✦
Knowledge Hub session — pi-teams prompt modes and cheap executor experiment
operations / inbox/2026/04/08/knowledge-hub-session-pi-teams-prompt-modes-and-cheap-executor-experiment.md
project-doc Live smoke test — 2026-04-08 project-doc Knowledge Hub session log project-doc Knowledge Hub simplification plan capture Knowledge Hub session — pi-teams prompt modes and cheap executor experiment lesson Pi-teams prompting and cheap executor lessons playbook Cheap executor decision checklist playbook Pi team lead playbook unknown Operations
Home / operations / inbox / 2026 / 04 / 08 / knowledge-hub-session-pi-teams-prompt-modes-and-cheap-executor-experiment.md

Knowledge Hub session — pi-teams prompt modes and cheap executor experiment

Properties
Type
capture
Status
active
Curation state
raw
Created
2026-04-08T02:30:48Z
Updated
2026-04-08T02:30:48Z
Tags
knowledge-hub pi-teams prompting cheap-executor codex review
ID
capture-2026-04-08-knowledge-hub-session-pi-teams-prompt-modes-and-cheap-executor-experiment
Source agents
ai:pi
Source machines
macos
operations / inbox/2026/04/08/knowledge-hub-session-pi-teams-prompt-modes-and-cheap-executor-experiment.md

Knowledge Hub session — pi-teams prompt modes and cheap executor experiment

Session summary

We used Knowledge Hub itself as the project context while designing and validating a reusable pi-teams prompting system for this repo.

Work completed

  • Reviewed and updated local pi-teams skill understanding
  • Created three project-local team modes:
  • frontier-open
  • mixed-guided
  • budget-scaffolded
  • Created project-local agent definitions for architect, backend, frontend, reviewer, and skills roles
  • Added project-local prompt-policy skill and reusable playbooks/checklists
  • Ran a mixed-guided live team experiment on a tags-related slice
  • Found and fixed a real backend integration gap in tag-filtered home navigation
  • Ran a cheaper-executor experiment with explicit openai-codex/gpt-5.4-mini
  • Completed and reviewed the cross-vault quick switcher home-page slice
  • Committed and pushed the resulting repo changes

Important observations

  • Outcome-first prompting with minimal justified constraints worked well
  • Mixed-guided was sufficient for small contract-bearing work
  • Repo audits are necessary before assigning roadmap slices because the branch may be ahead of the plan
  • Smaller models can work well for bounded, visible, low-risk slices when strong review is present
  • Provider-qualified model IDs should be used for cheaper-model experiments

Repo artifacts created

  • .pi/teams.yaml
  • .pi/agents/*.md
  • .pi/skills/knowledge-hub-team-prompting/SKILL.md
  • .context/TEAM-LEAD-PLAYBOOK.md
  • .context/CHEAP-EXECUTOR-CHECKLIST.md

Commits

  • 6d8825d Improve tag-filtered search and cross-vault quick switcher
  • f11997e Add Knowledge Hub team prompting playbooks