Skip to content

Pi Health Check — a continuous optimization ritual for the pi setup #13

Description

@huaxel

Destination

A manual on-demand, agent-assisted pi health check ritual that covers performance and cost review, configuration hygiene (detecting cruft and drift), and reliability assessment — delivered as a just pi-healthcheck task that produces a dual-format (terminal + JSON) report. The ritual identifies issues and recommends action but does not auto-fix.

Notes

  • Domain: Pi (coding agent) configuration, extensions, packages, models, cost tracking, session management
  • Skills: Systematic debugging for crash/failover patterns; wayfinder for decision tickets
  • Dotfiles repo is source of truth~/dotfiles/pi/agent/ is canonical, ~/.pi/agent/ contains legacy stale copies; PI_CODING_AGENT_DIR is the runtime mechanism
  • Agent preference: Subagent over workflow
  • Report format: Compact terminal summary (5 grades A–F) + rich JSON for machine consumption

Decisions so far

  • Health Check Data Sources — Catalogued all pi data sources: dynamic-footer observability (per-turn tokens/cost/TPS), crash/failover logs, session storage (523 MB across 89 project dirs), 11 extensions, 18 npm packages, config files with significant drift between dotfiles and runtime
  • Observability Pipeline Status — Root cause: trim({ keepLast: 10 }) in dynamic-footer intentionally caps history to 10 sessions; recommendation is to increase to 200+ and make configurable
  • Session Storage Audit — 522 MB total, 69 orphaned worktree sessions (196 MB, 38%), 281 files >30 days old (183 MB); 73% safe to auto-prune in 3 phases
  • Crash and Failover Pattern Analysis — 1 crash (context compaction at 175k tokens), failover is proactive load-balancing across 2 accounts (sub-2 used 58%), only 1 quota-exhaustion event in 9,606 requests
  • Config Sync InvestigationPI_CODING_AGENT_DIR makes dotfiles the runtime source of truth; ~/.pi/agent/ is largely stale/legacy; drift detection should compare against what PI_CODING_AGENT_DIR loads
  • Config Drift Detection Approach — Compare settings.json, models.json, trust.json, extensions/subagent/config.json via structured key-level Node.js diff. 7 drift categories with severity levels. Exclude auth.json and volatile files. Dual-format report.
  • Health Check Output Design — Terminal report: 5 grades (A–F) for Cost & Performance, Config Hygiene, Session Storage, Reliability, Observability; one-page summary with critical items. Rich JSON with full findings, grades, and recommendations. Dual-format output from just pi-healthcheck.

Not yet specified

The map is fully specified — all decisions needed to start building have been made.

Out of scope

  • Auto-fix — report only
  • Continuous/scheduled runs — manual on-demand
  • General pi feature work

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    wayfinder:mapWayfinder map for a navigation effort

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions