A hierarchical context system that gives AI agents persistent memory, enforced rules, and structured knowledge — across every session, every project, every conversation. No more re-explaining. No more lost instructions.
CLAUDE.md loads user-wide rules and preferences
Three-tier layering: global CLAUDE.md defines user-wide rules, project-level overrides add repo-specific instructions, and session context captures the current task. Each layer inherits from the one above it.
MEMORY.md index files point to topic-specific memory files. When a session starts, the SessionStart hook automatically recalls relevant memories using semantic similarity — no manual loading required.
Four structured categories — user (preferences and identity), feedback (correction patterns), project (architecture decisions), and reference (external knowledge). Each type has its own retention policy and recall priority.
When context windows fill, PreCompact hooks fire before compression occurs. Critical instructions, enforced rules, and active task state are marked as non-compressible. The agent retains what matters and discards safely.
Context doesn't die when a conversation ends. Memories stored during one session are available in the next. Trajectories capture multi-step workflows. Procedures record reusable processes. The agent builds knowledge over time, not per-session.
Non-negotiable instructions that survive context compression, session boundaries, and model switches. Privacy rules, approval gates, and code standards persist regardless of conversation length or complexity.
Without context management, every AI session starts from zero. You re-explain your coding standards, your tool preferences, your project architecture — every single time. At 30 seconds per re-explanation, across 40+ daily sessions, that's 20 minutes of wasted context-setting per day.
Worse, without enforced rules, critical instructions get compressed away during long conversations. Privacy constraints vanish mid-session. Approval gates disappear after context window rotation. The agent "forgets" the rules you set.
BulletproofSoftware.tech's context management solves both problems. The hierarchical system ensures agents always know who they're working with. The auto-memory system ensures they remember what they've learned. And PreCompact hooks ensure critical rules survive no matter how long the conversation runs.