← Project Log
Chronicle

What Happened

Retrofitted mystery-schools with infrastructure improvements developed for the Banyan project (banyanroot + banyancore). Focus was on formalizing Cursor agent management across branches and adding missing tracking systems.

Parallel Agent Workflow

Added git worktree support for managing multiple Cursor agent instances working on different branches simultaneously. Includes:

  • .cursor/worktrees.json for automatic worktree setup (lint configs, styles, deps)
  • Parallel Agent Workflow section in AGENTS.md with ownership rules, port isolation, and cleanup
  • Updated multi-agent-commits.mdc with worktree awareness

Provenance Tracking

Added project-log/provenance/ with schema and first entry (PROV-0001) documenting the human prompt that initiated this retrofit. The provenance system captures human prompts verbatim as evidence for project decisions.

Agent Cost Tracking

Added project-log/cost-tracking/ with pricing reference and March 2026 transaction log. Tracks agent session costs by model and provider.

Stale Branch Management

Added branch hygiene rules and triage guidance to AGENTS.md. The repo has several stale feature branches that should be evaluated:

  • feat/kg-viz — 28 behind development
  • corpus/rebalance-batch-002 — diverged
  • claude/review-project-docs-f6cUA — likely review artifact

Makefile

Added targets: provenance-new, cost-log, cost-summary, branch-status.

Files Changed

  • AGENTS.md (v2.3 → v2.4)
  • .cursor/worktrees.json (new)
  • .cursor/rules/multi-agent-commits.mdc (updated)
  • project-log/provenance/_schema.yaml (new)
  • project-log/provenance/PROV-0001_retrofit-banyan-improvements.md (new)
  • project-log/cost-tracking/_pricing.yaml (new)
  • project-log/cost-tracking/2026-03.yaml (new)
  • project-log/chronicle/2026-03-28_banyan-retrofit.md (new)
  • Makefile (updated with new targets)

Next Steps

  • Triage stale feature branches (feat/kg-viz, corpus/rebalance-batch-002, claude/review-project-docs-f6cUA)
  • Begin using provenance capture for significant human prompts
  • Begin logging agent session costs
  • Consider voice zone linting improvements (mystery-schools currently disables most ai-tells everywhere except published content; Banyan uses a graduated STRICT/MODERATE/RELAXED approach)
0:00
0:00