opencode-swarm 6.29.2 → 6.29.4

package/README.md

@@ -37,11 +37,12 @@ Swarm then:
 5. Sends that plan through a critic gate before coding starts.
 6. Executes one task at a time through the QA pipeline:
 
-   * coder writes code
-   * automated checks run
-   * reviewer checks correctness
-   * test engineer writes and runs tests
-   * failures loop back with structured feedback
+* coder writes code
+* automated checks run
+* reviewer checks correctness
+* test engineer writes and runs tests
+* architect runs regression sweep (scope:"graph" to find cross-task test regressions)
+* failures loop back with structured feedback
 
 7. After each phase, docs and retrospectives are updated.
 
@@ -49,10 +50,13 @@ All project state lives in `.swarm/`:
 
 ```text
 .swarm/
-├── plan.md
-├── context.md
-├── evidence/
-└── history/
+├── plan.md              # Project roadmap with phases and tasks
+├── plan.json            # Structured plan data
+├── context.md           # Technical decisions and SME guidance
+├── events.jsonl         # Event stream for diagnostics
+├── evidence/            # Review/test evidence bundles per task
+├── curator-summary.json # Curator system state (if enabled)
+└── drift-report-phase-N.json  # Plan-vs-reality drift reports
 ```
 
 That means Swarm is resumable by design. If you come back later and `.swarm/` already exists, the architect may go straight into **RESUME** or **EXECUTE** instead of replaying the full first-run discovery flow.
@@ -325,8 +329,9 @@ MODE: EXECUTE (per task)
 ├── 5i. @reviewer (security pass, if security-sensitive files changed)
 ├── 5j. @test_engineer (verification tests + coverage ≥70%)
 ├── 5k. @test_engineer (adversarial tests)
-├── 5l. ⛔ Pre-commit checklist (all 4 items required, no override)
-└── 5m. Task marked complete, evidence written
+├── 5l. architect regression sweep (scope:"graph" to find cross-task test regressions)
+├── 5m. ⛔ Pre-commit checklist (all 4 items required, no override)
+└── 5n. Task marked complete, evidence written
 ```
 
 If any step fails, the coder gets structured feedback and retries. After 5 failures on the same task, it escalates to you.
@@ -936,6 +941,33 @@ The following tools can be assigned to agents via overrides:
 
 ## Recent Changes
 
+### v6.29.3 — Curator Visibility + Documentation Refresh
+
+This release adds Curator status reporting to `/swarm diagnose` and refreshes documentation to reflect current behavior.
+
+- **Curator status in diagnose**: `/swarm diagnose` now reports whether Curator is enabled/disabled and validates the `curator-summary.json` artifact if present.
+- **README and config docs refreshed**: Updated `.swarm/` directory tree, Curator configuration options, and drift report artifacts to match current implementation.
+
+### v6.29.2 — Multi-Language Incremental Verify + Slop-Detector Hardening
+
+This release adds multi-language typecheck support and hardens detection of low-quality AI-generated code.
+
+- **Multi-language incremental_verify**: Post-coder typecheck now supports TypeScript/JavaScript, Go, Rust, and C#. Runs language-appropriate build/typecheck commands per task for faster feedback.
+- **Slop-detector hardening**: Enhanced detection of AI-generated placeholder code with multi-language heuristics (comments like `// TODO`, `// FIXME`, `// HACK`, language-specific patterns for Go/Rust/C#/Python). Advisory warnings injected into coder context when slop patterns detected.
+- **CODEBASE REALITY CHECK**: Architect dispatches Explorer before planning to verify current state of referenced items. Produces a CODEBASE REALITY REPORT with statuses: NOT STARTED, PARTIALLY DONE, ALREADY COMPLETE, or ASSUMPTION INCORRECT. Skipped for greenfield projects.
+- **Evidence schema fix**: Evidence bundles now correctly validate against schema; invalid entries are filtered.
+- **Curator/docs alignment**: Curator health check in `/swarm diagnose` reports curator.enabled status and validates curator-summary.json.
+
+### v6.29.1 — Advisory Hook Message Injection
+
+This release improves how advisory warnings are injected into agent context.
+
+- **Advisory hook message injection**: Enhanced message formatting for self-coding detection, partial gate tracking, batch detection, and scope violation warnings. Messages now include structured context for faster resolution.
+
+### v6.23 through v6.28 — Stability and Incremental Improvements
+
+Subsequent releases focused on stability hardening, incremental_verify scaffolding, and tooling improvements across the pipeline. Key areas: pre_check_batch parallelization, evidence bundle validation, and diagnostics tooling.
+
 ### v6.22 — Curator Background Analysis + Session State Persistence
 
 This release adds the optional Curator system for phase-level intelligence and fixes session snapshot persistence for task workflow states.
@@ -1118,12 +1150,16 @@ OpenCode Swarm v6.16+ ships with language profiles for 11 languages across three
 
 The Curator is an optional background analysis system that runs after each phase. It is **disabled by default** (`curator.enabled = false`) and never blocks execution — all Curator operations are wrapped in try/catch.
 
+To enable, set `"curator": { "enabled": true }` in your config. When enabled, it writes `.swarm/curator-summary.json` and `.swarm/drift-report-phase-N.json` files.
+
