opencode-swarm-plugin 0.43.0 → 0.44.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (208) hide show
  1. package/bin/cass.characterization.test.ts +422 -0
  2. package/bin/swarm.serve.test.ts +6 -4
  3. package/bin/swarm.test.ts +68 -0
  4. package/bin/swarm.ts +81 -8
  5. package/dist/compaction-prompt-scoring.js +139 -0
  6. package/dist/contributor-tools.d.ts +42 -0
  7. package/dist/contributor-tools.d.ts.map +1 -0
  8. package/dist/eval-capture.js +12811 -0
  9. package/dist/hive.d.ts.map +1 -1
  10. package/dist/index.d.ts +12 -0
  11. package/dist/index.d.ts.map +1 -1
  12. package/dist/index.js +7728 -62590
  13. package/dist/plugin.js +23833 -78695
  14. package/dist/sessions/agent-discovery.d.ts +59 -0
  15. package/dist/sessions/agent-discovery.d.ts.map +1 -0
  16. package/dist/sessions/index.d.ts +10 -0
  17. package/dist/sessions/index.d.ts.map +1 -0
  18. package/dist/swarm-orchestrate.d.ts.map +1 -1
  19. package/dist/swarm-prompts.d.ts.map +1 -1
  20. package/dist/swarm-review.d.ts.map +1 -1
  21. package/package.json +17 -5
  22. package/.changeset/swarm-insights-data-layer.md +0 -63
  23. package/.hive/analysis/eval-failure-analysis-2025-12-25.md +0 -331
  24. package/.hive/analysis/session-data-quality-audit.md +0 -320
  25. package/.hive/eval-results.json +0 -483
  26. package/.hive/issues.jsonl +0 -138
  27. package/.hive/memories.jsonl +0 -729
  28. package/.opencode/eval-history.jsonl +0 -327
  29. package/.turbo/turbo-build.log +0 -9
  30. package/CHANGELOG.md +0 -2255
  31. package/SCORER-ANALYSIS.md +0 -598
  32. package/docs/analysis/subagent-coordination-patterns.md +0 -902
  33. package/docs/analysis-socratic-planner-pattern.md +0 -504
  34. package/docs/planning/ADR-001-monorepo-structure.md +0 -171
  35. package/docs/planning/ADR-002-package-extraction.md +0 -393
  36. package/docs/planning/ADR-003-performance-improvements.md +0 -451
  37. package/docs/planning/ADR-004-message-queue-features.md +0 -187
  38. package/docs/planning/ADR-005-devtools-observability.md +0 -202
  39. package/docs/planning/ADR-007-swarm-enhancements-worktree-review.md +0 -168
  40. package/docs/planning/ADR-008-worker-handoff-protocol.md +0 -293
  41. package/docs/planning/ADR-009-oh-my-opencode-patterns.md +0 -353
  42. package/docs/planning/ROADMAP.md +0 -368
  43. package/docs/semantic-memory-cli-syntax.md +0 -123
  44. package/docs/swarm-mail-architecture.md +0 -1147
  45. package/docs/testing/context-recovery-test.md +0 -470
  46. package/evals/ARCHITECTURE.md +0 -1189
  47. package/evals/README.md +0 -768
  48. package/evals/compaction-prompt.eval.ts +0 -149
  49. package/evals/compaction-resumption.eval.ts +0 -289
  50. package/evals/coordinator-behavior.eval.ts +0 -307
  51. package/evals/coordinator-session.eval.ts +0 -154
  52. package/evals/evalite.config.ts.bak +0 -15
  53. package/evals/example.eval.ts +0 -31
  54. package/evals/fixtures/compaction-cases.ts +0 -350
  55. package/evals/fixtures/compaction-prompt-cases.ts +0 -311
  56. package/evals/fixtures/coordinator-sessions.ts +0 -328
  57. package/evals/fixtures/decomposition-cases.ts +0 -105
  58. package/evals/lib/compaction-loader.test.ts +0 -248
  59. package/evals/lib/compaction-loader.ts +0 -320
  60. package/evals/lib/data-loader.evalite-test.ts +0 -289
  61. package/evals/lib/data-loader.test.ts +0 -345
  62. package/evals/lib/data-loader.ts +0 -281
  63. package/evals/lib/llm.ts +0 -115
  64. package/evals/scorers/compaction-prompt-scorers.ts +0 -145
  65. package/evals/scorers/compaction-scorers.ts +0 -305
  66. package/evals/scorers/coordinator-discipline.evalite-test.ts +0 -539
  67. package/evals/scorers/coordinator-discipline.ts +0 -325
  68. package/evals/scorers/index.test.ts +0 -146
  69. package/evals/scorers/index.ts +0 -328
  70. package/evals/scorers/outcome-scorers.evalite-test.ts +0 -27
  71. package/evals/scorers/outcome-scorers.ts +0 -349
  72. package/evals/swarm-decomposition.eval.ts +0 -121
  73. package/examples/commands/swarm.md +0 -745
  74. package/examples/plugin-wrapper-template.ts +0 -2426
  75. package/examples/skills/hive-workflow/SKILL.md +0 -212
  76. package/examples/skills/skill-creator/SKILL.md +0 -223
  77. package/examples/skills/swarm-coordination/SKILL.md +0 -292
  78. package/global-skills/cli-builder/SKILL.md +0 -344
  79. package/global-skills/cli-builder/references/advanced-patterns.md +0 -244
  80. package/global-skills/learning-systems/SKILL.md +0 -644
  81. package/global-skills/skill-creator/LICENSE.txt +0 -202
  82. package/global-skills/skill-creator/SKILL.md +0 -352
  83. package/global-skills/skill-creator/references/output-patterns.md +0 -82
  84. package/global-skills/skill-creator/references/workflows.md +0 -28
  85. package/global-skills/swarm-coordination/SKILL.md +0 -995
  86. package/global-skills/swarm-coordination/references/coordinator-patterns.md +0 -235
  87. package/global-skills/swarm-coordination/references/strategies.md +0 -138
  88. package/global-skills/system-design/SKILL.md +0 -213
  89. package/global-skills/testing-patterns/SKILL.md +0 -430
  90. package/global-skills/testing-patterns/references/dependency-breaking-catalog.md +0 -586
  91. package/opencode-swarm-plugin-0.30.7.tgz +0 -0
  92. package/opencode-swarm-plugin-0.31.0.tgz +0 -0
  93. package/scripts/cleanup-test-memories.ts +0 -346
  94. package/scripts/init-skill.ts +0 -222
  95. package/scripts/migrate-unknown-sessions.ts +0 -349
  96. package/scripts/validate-skill.ts +0 -204
  97. package/src/agent-mail.ts +0 -1724
  98. package/src/anti-patterns.test.ts +0 -1167
  99. package/src/anti-patterns.ts +0 -448
  100. package/src/compaction-capture.integration.test.ts +0 -257
  101. package/src/compaction-hook.test.ts +0 -838
  102. package/src/compaction-hook.ts +0 -1204
  103. package/src/compaction-observability.integration.test.ts +0 -139
  104. package/src/compaction-observability.test.ts +0 -187
  105. package/src/compaction-observability.ts +0 -324
  106. package/src/compaction-prompt-scorers.test.ts +0 -475
  107. package/src/compaction-prompt-scoring.ts +0 -300
  108. package/src/dashboard.test.ts +0 -611
  109. package/src/dashboard.ts +0 -462
  110. package/src/error-enrichment.test.ts +0 -403
  111. package/src/error-enrichment.ts +0 -219
  112. package/src/eval-capture.test.ts +0 -1015
  113. package/src/eval-capture.ts +0 -929
  114. package/src/eval-gates.test.ts +0 -306
  115. package/src/eval-gates.ts +0 -218
  116. package/src/eval-history.test.ts +0 -508
  117. package/src/eval-history.ts +0 -214
  118. package/src/eval-learning.test.ts +0 -378
  119. package/src/eval-learning.ts +0 -360
