its domain risks matter ## Software Scenario Profiles; use only the relevant profile
Doc Necessity Gate selects a spec/workspace artifact ## Documentation And Workspace Bootstrap; a named baseline template only when initializing that exact file
A written Design Spec is ready for review ## Spec Self-Review
A baseline-dependent recommendation needs structured usage evidence ### Baseline Usage Draft under ## Design Artifact Templates
Requirements, scenarios or acceptance remain incomplete or disputed before a design recommendation ### Requirement Ready Check; stay at proposal/spec clarification until decision-changing gaps are resolved
An approach introduces an owner, skill, artifact, adapter, fallback, compatibility path, workflow step or metric ### Existence Check; reuse the existing owner when sufficient
Ambiguous product, feature, workflow or architecture choices ### Product Risk Lens; keep value, non-goals, trade-offs and the decision compact
Choosing a medium/high-complexity implementation direction ### Plan-Time Complexity Check And Complexity Budget; inspect actual owners/file shape before choosing
Novel mechanism, protocol, artifact or interaction; unresolved alternatives after internal reuse checks; domain outside repository evidence ### Prior-Art And Reuse Lens (Prior-Art & Reuse Lens); routine established patterns need no external research ceremony
Both what to build and where it belongs are at issue ### Baseline Role Alignment; distinguish requirement truth from architecture truth

Before approach selection, unresolved new/duplicate owners, fallback, adapter, compat-only carrier, delete-first questions, unverified assumptions or a "long-term stable" claim after Existence Check escalate to first-principles-review and its Decision Hygiene Review; return here once that decision surface is clean. Do not make this a universal design ceremony.

For narrower architecture coherence risks, use Architecture Integrity Lens: invariant, canonical owner/contract, responsibility overlap, higher-level simplification, retirement/falsifier and verdict. For internal retirement vs compatibility retention vs persistent-state confirmation, compose anti-entropy-governance; it chooses delete-first | compat-exception | confirmation-first without transferring destructive authority to this skill.

For user-facing interface or interaction design, compose ui-ux-governance and carry scoped experience rules into existing acceptance criteria. Design approval and handoff remain here.

Design Ready And Design Complete

Approach selection is ready when:

  • the desired outcome and primary scenario are known;
  • scope and non-goals are explicit;
  • current behavior and the target delta are grounded;
  • key invariants and the likely canonical owner are identified;
  • at least one observable acceptance criterion exists;
  • no open unknown can still change the approach category.

Not every unknown must be eliminated; only decision-changing unknowns block convergence.

The design can hand off when all applicable conditions hold:

  • the selected approach and canonical owner are explicit;
  • fixed behavior/contract and implementation-owned choices are separated;
  • alternatives were materially compared or excluded by evidence;
  • critical assumptions have evidence or explicit acceptance;
  • failure/recovery and consumer impact are covered where applicable;
  • acceptance criteria are observable and usable by verification;
  • new owners, fallbacks, adapters, compatibility, or persistence have creation proof and retirement/rollback treatment;
  • every user-owned decision has real user approval.

Design Complete is method readiness, not completion authority. Transition to writing-plans only after these conditions hold; do not carry unresolved decision-changing unknowns into the plan.

04 同仓库更多 Skills

hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-review-with-annotations

Create, inspect, reply to, update, or resolve situated Worktable annotations. Use when feedback belongs to a Doc section, block, text range, or HTML Doc, or when the user asks to address existing review feedback and coordinate a requested artifact edit.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-find-and-synthesize-context

Find, assess, and synthesize context from Worktable Docs and Records. Use when the user asks what a workspace contains, wants workspace-backed research, needs current facts reconciled across artifacts, or requests a cited synthesis without changing workspace state.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-create-or-update-docs

Create, edit, rename, or maintain narrative Worktable Docs. Use for plans, decisions, briefs, notes, research syntheses, and other durable prose when the user requests a new artifact or a safe update to an existing one.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-create-or-manage-records

Organize repeated structured information as durable Worktable Records. Use for research datasets, inventories, directories, pipelines, trackers, and other collections whose items need identity, fields, filtering, links, validation, or individual updates.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-create-or-manage-interactive-html

Create or improve durable visual and interactive artifacts in Worktable. Use for dashboards, trackers, calculators, mini apps, slide presentations, and other requests that benefit from a polished HTML Doc.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-collaborate-threads

Read, answer, and hand off messages through durable Worktable threads with users or other authenticated participants. Use when the user asks to check addressed messages, reply to a participant, continue a thread, request another agent's help, wait briefly for a response, or watch for comments and messages sent to the agent.

★ 1.6k