 ### What the Curator Does
 
-- **Init** (`phase-monitor.ts`): On the first phase, initializes a curator summary file at `.swarm/curator-summary.json`.
+- **Init** (`phase-monitor.ts`): On the first phase, initializes a curator summary file at `.swarm/curator-summary.json` and persists the init briefing to `.swarm/curator-briefing.md`.
 - **Phase analysis** (`phase-complete.ts`): After each phase completes, collects phase events, checks compliance, and optionally invokes the curator explorer to summarize findings.
+- **Compliance surfacing** (`phase-complete.ts`): Compliance observations are surfaced in the return value's warnings array (unless `suppress_warnings` is true).
 - **Knowledge updates** (`phase-complete.ts`): Merges curator findings into the knowledge base up to the configured `max_summary_tokens` cap.
-- **Drift injection** (`knowledge-injector.ts`): Prepends the latest drift report summary to the architect's knowledge context at phase start, up to `drift_inject_max_chars` characters.
+- **Briefing injection** (`knowledge-injector.ts`): The curator-briefing.md content is injected into the architect's context at session start.
+- **Drift injection** (`knowledge-injector.ts`): Prepends the latest drift report summary to the architect's knowledge context at phase start, up to `drift_inject_max_chars` characters. Drift reports now inject even when no knowledge entries exist.
 
 ### Configuration
 

package/dist/cli/index.js

@@ -34775,7 +34775,8 @@ var test_runner = createSwarmTool({
     scope: tool.schema.enum(["all", "convention", "graph"]).optional().describe('Test scope: "all" runs full suite, "convention" maps source files to test files by naming, "graph" finds related tests via imports'),
     files: tool.schema.array(tool.schema.string()).optional().describe("Specific files to test (used with convention or graph scope)"),
     coverage: tool.schema.boolean().optional().describe("Enable coverage reporting if supported"),
-    timeout_ms: tool.schema.number().optional().describe("Timeout in milliseconds (default 60000, max 300000)")
+    timeout_ms: tool.schema.number().optional().describe("Timeout in milliseconds (default 60000, max 300000)"),
+    allow_full_suite: tool.schema.boolean().optional().describe('Explicit opt-in for scope "all". Required because full-suite output can destabilize SSE streaming.')
   },
   async execute(args, directory) {
     const workingDir = directory.trim() || directory;
@@ -34827,14 +34828,16 @@ var test_runner = createSwarmTool({
     }
     const scope = args.scope || "all";
     if (scope === "all") {
-      const errorResult = {
-        success: false,
-        framework: "none",
-        scope: "all",
-        error: 'Full-suite test execution (scope: "all") is prohibited in interactive sessions',
-        message: 'Use scope "convention" or "graph" with explicit files to run targeted tests in interactive mode. Full-suite runs are restricted to prevent excessive resource consumption.'
-      };
-      return JSON.stringify(errorResult, null, 2);
+      if (!args.allow_full_suite) {
+        const errorResult = {
+          success: false,
+          framework: "none",
+          scope: "all",
+          error: 'Full-suite test execution (scope: "all") requires allow_full_suite: true',
+          message: 'Set allow_full_suite: true to confirm intentional full-suite execution. Use scope "convention" or "graph" for targeted tests. Full-suite output is large and may destabilize SSE streaming on some opencode versions.'
+        };
+        return JSON.stringify(errorResult, null, 2);
+      }
     }
     if ((scope === "convention" || scope === "graph") && (!args.files || args.files.length === 0)) {
       const errorResult = {
@@ -34869,7 +34872,7 @@ var test_runner = createSwarmTool({
     let testFiles = [];
     let graphFallbackReason;
     let effectiveScope = scope;
-    if (scope === "convention") {
+    if (scope === "all") {} else if (scope === "convention") {
       const sourceFiles = args.files.filter((f) => {
         const ext = path13.extname(f).toLowerCase();
         return SOURCE_EXTENSIONS.has(ext);
@@ -34909,7 +34912,7 @@ var test_runner = createSwarmTool({
         testFiles = getTestFilesFromConvention(sourceFiles);
       }
     }
-    if (testFiles.length === 0) {
+    if (scope !== "all" && testFiles.length === 0) {
       const errorResult = {
         success: false,
         framework,
@@ -34919,7 +34922,7 @@ var test_runner = createSwarmTool({
       };
       return JSON.stringify(errorResult, null, 2);
     }
-    if (testFiles.length > MAX_SAFE_TEST_FILES) {
+    if (scope !== "all" && testFiles.length > MAX_SAFE_TEST_FILES) {
       const sampleFiles = testFiles.slice(0, 5);
       const errorResult = {
         success: false,

package/dist/hooks/incremental-verify.d.ts

@@ -24,7 +24,8 @@ export declare function resetAdvisoryDedup(): void;
  * or null overall if no supported language is detected.
  * Checks in order: TypeScript (package.json) → Go (go.mod) → Rust (Cargo.toml)
  * → Python (pyproject.toml/requirements.txt/setup.py) → C# (*.csproj/*.sln)
- * First match wins; package.json presence means Node/Bun project — no fallthrough.
+ * First match wins; package.json only short-circuits when TypeScript markers are present.
+ * Otherwise the function falls through to Go/Rust/Python/C# detection.
  */
 declare function detectTypecheckCommand(projectDir: string): {
     command: string[] | null;