  120. package/src/eval-runner.test.ts +0 -223
  121. package/src/eval-runner.ts +0 -402
  122. package/src/export-tools.test.ts +0 -476
  123. package/src/export-tools.ts +0 -257
  124. package/src/hive.integration.test.ts +0 -2241
  125. package/src/hive.ts +0 -1628
  126. package/src/index.ts +0 -935
  127. package/src/learning.integration.test.ts +0 -1815
  128. package/src/learning.ts +0 -1079
  129. package/src/logger.test.ts +0 -189
  130. package/src/logger.ts +0 -135
  131. package/src/mandate-promotion.test.ts +0 -473
  132. package/src/mandate-promotion.ts +0 -239
  133. package/src/mandate-storage.integration.test.ts +0 -601
  134. package/src/mandate-storage.test.ts +0 -578
  135. package/src/mandate-storage.ts +0 -794
  136. package/src/mandates.ts +0 -540
  137. package/src/memory-tools.test.ts +0 -195
  138. package/src/memory-tools.ts +0 -344
  139. package/src/memory.integration.test.ts +0 -334
  140. package/src/memory.test.ts +0 -158
  141. package/src/memory.ts +0 -527
  142. package/src/model-selection.test.ts +0 -188
  143. package/src/model-selection.ts +0 -68
  144. package/src/observability-tools.test.ts +0 -359
  145. package/src/observability-tools.ts +0 -871
  146. package/src/output-guardrails.test.ts +0 -438
  147. package/src/output-guardrails.ts +0 -381
  148. package/src/pattern-maturity.test.ts +0 -1160
  149. package/src/pattern-maturity.ts +0 -525
  150. package/src/planning-guardrails.test.ts +0 -491
  151. package/src/planning-guardrails.ts +0 -438
  152. package/src/plugin.ts +0 -23
  153. package/src/post-compaction-tracker.test.ts +0 -251
  154. package/src/post-compaction-tracker.ts +0 -237
  155. package/src/query-tools.test.ts +0 -636
  156. package/src/query-tools.ts +0 -324
  157. package/src/rate-limiter.integration.test.ts +0 -466
  158. package/src/rate-limiter.ts +0 -774
  159. package/src/replay-tools.test.ts +0 -496
  160. package/src/replay-tools.ts +0 -240
  161. package/src/repo-crawl.integration.test.ts +0 -441
  162. package/src/repo-crawl.ts +0 -610
  163. package/src/schemas/cell-events.test.ts +0 -347
  164. package/src/schemas/cell-events.ts +0 -807
  165. package/src/schemas/cell.ts +0 -257
  166. package/src/schemas/evaluation.ts +0 -166
  167. package/src/schemas/index.test.ts +0 -199
  168. package/src/schemas/index.ts +0 -286
  169. package/src/schemas/mandate.ts +0 -232
  170. package/src/schemas/swarm-context.ts +0 -115
  171. package/src/schemas/task.ts +0 -161
  172. package/src/schemas/worker-handoff.test.ts +0 -302
  173. package/src/schemas/worker-handoff.ts +0 -131
  174. package/src/skills.integration.test.ts +0 -1192
  175. package/src/skills.test.ts +0 -643
  176. package/src/skills.ts +0 -1549
  177. package/src/storage.integration.test.ts +0 -341
  178. package/src/storage.ts +0 -884
  179. package/src/structured.integration.test.ts +0 -817
  180. package/src/structured.test.ts +0 -1046
  181. package/src/structured.ts +0 -762
  182. package/src/swarm-decompose.test.ts +0 -188
  183. package/src/swarm-decompose.ts +0 -1302
  184. package/src/swarm-deferred.integration.test.ts +0 -157
  185. package/src/swarm-deferred.test.ts +0 -38
  186. package/src/swarm-insights.test.ts +0 -214
  187. package/src/swarm-insights.ts +0 -459
  188. package/src/swarm-mail.integration.test.ts +0 -970
