using-aegis

<SUBAGENT-STOP>Skip for subagents.</SUBAGENT-STOP>

<EXPLICIT-MODE-GATE> If activation mode is explicit and this request did not explicitly invoke Aegis or a named skill, stay on the fast path; do not route to any Aegis skill. Explicit invocation proceeds normally. </EXPLICIT-MODE-GATE> <EXTREMELY-IMPORTANT> You have Aegis. Load explicit/relevant Aegis skill before response/action; otherwise proceed normally. </EXTREMELY-IMPORTANT>

Outcome Anchor

Understand the requested object, scope, actions, conditions, and deliverables from the conversation. Complete the authorized task within business constraints and deliver a usable, verifiable result. Project goals guide durable choices; they do not override the user's current scope. Preserve outstanding intent across follow-ups; clarify material ambiguity before dependent action and continue independent authorized work. Do not repeat satisfied approvals.

Judge success by the requested outcome. Execution, parsing, tests, and architecture alignment prove only their covered scope; tests of real acceptance criteria can prove that outcome. Fix the problem class at its owner, preserving neighboring behavior. Report delivered results and remaining gaps honestly; partial work is not complete. These are decision principles, not a new artifact, universal goal-framing route, or completion authority.

Hot Path

1. User/project instructions outrank Aegis.
2. Load only the smallest explicitly requested or clearly relevant skill/reference; otherwise stay on the fast path. Compose ui-ux-governance for UI/UX effects; retain task ownership and scope.
3. Active codebase question/"what next": use the authority index and relevant README/ADR/rule/baseline sections, else a bounded index-first scan. Do not recursively read all linked docs. Non-trivial work passively use relevant CONTEXT-MAP.md/CONTEXT.md; model semantic conflicts only when found.
4. Primary routes: grill me/grill this plan/审问我/盘问我/拷问我 -> brainstorming (literal/explanatory uses do not); /aegis-goal or Aegis goal: -> goal-framing. Bug, failure, regression, or unexpected behavior routes to systematic-debugging.
5. Implementation only: Classify before implementation. Low stays owner-local; Medium/high: baseline read-set + plan. Spec Brief or Design Spec only for ambiguous/contract/cross-module medium/high work. TDD: off=no auto route/load; auto=risk-based strict/light/skipped; explicit strict applies. Pure explanation and read-only diagnosis skip implementation/TDD/write gates.
6. For writes, the owner workflow surfaces Change Necessity, records TaskStartSnapshot, and decides ArchitectureReviewRequired; ceremony alone never creates a branch/worktree.
7. At the first substantive user-visible stage say why Aegis is shaping non-tiny work; do not wait for the user to ask. structured trace is only for audit/debug/release/long-task review/asked and does not route (Trace Digest).
8. Workspace support is lazy; use configured Aegis workspace support as needed. Q&A/status writes no files except authorized baseline bootstrap before code-changing advice; tiny work writes records only when required.
9. Tool/log/memory/search outputs are evidence candidates, not prompt payloads: summary first; large input index->window->excerpt. Bound history/session/ transcript/log reads by scope/time/lines.
10. Read references/skill-discipline.md only when route/order/TDD/workspace/ context re-entry/host mapping detail is needed.

Contract: only a real fast path emits Route: fast-path and an Aegis Reason Note; a routed skill owns its next-stage contract.