Skip to content

STATE.md

Latest commit

 

History

History
99 lines (76 loc) · 4.17 KB

STATE.md

File metadata and controls

99 lines (76 loc) · 4.17 KB

State

Now (2026-03-25)

remove-active-session branch — shared current-session bridge removed from agent-session

The refactor target is now implemented.

Completed architecture

  • :active-session-id is gone from canonical root state
  • shared ctx no longer carries :current-session-id*
  • core has no shared current session concept
  • RPC owns its own focus pointer
  • TUI owns its own focus pointer
  • service is explicit-only
  • core/session logic routes through explicit session-id or scoped :target-session-id ctx

What changed

Core atom / ctx shape

  • removed root-state :active-session-id
  • removed shared ctx :current-session-id*
  • ss/active-session-id-in now reads only :target-session-id
  • added explicit ss/retarget-ctx helper for scoped execution

RPC

  • RPC runtime state owns focus-session-id*
  • targetable ops are scoped from explicit request :session-id or RPC-local focus
  • per-request session routing is explicit; RPC no longer uses dynamic *request-session-id* binding
  • lifecycle flows update RPC focus locally
  • no shared ctx focus writes remain

TUI / console

  • TUI owns :focus-session-id in UI state
  • main runtime closures scope ctx explicitly from adapter-local focus
  • CLI/TUI no longer depend on shared ctx current-session fallback

Resolvers / service

  • removed :psi.agent-session/context-active-session-id from resolver surface
  • service APIs require explicit session-id

Lifecycle

  • new/resume/fork return session data and re-scope locally
  • no lifecycle operation mutates a shared current-session pointer

Tests / harness

  • tests updated to re-scope ctx explicitly after lifecycle operations
  • test support now seeds :target-session-id directly

Truthful invariant

agent-session core has no shared current-session concept. Session routing is explicit: either adapter-local focus or scoped :target-session-id.

Tests

All current Clojure tests are passing:

  • bb clojure:test:unit ✓ 935 tests, 5660 assertions, 0 failures
  • bb clojure:test ✓ 935 unit + 86 extension tests green

What remains next

The bridge-removal plan is complete. Remaining work is follow-on cleanup / next design steps:

  • extension instance state storage
  • retire remaining agent-core coupling in some consumers
  • configuration scoping
  • cache stability improvements
  • prompt request-preparation convergence (agent profile / skill injection)
  • model selection hierarchy for helper/background task execution
  • post-Wave-B executor thinning is complete:
    • psi.agent-session.executor has been deleted
    • prompt/streaming ownership now lives in prompt-turn / prompt-loop

Completed follow-on cleanup since Wave B:

  • psi.ai.providers.openai.common has been removed after migrating direct callers
  • psi.app-runtime.background-jobs has been removed after migrating commands, RPC, widgets, and tests to background-job-view
  • auto-session-name extension vertical slice is now landed:
    • prompt lifecycle emits extension event session_turn_finished
    • extension runtime supports delayed extension dispatch via psi.extension/schedule-event
    • extension API supports explicit source-session targeting via :query-session and :mutate-session
    • extensions.auto-session-name now infers titles from journal-backed visible session content through a helper child session
    • extension-local guards prevent helper recursion, stale checkpoint overwrite, and overwrite when the current session name diverges from the last auto-applied name

Post-Wave-B check

Wave B of the Gordian refactor is complete.

Completed commits:

  • da746cef — ⚒ prompts: extract shared prompt streaming helpers
  • 177a3551 — ⚒ prompts: split single-turn and loop execution
  • 7221fa57 — ⚒ prompts: narrow prompt-runtime to prepared-request execution
  • 1530e0b9 — ⚒ dispatch: consolidate prompt lifecycle handler registration
  • 5ea34cc0 — ⚒ background-jobs: extract canonical view model
  • 0614907f — ⚒ openai: separate transport content and reasoning helpers

Gordian wave-boundary snapshot:

  • src: propagation cost 13.7%, cycles none, namespaces 181
  • src+test: propagation cost 12.9%, cycles none, namespaces 306