  189. package/src/swarm-mail.ts +0 -739
  190. package/src/swarm-orchestrate.integration.test.ts +0 -282
  191. package/src/swarm-orchestrate.test.ts +0 -548
  192. package/src/swarm-orchestrate.ts +0 -3084
  193. package/src/swarm-prompts.test.ts +0 -1270
  194. package/src/swarm-prompts.ts +0 -2077
  195. package/src/swarm-research.integration.test.ts +0 -701
  196. package/src/swarm-research.test.ts +0 -698
  197. package/src/swarm-research.ts +0 -472
  198. package/src/swarm-review.integration.test.ts +0 -285
  199. package/src/swarm-review.test.ts +0 -879
  200. package/src/swarm-review.ts +0 -709
  201. package/src/swarm-strategies.ts +0 -407
  202. package/src/swarm-worktree.test.ts +0 -501
  203. package/src/swarm-worktree.ts +0 -575
  204. package/src/swarm.integration.test.ts +0 -2377
  205. package/src/swarm.ts +0 -38
  206. package/src/tool-adapter.integration.test.ts +0 -1221
  207. package/src/tool-availability.ts +0 -461
  208. package/tsconfig.json +0 -28
@@ -1,817 +0,0 @@
1
- /**
2
- * Integration tests for structured_* tools
3
- *
4
- * Tests the complete tool execution flow:
5
- * - Tool argument parsing and validation
6
- * - JSON extraction from various formats
7
- * - Schema validation with structured output
8
- * - Error handling and feedback formatting
9
- *
10
- * ## Test Pattern
11
- * - Call tool.execute() directly (simulates plugin invocation)
12
- * - Use mock ToolContext
13
- * - Test happy paths with real-world scenarios
14
- * - No external dependencies (pure JSON parsing)
15
- */
16
-
17
- import type { ToolContext } from "@opencode-ai/plugin";
18
- import { describe, expect, it } from "bun:test";
19
- import {
20
- structured_extract_json,
21
- structured_parse_cell_tree,
22
- structured_parse_decomposition,
23
- structured_parse_evaluation,
24
- structured_validate,
25
- } from "./structured";
26
-
27
- // Mock ToolContext for tool execution
28
- const mockCtx = {} as ToolContext;
29
-
30
- // ============================================================================
31
- // structured_extract_json - JSON Extraction
32
- // ============================================================================
33
-
34
- describe("structured_extract_json integration", () => {
35
- it("extracts clean JSON and returns success", async () => {
36
- const result = await structured_extract_json.execute(
37
- { text: '{"name": "Joel", "role": "developer"}' },
38
- mockCtx,
39
- );
40
-
41
- const parsed = JSON.parse(result);
42
- expect(parsed.success).toBe(true);
43
- expect(parsed.data).toEqual({ name: "Joel", role: "developer" });
44
- expect(parsed.extraction_method).toBe("direct_parse");
45
- });
46
-
47
- it("extracts JSON from markdown code block", async () => {
48
- const markdown = `Here is the response:\n\`\`\`json\n{"status": "complete"}\n\`\`\``;
49
-
50
- const result = await structured_extract_json.execute(
51
- { text: markdown },
52
- mockCtx,
53
- );
54
-
55
- const parsed = JSON.parse(result);
56
- expect(parsed.success).toBe(true);
57
- expect(parsed.data).toEqual({ status: "complete" });
58
- expect(parsed.extraction_method).toBe("json_code_block");
59
- });
60
-
61
- it("extracts JSON from mixed content with multiple code blocks", async () => {
62
- const mixed = `
63
- Some text here.
