refactor: own semantic boundary and result projection strategies #10

Merged

severus-snape merged 1 commit from refactor/context-strategies into main

2026-10-27 07:41:13 +00:00

severus-snape commented

2026-10-27 07:41:00 +00:00

Collaborator

Scope

Extract actual semantic boundary assessor and pure tool-result projection algorithm into source plugin repository. Core retains contracts, compatibility validation/defaults, durable context state machine/persistence/usage/cooldowns/authority/provider lifecycle. Generic startup contextStrategy ESM API1 selects assess+project together, validates and binds receivers, no silent fallback. CLI injects selected projector into PiRuntime instead of adapter importing concrete algorithm. Default algorithm/prompt unchanged, transcript never rewritten.

Validation

587/587 full tests, typecheck/build/diffcheck PASS. Existing23 context/pruning focused tests PASS incl unicode, ID correlation, duplicates/errors/nontext/recent preservation, coexistence. Two new actual local ESM strategy tests assert custom assessment/project selection without baseline or model calls, receiver retention, missing/malformed startup rejection, default selection. Existing runtime/context suites included. No state migration or live settings changes.

Limits/review

Trusted startup code no sandbox/hot reload/resources. Embedders using PiRuntime must explicitly inject projectToolResults to enable projection; absence leaves messages unchanged, documented. Threshold/action scheduling remains Core, only concrete semantic and pruning decisions extracted. No universal policy config or completion/nightly extraction claimed. Author counterchecked; current persistent review blockers0, no additional independent reviewer/reset. No generated edits or original dirty worktrees altered. Immutable plugin pin. Merge/build not deployment; final activation separately verified.

## Scope Extract actual semantic boundary assessor and pure tool-result projection algorithm into source plugin repository. Core retains contracts, compatibility validation/defaults, durable context state machine/persistence/usage/cooldowns/authority/provider lifecycle. Generic startup contextStrategy ESM API1 selects assess+project together, validates and binds receivers, no silent fallback. CLI injects selected projector into PiRuntime instead of adapter importing concrete algorithm. Default algorithm/prompt unchanged, transcript never rewritten. ## Validation 587/587 full tests, typecheck/build/diffcheck PASS. Existing23 context/pruning focused tests PASS incl unicode, ID correlation, duplicates/errors/nontext/recent preservation, coexistence. Two new actual local ESM strategy tests assert custom assessment/project selection without baseline or model calls, receiver retention, missing/malformed startup rejection, default selection. Existing runtime/context suites included. No state migration or live settings changes. ## Limits/review Trusted startup code no sandbox/hot reload/resources. Embedders using PiRuntime must explicitly inject projectToolResults to enable projection; absence leaves messages unchanged, documented. Threshold/action scheduling remains Core, only concrete semantic and pruning decisions extracted. No universal policy config or completion/nightly extraction claimed. Author counterchecked; current persistent review blockers0, no additional independent reviewer/reset. No generated edits or original dirty worktrees altered. Immutable plugin pin. Merge/build not deployment; final activation separately verified.

severus-snape added 1 commit

2026-10-27 07:41:00 +00:00

refactor: own context assessment and tool-result projection strategies aff5e9f658

severus-snape merged commit c7e1ac4f26 into main

2026-10-27 07:41:13 +00:00

severus-snape referenced this pull request from a commit

2026-10-27 07:41:15 +00:00

Merge pull request 'refactor: own semantic boundary and result projection strategies' (#10) from refactor/context-strategies into main