When working on sensitive projects, proprietary code, or systems where context leakage is a concern. Use when the user says "protect this," "sensitive project," "don't leak," "sanitize output," "clean context," "proprietary," "confidential," or when working across multiple projects where context bleed between them could be problematic. Implements defensive patterns to prevent sensitive context from leaking into outputs, commits, or cross-project contamination.
When ending a session, consolidating memory, or when memory files feel stale or contradictory. Use when the user says "dream," "consolidate memory," "clean up memory," "merge notes," "memory maintenance," "what do you remember," "update your knowledge," "session end," or when memory/context files have accumulated noise over multiple sessions. Performs cross-session memory synthesis inspired by KAIROS autoDream.
When the user needs to perform the same operation across multiple files, components, or entities. Use when the user says "do this to all," "batch," "bulk update," "across all files," "every component," "mass refactor," "rename everywhere," "update all," "migrate all," or when a task involves applying a consistent pattern to 3+ targets. Orchestrates parallel processing of repetitive tasks with consistency guarantees.
When the user wants to find bugs, test edge cases, stress-test code, or perform adversarial analysis of their codebase. Use when the user says "find bugs," "break this," "what could go wrong," "red team," "hunt bugs," "adversarial test," "edge cases," "stress test," "what did I miss," "pen test this logic," or before any major release. Operates with red team psychology — assumes everything is broken until proven otherwise.
When the user wants a thorough code review of changes, a PR, a file, or a feature branch. Use when the user says "review this," "code review," "check my code," "what's wrong with this," "review my PR," "look over this," "critique this," "review before merge," or when submitting code for quality assessment. Performs multi-pass review covering correctness, security, performance, maintainability, and edge cases.
When the user wants to deeply understand a codebase, map its architecture, or build a persistent mental model for future sessions. Use when the user says "explain this codebase," "map the architecture," "codebase insight," "understand this repo," "what does this project do," "how does this work," "analyze the codebase," or when starting work on an unfamiliar project. Produces a persistent insight file that survives across sessions.
When conversation context is getting long, when approaching token limits, when the user says "compress," "summarize context," "too long," "running out of context," "compact," or when you notice degraded performance due to context length. Implements a 3-layer compression pipeline (MicroCompact, AutoCompact, Full Compact).
When the user has a complex task that would benefit from multiple agents working in parallel or in sequence. Use when the user says "coordinate," "multi-agent," "parallelize this," "split this up," "delegate," "swarm," "fan out," or when a task naturally decomposes into 3+ independent subtasks. Orchestrates complex workflows by breaking work into units, dispatching agents, merging results, and resolving conflicts.