64
-
65
- \`\`\`typescript
66
- const code = "not json";
67
- \`\`\`
68
-
69
- Result:
70
-
71
- \`\`\`json
72
- {"result": "success"}
73
- \`\`\`
74
- `;
75
-
76
- const result = await structured_extract_json.execute(
77
- { text: mixed },
78
- mockCtx,
79
- );
80
-
81
- const parsed = JSON.parse(result);
82
- expect(parsed.success).toBe(true);
83
- expect(parsed.data).toEqual({ result: "success" });
84
- expect(parsed.extraction_method).toBe("json_code_block");
85
- });
86
-
87
- it("handles complex nested structures", async () => {
88
- const complex = JSON.stringify({
89
- epic: {
90
- title: "Add authentication",
91
- description: "OAuth + session management",
92
- },
93
- subtasks: [
94
- {
95
- title: "OAuth integration",
96
- files: ["src/auth/oauth.ts"],
97
- estimated_complexity: 3,
98
- },
99
- {
100
- title: "Session store",
101
- files: ["src/auth/sessions.ts"],
102
- estimated_complexity: 2,
103
- },
104
- ],
105
- });
106
-
107
- const result = await structured_extract_json.execute(
108
- { text: complex },
109
- mockCtx,
110
- );
111
-
112
- const parsed = JSON.parse(result);
113
- expect(parsed.success).toBe(true);
114
- expect(parsed.data.epic.title).toBe("Add authentication");
115
- expect(parsed.data.subtasks).toHaveLength(2);
116
- });
117
-
118
- it("returns error for non-JSON input", async () => {
119
- const result = await structured_extract_json.execute(
120
- { text: "This is not JSON at all" },
121
- mockCtx,
122
- );
123
-
124
- const parsed = JSON.parse(result);
125
- expect(parsed.success).toBe(false);
126
- expect(parsed.error).toContain("Could not extract");
127
- expect(Array.isArray(parsed.attempted_strategies)).toBe(true);
128
- expect(parsed.attempted_strategies.length).toBeGreaterThan(0);
129
- });
130
-
131
- it("includes input preview in error for long text", async () => {
132
- const longText = "x".repeat(300);
133
-
134
- const result = await structured_extract_json.execute(
135
- { text: longText },
136
- mockCtx,
137
- );
138
-
139
- const parsed = JSON.parse(result);
140
- expect(parsed.success).toBe(false);
141
- expect(parsed.raw_input_preview).toBeDefined();
142
- expect(parsed.raw_input_preview.length).toBeLessThanOrEqual(200);
143
- });
144
- });
145
-
146
- // ============================================================================
147
- // structured_validate - Schema Validation
148
- // ============================================================================
149
-
150
- describe("structured_validate integration", () => {
151
- describe("evaluation schema", () => {
152
- it("validates correct evaluation structure", async () => {
153
- const evalObj = {
154
- passed: true,
155
- criteria: {
156
- type_safety: { passed: true, feedback: "All types validated" },
157
- no_bugs: { passed: true, feedback: "No issues found" },
158
- },
159
- overall_feedback: "Excellent work",
160
- retry_suggestion: null,
161
- };
162
-
163
- const result = await structured_validate.execute(
164
- {
165
- response: JSON.stringify(evalObj),
166
- schema_name: "evaluation",
167
- },
168
- mockCtx,
169
- );
170
-
171
- const parsed = JSON.parse(result);
172
- expect(parsed.success).toBe(true);
173
- expect(parsed.data.passed).toBe(true);
174
- expect(Object.keys(parsed.data.criteria)).toHaveLength(2);
175
- });
176
-
177
- it("extracts JSON from markdown before validating", async () => {
178
- const evalObj = {
179
- passed: false,
180
- criteria: {
181
- test: { passed: false, feedback: "Test failed" },
182
- },
183
- overall_feedback: "Needs fixes",
184
- retry_suggestion: "Add tests",
185
- };
186
-
187
