Skip to content

file-lifecycle.md

Latest commit

 

History

History
49 lines (39 loc) · 2.03 KB

file-lifecycle.md

File metadata and controls

49 lines (39 loc) · 2.03 KB
Raw

File Lifecycle

This repo treats tracked files as the durable project memory. Keep runtime instructions compact and keep large rationale in docs.

Track

  • Root AGENTS.md startup instructions.
  • .opencode/instructions/path-rules/*.md files that OpenCode must read before creating or editing matching paths.
  • .opencode/docs/ reference material that explains procedures, rationale, and lookup tables.
  • Design, architecture, production, QA, and sprint artifacts that represent project decisions.
  • Review logs and handoffs when a skill explicitly asks for them.
  • production/resume-index.md when /handoff derives it from canonical state; keep it compact, tracked, and disposable. The installer allowlists this path for tracking but does not ship, own, or remove the project-created file.

Ignore Or Keep Local

  • Temporary command output, raw logs, caches, generated build artifacts, and local editor files.
  • production/session-logs/session-baseline.json; the session-start hook rewrites this local review-scope anchor for each session.
  • One-off transcripts unless converted into a concise project decision or handoff.
  • Legacy Claude runtime files as OpenCode dependencies. They may exist for coexistence or migration history, but OpenCode instructions must not depend on them at runtime.

Anti-Redundancy Policy

  • Root AGENTS.md holds the hot-path startup contract only.
  • Path-scoped Markdown files in .opencode/instructions/path-rules/ hold discipline needed before editing matching paths.
  • Long procedures belong in .opencode/docs/.
  • If two files need the same rule, keep the short operational rule in the hot path and link to one longer explanation.

Pause Audit

Before ending a work unit, check:

  • Did every changed file belong to the requested scope?
  • Are generated or temporary files left untracked?
  • Are verification results labeled accurately?
  • Is the next action discoverable from session state, handoff, or final reply?
  • Would a fresh session know which docs to read without scanning the whole repo?