GanyuanRan/Aegis

Aegis Goal Framing creates a thin goal frame before execution when a user sets an Aegis goal or asks for scope, success evidence, or stop conditions. It is used to define boundaries before routing into the next workflow.

Anti-Entropy Governance helps decide when to delete old code paths, collapse duplicate owners, or remove fallbacks safely. It also guards schema, persistence, and source-of-truth changes that need explicit confirmation.

Brainstorming helps turn ambiguous or high-complexity ideas into clearer designs by asking focused questions, comparing approaches, and shaping a spec before implementation. Use it for new features, architecture, contracts, or plan pressure-tests.

Code Review Reception helps you evaluate code review feedback before making changes. Use it when comments are unclear, risky, disputed, or may conflict with the codebase or prior decisions.

Communicating Concisely shortens responses when the user asks for caveman mode, fewer tokens, or brief answers. It drops filler and keeps technical details exact until turned off.

Dispatching Parallel Agents delegates unrelated tasks to separate agents with isolated context. Use it when 2+ independent problems can be worked in parallel; otherwise handle them inline or with a planned subagent workflow.

Establishing Project Context maintains canonical domain language in CONTEXT.md so humans and agents use the same terms. Use it when project work reveals a renamed, conflicting, or deprecated term that needs active semantic modeling.

Executing Plans loads a written implementation plan, checks it for fit and risk, and carries it through execution with checkpoints and verification. Use it for multi-step work across sessions or when review gates are needed.

Finishing A Development Branch handles verification, merge or PR decisions, and ownership-aware cleanup for a task-created Git branch or worktree. Use it when verified work needs integration or the user asks to manage branch lifecycle steps.

First Principles Review helps evaluate ambiguous design, debugging, or architecture choices using first-principles and Occam's-razor reasoning. Use it when competing constraints, fallback growth, or ownership confusion could send work down the wrong path.

Long Task Continuation keeps multi-step work checkpointed and resumable across pauses, handoffs, or context resets. Use it when a task spans sessions, uses subagents, or needs drift and evidence tracking.

Recording Architecture Decisions helps decide when an ADR or baseline update is needed after an architecture change, and guides create, amend, supersede, or skip actions. It is used to keep decision records and current-state architecture aligned.

Requesting Code Review dispatches a reviewer subagent with a focused evidence pack instead of your session history. Use it after implementation slices, before merges, or when you need an independent review of risks and compatibility.

Subagent-Driven Development uses fresh subagents for each independent plan task, followed by spec and code-quality reviews. Use it when coordinating written implementation plans in the current session; otherwise keep work inline.

Systematic Debugging helps diagnose bugs, test failures, and unexpected behavior before proposing fixes. It guides isolation, root-cause tracing, owner identification, and minimal verification at the canonical owner.

Test-Driven Development defines when to use strict TDD and how to run RED/GREEN/REFACTOR with Aegis. Use it for explicit test-first work, bug fixes, and behavior changes that need failing tests before code.

Update Aegis updates the installed Aegis Method Pack for the current AI coding host. Use it to check whether Aegis is current, register a host, or perform an explicit update without changing the target project.

Using Aegis checks routing rules at the start of a turn and decides when to load a specific Aegis skill. It is used to stay on the fast path or switch into a skill based on the task.

Using Git Worktrees helps when a task needs a separate checkout, dirty state blocks branch switching, or the repository explicitly requires a worktree. It guides safe creation, placement, setup, and cleanup of task-owned worktrees.

Verification Before Completion requires a fresh falsifying check and explicit evidence before claiming work is complete, fixed, or ready to hand off. Use it when preparing to commit, merge, publish, release, or otherwise declare success.

Writing Plans turns approved specs or requirements into durable implementation plans with scope, file boundaries, verification, and risks. Use it for multi-step work before changing code.

Writing Skills uses pressure-scenario validation to check whether a skill changes agent behavior before deployment. Use it when creating, editing, or verifying skills and you need baseline-first validation.