- const markdown = `Analysis complete:\n\`\`\`json\n${JSON.stringify(evalObj)}\n\`\`\``;
188
-
189
- const result = await structured_validate.execute(
190
- {
191
- response: markdown,
192
- schema_name: "evaluation",
193
- },
194
- mockCtx,
195
- );
196
-
197
- const parsed = JSON.parse(result);
198
- expect(parsed.success).toBe(true);
199
- expect(parsed.extractionMethod).toBe("json_code_block");
200
- });
201
-
202
- it("returns structured errors for invalid evaluation", async () => {
203
- const result = await structured_validate.execute(
204
- {
205
- response: '{"invalid": true}',
206
- schema_name: "evaluation",
207
- },
208
- mockCtx,
209
- );
210
-
211
- const parsed = JSON.parse(result);
212
- expect(parsed.success).toBe(false);
213
- expect(Array.isArray(parsed.errors)).toBe(true);
214
- expect(parsed.errors.length).toBeGreaterThan(0);
215
- });
216
-
217
- it("handles empty response gracefully", async () => {
218
- const result = await structured_validate.execute(
219
- {
220
- response: "",
221
- schema_name: "evaluation",
222
- },
223
- mockCtx,
224
- );
225
-
226
- const parsed = JSON.parse(result);
227
- expect(parsed.valid).toBe(false);
228
- expect(parsed.error).toContain("empty");
229
- });
230
- });
231
-
232
- describe("task_decomposition schema", () => {
233
- it("validates correct decomposition structure", async () => {
234
- const decompObj = {
235
- task: "Implement authentication",
236
- reasoning: "Split by feature layer",
237
- subtasks: [
238
- {
239
- title: "Auth service",
240
- description: "Core logic",
241
- files: ["src/auth.ts"],
242
- estimated_effort: "medium",
243
- },
244
- {
245
- title: "Auth UI",
246
- description: "Login form",
247
- files: ["src/components/Login.tsx"],
248
- estimated_effort: "small",
249
- },
250
- ],
251
- };
252
-
253
- const result = await structured_validate.execute(
254
- {
255
- response: JSON.stringify(decompObj),
256
- schema_name: "task_decomposition",
257
- },
258
- mockCtx,
259
- );
260
-
261
- const parsed = JSON.parse(result);
262
- expect(parsed.success).toBe(true);
263
- expect(parsed.data.task).toBe("Implement authentication");
264
- expect(parsed.data.subtasks).toHaveLength(2);
265
- });
266
- });
267
-
268
- describe("cell_tree schema", () => {
269
- it("validates correct cell tree structure", async () => {
270
- const treeObj = {
271
- epic: {
272
- title: "Add authentication",
273
- description: "OAuth + sessions",
274
- },
275
- subtasks: [
276
- {
277
- title: "OAuth integration",
278
- description: "Connect to provider",
279
- files: ["src/auth/oauth.ts"],
280
- dependencies: [],
281
- estimated_complexity: 3,
282
- },
283
- {
284
- title: "Session store",
285
- files: ["src/auth/sessions.ts"],
286
- dependencies: [0],
287
- estimated_complexity: 2,
288
- },
289
- ],
290
- };
291
-
292
- const result = await structured_validate.execute(
293
- {
294
- response: JSON.stringify(treeObj),
295
- schema_name: "cell_tree",
296
- },
297
- mockCtx,
298
- );
299
-
300
- const parsed = JSON.parse(result);
301
- expect(parsed.success).toBe(true);
302
- expect(parsed.data.epic.title).toBe("Add authentication");
303
- expect(parsed.data.subtasks).toHaveLength(2);
304
- });
305
- });
306
-
307
- it("includes retry hint when max_retries provided", async () => {
308
- const result = await structured_validate.execute(
309
- {
310
- response: '{"invalid": true}',
311
- schema_name: "evaluation",
312
- max_retries: 3,
313
- },
314
- mockCtx,
315
- );
316
-
317
- const parsed = JSON.parse(result);
318
- expect(parsed.success).toBe(false);
319
- const hasRetryHint = parsed.errors.some((e: string) =>
