Feat files - #2
Closed
arantespp wants to merge 27 commits into
Closed
Conversation
5 tasks
arantespp
added a commit
that referenced
this pull request
Jun 24, 2026
…y embedding docs (#252) Fix #2 — memory_name in knowledge results: - Add memoryName to MemoryKnowledgeResult (both semantic and non-search paths in knowledgeMemory.ts); the Memory association is already loaded, so no extra query. - Add required memory_name to the OpenAPI MemoryKnowledgeResult schema; regenerate SDK/CLI (generated files are gitignored, rebuilt in CI). - Inject context label now uses the human name: [Memory: <name>] instead of the opaque [Memory: <id>] (agentKnowledge.ts). - Tests: assert memory_name in REST knowledge search; label test expects the memory name. Fix #1 — memory embedding resilience (docs only): - The code already aligns memory to documents (MemoryEntry.embedding is nullable and writeMemoryEntry is non-fatal on embedding failure). The PRD was stale: correct prd-memories.md (NOT NULL -> NULL, write-algorithm STEP 1 non-fatal note, design-principle wording). Docs: prd-knowledge.md response examples and knowledge.md data model gain memory_name. Verified: pnpm --filter @soat/server typecheck, eslint --fix, and the full test suite (76 suites / 1641 tests) pass against a local Postgres + pgvector. Claude-Session: https://claude.ai/code/session_01ErUBchseZrpYysrq5rZMNF Co-authored-by: Claude <noreply@anthropic.com>
arantespp
added a commit
that referenced
this pull request
Jun 28, 2026
…n fields, trace_id) (#294) * fix(agents): address UX/docs issues from issue #293 - Fix #1: parseBedrockSecret now treats plain ABSK strings as { apiKey: value } instead of silently discarding them - Fix #2: buildBedrockModel now falls back to config.apiKey when no secret is linked, so api_key in the provider config object is honoured instead of being silently ignored - Fix #3: Add PATCH /agents/:agent_id as an alias for PUT, so both HTTP methods work for partial agent updates; update OpenAPI spec accordingly and regenerate SDK + CLI manifest - Fix #4: POST and PUT/PATCH /agents reject unknown request body fields with 400 "Unknown field(s): ..." to prevent silent data loss (e.g. sending prompt instead of instructions) - Fix #5: recordGenerationFailure now wraps non-DomainErrors in a new GENERATION_FAILED DomainError that carries generation_id and trace_id in meta, so 500 responses from /agents/:id/generate include a trace_id the caller can query for the root cause - Update ai-providers docs with JSON-format warning and config.api_key note Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01JTwheo15iV5tcGGWjk8V1z * test(generationLifecycle): cover non-Error thrown value to fix branch threshold Add a test that passes a plain object (not an Error or DomainError) to recordGenerationFailure, covering the false branch of the instanceof Error ternary that was sitting below the 70% branch coverage threshold. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01JTwheo15iV5tcGGWjk8V1z --------- Co-authored-by: Claude <noreply@anthropic.com>
arantespp
pushed a commit
that referenced
this pull request
Jul 18, 2026
…ation Restores global function coverage to 100%: - exercise emitApproval's create-time unique-violation backstop via a sanctioned force-failure spy (tests.md exception #2) - make runToolCallContinuation self-contained (never rejects) so the resume handler needs no separate .catch arrow; cover the swallow path Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AWGkJE9gG4tdJkaPxmtEFT
arantespp
added a commit
that referenced
this pull request
Jul 18, 2026
…ace (Milestone 1) (#581) * feat(agents,approvals): tool-call approval interception on every surface (Milestone 1) Milestone 1 of the manage-by-exception roadmap — extends the approval queue from orchestrated (DAG) runs to every execution surface (chat sessions, direct generations, MCP) via an `approval_policy` enforced in the platform tool-dispatch path. Task 1.0 (canonical tool_bindings + approval_policy validation) shipped earlier; this implements tasks 1.1–1.5. - 1.1 Policy evaluation in the dispatch path. A new pure module (`agentToolApproval.ts`) evaluates a binding's allow / require_approval / deny policy (JSON Logic over `{ action, arguments }`, first match wins) and gates each resolved tool's `execute`. Threaded into the resolver via a `ResolverApprovalContext` built at the generation entry point (and recovery path); zero-overhead when no binding declares a policy. - 1.2 Return-pending. `require_approval` freezes the call into an approval item and returns `{ status: "pending_approval", approval_id, expires_at }` as the tool result — the turn completes normally, no `requires_action` pause. The optional `approval_reasoning` / `approval_evidence` / `approval_predicted_impact` justification fields are injected into the model-visible schema (http/pipeline/inline), stripped from executed args, and frozen onto the item. - 1.3 Continuation on resolution. A tool-call resume handler (`agentToolApprovalContinuation.ts`, registered at startup) executes the frozen (or edited) action on approval, populates `DecisionOutput.result`, and fires a continuation generation linked via `initiator_generation_id` — routed into the originating session thread when there was one, else a standalone linked generation. Rejection/expiry carry the decision and execute nothing. - 1.4 Dedup. A `(project, agent, tool, action, args)` digest returns the existing pending item on re-proposal; the partial unique index backs the concurrent-emit race. - 1.5 `origin: tool_call` wired end-to-end; the lifecycle never branches on it. Model: `ApprovalItem` gains `session_id` and `proposed_action.action`; OpenAPI + docs (`approvals.md`, `agents.md`) updated; smoke-test flow added. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AWGkJE9gG4tdJkaPxmtEFT * fix(postgresdb): format ApprovalItem proposedAction type (lint) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AWGkJE9gG4tdJkaPxmtEFT * test(approvals): cover dedup-race backstop and self-contained continuation Restores global function coverage to 100%: - exercise emitApproval's create-time unique-violation backstop via a sanctioned force-failure spy (tests.md exception #2) - make runToolCallContinuation self-contained (never rejects) so the resume handler needs no separate .catch arrow; cover the swallow path Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AWGkJE9gG4tdJkaPxmtEFT * refactor(approvals): address PR review — gate semantics, provenance, registration - evaluateApprovalEffect: use plain JS truthiness for rule matching, so a `when` yielding 0/''/null/NaN is falsy (matches the JSON Logic convention used by orchestrationPollNode), instead of firing on 0. - Freeze the effective (preset-merged) arguments the policy evaluated onto the item (and dedup key), so the approver sees the full call and the platform re-executes exactly what was decided. - gateResolvedTools: spread the resolved tool in the schema-injection branch so properties beyond description/inputSchema/execute survive gating; fix the JSDoc to match localInjectableSchema (http/pipeline + inline http). - Register the tool-call resume handler from the approval expiry sweeper too, not only the REST router, so a scheduler-only process still fires continuations for items it expires. - Clarify the session-continuation provenance in JSDoc (thread linkage today; initiator_generation_id FK-threading through the session stack is a follow-up). - Document the inline-policy ordering invariant shared with deriveLegacyToolFields. - Add an end-to-end test that resolves an item through the real notifyResume path and asserts the registered handler fires the continuation. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AWGkJE9gG4tdJkaPxmtEFT --------- Co-authored-by: Claude <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.
No description provided.