oxidizer(#1420): AIShouldRespondServerCommand — delegate to cognition/should-respond, drop parallel reimpl#1421
Merged
Conversation
…/should-respond, drop parallel reimpl
AIShouldRespondServerCommand carried a separate gating implementation
(custom prompt + JSON-repair-via-second-LLM retry) parallel to the
canonical cognition/should-respond Rust path that AIDecisionService.
evaluateGating already used. Two paths could drift independently;
the TS prompt was already stale relative to the Rust template.
This PR collapses both into one: the TS command now constructs the
AIDecisionContext from the public AIShouldRespondParams shape and
delegates to client.cognitionShouldRespond. Rust owns the prompt,
model call, parser, and typed AIGatingDecision contract.
## Diff
- AIShouldRespondServerCommand.ts: -85 / +59 LOC (net -26)
- ESLint baseline ratchet: 5435 -> 5433 (-2)
## Dead TS deleted
- `import { AIProviderDaemon }` (no longer referenced in this file)
- `import { TextGenerationRequest }` (parent-class type only;
not used by the delegation)
- `import { LOCAL_MODELS }` (Rust evaluate_gating carries its own
DEFAULT_GATING_MODEL constant; missing model defaults Rust-side)
- Inline gating instruction build + message-array construction
(Rust cognition/should_respond.rs::build_gating_prompt owns it)
- JSON-repair-via-second-LLM retry path (Rust returns typed errors;
caller decides retry policy at the IPC seam)
- Stale `>>> trigger <<<` marking logic (Rust handles trigger
marking inside build_gating_prompt — verified parity)
## What stays
- The thin TS shim: param -> RustAIDecisionContext mapping +
AIShouldRespondResult construction.
- Verbose debug mode: still emits ragContext.messageCount +
conversationPreview (TS-derivable, no Rust round-trip needed).
`promptSent` / `aiResponse` debug fields now sentinel-pointer
to the Rust logs (`cognition::should_respond`) where they
actually live.
- Catch-around-throw error path (matches sibling shim discipline).
## Discipline
- Cast `params -> RustAIDecisionContext` mirrors the existing
pattern in AIDecisionService.evaluateGating (`as unknown as`
for the structurally-matching surface).
- Synthetic `triggerMessage.id` derived from the timestamp so
repeat calls don't multiply observability noise (params don't
carry one; Rust requires it).
- No fail-open default — failures throw, caller catches via
the existing error-return path.
## Refs
- #1420 sub-card (just filed)
- Existing Rust: cognition/should_respond.rs::evaluate_gating
(already shipped, in production via AIDecisionService.evaluateGating)
- Sibling pattern: codex's #1383 check_redundancy delegation,
my #1402 generate_response delegation
- #1248 umbrella
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Merged
6 tasks
joelteply
added a commit
that referenced
this pull request
May 18, 2026
…se-decision (one PR) (#1426) * oxidizer: AIValidateResponseServerCommand → cognition/validate-response-decision (one PR per zero-users directive) Single-PR oxidizer per Joel 2026-05-18 19:44Z directive (zero users, full-blown Rust-driven dev, no migration ceremony). Adds the Rust cognition path AND replaces the TS parallel reimplementation in the same commit — no 4-PR cadence. Renamed command + binding from `cognition/validate-response` to `cognition/validate-response-decision` to avoid collision with the existing persona-validator surface (which already owns `cognition/validate-response` in modules/cognition.rs:814). ## What this ships Rust: - `cognition/validate_response.rs` (~380 LOC + 14 tests): - `ValidateResponseRequest` / `ValidateResponseDecision` / `ResponseDecision` (ts-rs exported) - `ValidateResponseError` typed enum (NoAdapter / Generation) - `build_validate_prompt` — pure prompt builder (mirrors TS template byte-for-byte modulo substitutions) - `parse_decision` — pure one-word parser (SUBMIT/CLARIFY/SILENT) with TS-parity precedence: CLARIFY > SILENT > Submit (fail-open default) - `reason_for` — canonical reason strings - `evaluate_validate_response` — async orchestrator (Groq via existing registry, llama-3.1-8b-instant default, temp 0.1, max 10 tokens) - `modules/cognition.rs`: `cognition/validate-response-decision` IPC arm - ts-rs barrel adds 3 new types (cognition/{ResponseDecision, ValidateResponseDecision, ValidateResponseRequest}.ts) TS: - `bindings/modules/cognition.ts`: new `cognitionValidateResponseDecision` binding method. - `commands/ai/validate-response/server/AIValidateResponseServerCommand.ts`: thin shim. Deletes inline prompt template, parseDecision, getReasonForDecision, AIProviderDaemon/TextGenerationRequest/ LOCAL_MODELS imports. ## Discipline - One PR carries Rust + TS shim + dead-TS delete (zero-users mode). - All errors typed (NoAdapter, Generation). No silent default-on-error. - Fail-open SUBMIT default in parser matches TS behavior (silence more user-hostile than off-topic). - Clippy held at 157 baseline (resolved 1 new unreachable-pattern warning by renaming colliding match arm). ## Tests - 14 logic + ts-rs tests pass (`cognition::validate_response::*`) - npm run build:ts clean - Clippy at baseline ## Refs - Joel 2026-05-18 19:44Z: zero-users full-blown-Rust-dev mode → one PR per oxidizer - Sibling: codex's #1383 + my #1402/#1421 pattern (one-PR delegation) - #1248 umbrella Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * chore(eslint): lock baseline drop 5433→5432 from validate-response TS deletion --------- Co-authored-by: Test <test@test.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