320
- e.includes("try again"),
321
- );
322
- expect(hasRetryHint).toBe(true);
323
- });
324
- });
325
-
326
- // ============================================================================
327
- // structured_parse_evaluation - Typed Evaluation Parsing
328
- // ============================================================================
329
-
330
- describe("structured_parse_evaluation integration", () => {
331
- it("parses valid evaluation with summary", async () => {
332
- const evalObj = {
333
- passed: true,
334
- criteria: {
335
- type_safe: { passed: true, feedback: "All types validated" },
336
- no_bugs: { passed: true, feedback: "No issues found" },
337
- patterns: { passed: true, feedback: "Follows best practices" },
338
- },
339
- overall_feedback: "Excellent implementation",
340
- retry_suggestion: null,
341
- };
342
-
343
- const result = await structured_parse_evaluation.execute(
344
- { response: JSON.stringify(evalObj) },
345
- mockCtx,
346
- );
347
-
348
- const parsed = JSON.parse(result);
349
- expect(parsed.success).toBe(true);
350
- expect(parsed.data.passed).toBe(true);
351
- expect(parsed.summary.passed).toBe(true);
352
- expect(parsed.summary.criteria_count).toBe(3);
353
- expect(parsed.summary.failed_criteria).toEqual([]);
354
- });
355
-
356
- it("identifies failed criteria in summary", async () => {
357
- const evalObj = {
358
- passed: false,
359
- criteria: {
360
- type_safe: { passed: true, feedback: "OK" },
361
- no_bugs: { passed: false, feedback: "Found null pointer" },
362
- patterns: { passed: false, feedback: "Missing error handling" },
363
- },
364
- overall_feedback: "Needs fixes",
365
- retry_suggestion: "Add null checks and error handling",
366
- };
367
-
368
- const result = await structured_parse_evaluation.execute(
369
- { response: JSON.stringify(evalObj) },
370
- mockCtx,
371
- );
372
-
373
- const parsed = JSON.parse(result);
374
- expect(parsed.success).toBe(true);
375
- expect(parsed.summary.passed).toBe(false);
376
- expect(parsed.summary.failed_criteria).toContain("no_bugs");
377
- expect(parsed.summary.failed_criteria).toContain("patterns");
378
- expect(parsed.summary.failed_criteria).not.toContain("type_safe");
379
- });
380
-
381
- it("returns structured error for malformed JSON", async () => {
382
- const result = await structured_parse_evaluation.execute(
383
- { response: "not json" },
384
- mockCtx,
385
- );
386
-
387
- const parsed = JSON.parse(result);
388
- expect(parsed.success).toBe(false);
389
- expect(parsed.error).toContain("extract JSON");
390
- expect(parsed.feedback).toBeDefined();
391
- expect(Array.isArray(parsed.attempted_strategies)).toBe(true);
392
- });
393
-
394
- it("returns structured error for invalid schema", async () => {
395
- const invalidEval = {
396
- passed: "not a boolean",
397
- criteria: {},
398
- overall_feedback: "test",
399
- retry_suggestion: null,
400
- };
401
-
402
- const result = await structured_parse_evaluation.execute(
403
- { response: JSON.stringify(invalidEval) },
404
- mockCtx,
405
- );
406
-
407
- const parsed = JSON.parse(result);
408
- expect(parsed.success).toBe(false);
409
- expect(parsed.error).toContain("does not match schema");
410
- expect(Array.isArray(parsed.validation_errors)).toBe(true);
411
- expect(parsed.expected_shape).toBeDefined();
412
- });
413
- });
414
-
415
- // ============================================================================
416
- // structured_parse_decomposition - Typed Decomposition Parsing
417
- // ============================================================================
418
-
419
