Personal Projects (Side)

A public launcher managing project-scoped execution boundaries for Claude Code, Codex, Kiro, and OMP, with a durable queue for ordered delivery of parallel work

• Shared global AI CLI state and profile inference from names or remotes risked mixing sessions and rules or starting in the wrong workspace — Generated project-scoped runtime homes and implemented harness-auto to select only the single most-specific registered boundary from the canonical current path, Blocked outside paths, symlink escapes, and ambiguous registrations before startup; public contract demo and CI reproduce boundary, source SHA-256, and relationship-integrity checks
• Holding a shared lock through validation, push, and readback left parallel integration waits failing even after 600 and 3,600 seconds — Separated durable enqueue/wait and implemented cross-pool per-destination FIFO, idempotent submission, worker/test budgets, frozen snapshots, and crash-after-push reconciliation, Both real jobs from two pools reached DELIVERED on attempt 1 with exit 0; duplicate submissions returned the same ID and wait timeout left jobs queued; public PR #146 preserves implementation evidence
• OMP's local fork and mismatched HOME, authentication, and resume metadata interrupted shared execution or bypassed tool policy — Replaced the local fork with upstream extensions and configuration overlays; unified runtime, auth/resume, URL-read, pre-model-call, terminal-process, and termination-signal policies, On v0.77.7, confirmed 0 GETs to denied native URL targets, 0 provider calls on print denial, and Ctrl-C exit 130 with 0 remaining native processes; upstream MCP/LSP activation and full global DB SSOT remain outside this scope

Built the deployment and scheduling infrastructure for a side project that auto-collects and normalizes job postings

• Free-tier hosting lacked cron support and the managed DB's free tier auto-paused, blocking scheduled jobs — Adopted a self-hosted Mac mini workflow-automation (n8n) scheduler instead of hosting cron, and standardized warm-up endpoints and secrets, Kept the free tier alive via daily DB warm-up and restructured manifest-based sync