5 tasks
joelteply
added a commit
that referenced
this pull request
May 18, 2026
…h owns gating now (#1428) After #1421 + #1424 + the should-respond oxidizer wave, the `AIShouldRespondCommand` parent class's protected helpers (`buildGatingInstruction`, `parseGatingResponse`) became dead code. The Server impl now delegates to `RustCoreIPCClient.cognitionShouldRespond` which routes through the Rust `cognition/should_respond.rs::evaluate_gating` path; nothing calls the TS helpers anymore. Per Joel's zero-users no-migration-ceremony directive: delete now, not "deprecate for follow-up." Single PR. ## Diff - AIShouldRespondCommand.ts: -172 LOC (kept 7-LOC shell for inheritance) ## What gets deleted - `protected buildGatingInstruction(params): string` (~95 LOC) — Rust `cognition/should_respond.rs::build_gating_prompt` is the prompt source of truth. - `protected parseGatingResponse(aiText): Partial<...>` (~65 LOC) — Rust `cognition/should_respond.rs::parse_gating_response` is the parser source of truth. ## What stays - Class shell (`AIShouldRespondCommand extends CommandBase`) — still the inheritance base for both `AIShouldRespondServerCommand` and `AIShouldRespondBrowserCommand`. - `static readonly commandName = 'ai/should-respond'` — used by the command registry to discover the command name. ## Verification - `npm run build:ts` — clean (no remaining references to deleted helpers). - `grep` for callers of the deleted methods: zero (only the now-deleted definitions themselves matched in the prior wave). - ESLint baseline check: no change (the deleted methods were lint-clean). ## Refs - continuum#1421 (AIShouldRespondServerCommand delegation that orphaned these helpers) - Joel 2026-05-18 19:44Z (zero-users, no migration ceremony → delete-loser-in-one-PR) Co-authored-by: Test <test@test.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
joelteply
added a commit
that referenced
this pull request
May 25, 2026
…1438) RAG-mode generation now delegates to Rust via AIDecisionService — the same IPC seam PersonaUser's response path already uses. Rust owns prompt assembly (system prompt + history + time prefixes + hour-gap markers + identity reminder), provider routing, admission gating, timeout, and token-usage stamping (build_response_messages + build_response_generation_request in cognition/generate_response.rs). Direct-message + preview modes stay TS-side: - Direct mode is an introspection/test path that bypasses admission; Rust intentionally does not expose a "skip the gate" code path. - Preview mode reconstructs the request Rust would build as a local mirror. Source of truth is the Rust path; if assembly drifts a `cognition/preview-request` IPC is the fix. Mirrors the pattern from #1421 (should-respond) and #1426 (validate-response-decision). The 100-line of TS message-building that duplicated build_response_messages now lives only in Rust. Co-authored-by: Test <test@test.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
AIShouldRespondServerCommandcarried a parallel reimplementation of gating — separate prompt template + JSON-repair-via-second-LLM retry loop, all running outside the canonicalcognition/should-respondRust path thatAIDecisionService.evaluateGatingalready uses. Two paths drifted independently; TS prompt was stale relative to Rust.This PR collapses both into one: the TS command constructs the
AIDecisionContextfrom the publicAIShouldRespondParamsshape and delegates toclient.cognitionShouldRespond. Rust owns the prompt, model call, parser, and typedAIGatingDecisioncontract.Diff
AIShouldRespondServerCommand.ts: -85 / +59 LOC (net -26)Dead TS deleted
import { AIProviderDaemon }(no longer referenced in this file)import { TextGenerationRequest }(only used by deleted helper)import { LOCAL_MODELS }(Rustevaluate_gatingcarries its ownDEFAULT_GATING_MODEL)cognition/should_respond.rs::build_gating_promptowns it)>>> trigger <<<marking logic (Rust handles trigger marking insidebuild_gating_prompt)What stays
RustAIDecisionContextmapping +AIShouldRespondResultconstruction.ragContext.messageCount+conversationPreview(TS-derivable, no Rust round-trip).promptSent/aiResponsedebug fields now sentinel-pointer to Rust logs (cognition::should_respond).Discipline
params → RustAIDecisionContextmirrorsAIDecisionService.evaluateGatingpattern (as unknown asfor structurally-matching surface).triggerMessage.idderived from timestamp (params don't carry one; Rust requires it).Refs
cognition/should_respond.rs::evaluate_gating(already in production viaAIDecisionService.evaluateGating)Test plan
npm run build:ts— clean./jtag pingboot receipt)🤖 Generated with Claude Code