History system folders still empty after PAI 2.0 Packs installation (follow-up to #230)

[bustakar]

Related Issue

This is a follow-up to #230 which was closed with the explanation that PAI 2.0 Packs would address the history system gaps.

Problem

After installing PAI 2.0 with the following packs, the history folders remain empty:

• kai-hook-system
• kai-history-system
• kai-core-install
• kai-agents-skill

What's working:

• raw-outputs/ - populated ✅
• sessions/ - populated ✅

What's not working:

• learnings/ - empty
• research/ - empty
• decisions/ - empty
• execution/features/ - empty

Investigation

I did some research with Claude Code and found that subagent-stop-hook.ts expects agent types like researcher, architect, engineer, designer. However there were no agents of this type created in my .claude folder.

Additionally, the hook looks for a 🎯 COMPLETED: [AGENT:type] pattern in the output, but the DynamicAgent.hbs template doesn't instruct agents to output this format.

Questions

1. Am I missing some configuration or installation step?
2. Is there a specific way I should be spawning agents to populate these folders?
3. Or has this functionality not been fully implemented yet in PAI 2.0?

Environment

• PAI Packs: Latest from main branch
• Platform: macOS

[danielmiessler]

Thanks for following up on this @bustakar!

Quick update on v2.1.0: The new release includes a complete MEMORY/ skeleton structure with 11 pre-built directories. This ensures the folder structure exists from installation.

On your specific findings:
You're right that the specialized folders (learnings, research, decisions, execution) depend on specific agent output patterns. The subagent-stop-hook.ts looks for markers like 🎯 COMPLETED: [AGENT:type] which the standard prompts don't emit.

What's populated automatically:

• raw-outputs/ - All tool events (via capture-all-events.ts)
• sessions/ - Session summaries (via capture-session-summary.ts)

What requires specific patterns:

• learnings/ - Needs 🎯 LEARNING: or 📚 INSIGHT: markers
• research/ - Needs researcher agent output
• decisions/ - Needs decision markers
• execution/ - Needs execution tracking

This is tracked as a UX improvement - the history system should work more automatically without requiring specific output patterns.

For now, the working folders (raw-outputs, sessions) provide the core value. The specialized folders are for advanced usage patterns.

See the v2.1.0 release - the new MEMORY skeleton ensures the structure exists even if not all folders get populated immediately.

[danielmiessler]

Thank you @bustakar for the follow-up! 🙏

PAI v2.1 includes a complete MEMORY/ skeleton structure in pai-core-install with 11 organized directories (research/, sessions/, learnings/, decisions/, etc.). The VERIFY.md now includes checks for this structure.

See the release: https://github.com/danielmiessler/PAI/releases/tag/v2.1.0