Your agent knows your preferences. It just never uses them

Short summary

Agent memory systems fail silently when preferences are stored correctly but never retrieved at the moment they matter — similarity search can't connect 'fix the login bug' with 'use conventional commit style.' The fix is action-triggered retrieval: tag rules with the action they govern (trigger:git-commit) and fetch them deterministically right before that action runs. A Claude Code hook injects matching rules into context before commits and PRs, with hard failure-to-silence guarantees so memory never breaks your workflow.

• •Agent preferences are rules about actions, not facts about topics, so similarity search misses them at the critical moment
• •Action-triggered retrieval tags rules with the action they govern and fetches them deterministically before that action runs
• •Claude Code hooks inject procedural rules before git commits and PRs, failing silently to avoid breaking workflows

Generated with AI, which can make mistakes.