- describe("structured_parse_decomposition integration", () => {
420
- it("parses valid decomposition with summary", async () => {
421
- const decompObj = {
422
- task: "Implement authentication",
423
- reasoning: "Split by feature layer for parallel development",
424
- subtasks: [
425
- {
426
- title: "Auth service",
427
- description: "Core authentication logic",
428
- files: ["src/auth/service.ts", "src/auth/types.ts"],
429
- estimated_effort: "medium",
430
- },
431
- {
432
- title: "Auth UI",
433
- description: "Login and signup forms",
434
- files: ["src/components/Login.tsx", "src/components/Signup.tsx"],
435
- estimated_effort: "small",
436
- },
437
- {
438
- title: "Auth API",
439
- description: "REST endpoints",
440
- files: ["src/api/auth.ts"],
441
- estimated_effort: "small",
442
- },
443
- ],
444
- };
445
-
446
- const result = await structured_parse_decomposition.execute(
447
- { response: JSON.stringify(decompObj) },
448
- mockCtx,
449
- );
450
-
451
- const parsed = JSON.parse(result);
452
- expect(parsed.success).toBe(true);
453
- expect(parsed.data.task).toBe("Implement authentication");
454
- expect(parsed.summary.subtask_count).toBe(3);
455
- expect(parsed.summary.total_files).toBe(5);
456
- expect(parsed.summary.files).toContain("src/auth/service.ts");
457
- });
458
-
459
- it("includes effort breakdown in summary", async () => {
460
- const decompObj = {
461
- task: "Test task",
462
- subtasks: [
463
- {
464
- title: "T1",
465
- description: "D1",
466
- files: ["a.ts"],
467
- estimated_effort: "small",
468
- },
469
- {
470
- title: "T2",
471
- description: "D2",
472
- files: ["b.ts"],
473
- estimated_effort: "small",
474
- },
475
- {
476
- title: "T3",
477
- description: "D3",
478
- files: ["c.ts"],
479
- estimated_effort: "medium",
480
- },
481
- {
482
- title: "T4",
483
- description: "D4",
484
- files: ["d.ts"],
485
- estimated_effort: "large",
486
- },
487
- ],
488
- };
489
-
490
- const result = await structured_parse_decomposition.execute(
491
- { response: JSON.stringify(decompObj) },
492
- mockCtx,
493
- );
494
-
495
- const parsed = JSON.parse(result);
496
- expect(parsed.success).toBe(true);
497
- expect(parsed.summary.effort_breakdown.small).toBe(2);
498
- expect(parsed.summary.effort_breakdown.medium).toBe(1);
499
- expect(parsed.summary.effort_breakdown.large).toBe(1);
500
- });
501
-
502
- it("deduplicates files in summary", async () => {
503
- const decompObj = {
504
- task: "Test task",
505
- subtasks: [
506
- {
507
- title: "T1",
508
- description: "D1",
509
- files: ["shared.ts", "a.ts"],
510
- estimated_effort: "small",
511
- },
512
- {
513
- title: "T2",
514
- description: "D2",
515
- files: ["shared.ts", "b.ts"],
516
- estimated_effort: "small",
517
- },
518
- ],
519
- };
520
-
521
- const result = await structured_parse_decomposition.execute(
522
- { response: JSON.stringify(decompObj) },
523
- mockCtx,
524
- );
525
-
526
- const parsed = JSON.parse(result);
527
- expect(parsed.success).toBe(true);
528
- expect(parsed.summary.total_files).toBe(3);
529
- expect(parsed.summary.files).toEqual(["shared.ts", "a.ts", "b.ts"]);
530
- });
531
-
532
- it("handles dependencies in summary", async () => {
533
- const decompObj = {
534
- task: "Test task",
535
- subtasks: [
536
- {
537
- title: "T1",
538
- description: "D1",
539
- files: ["a.ts"],
540
- estimated_effort: "small",
541
- },
542
- {
543
- title: "T2",
544
- description: "D2",
545
- files: ["b.ts"],
546
- estimated_effort: "small",
547
- },
548
- ],
549
- dependencies: [
550
- { from: 0, to: 1, type: "blocks" },
551
- { from: 1, to: 0, type: "requires" },
552
- ],
553
- };
554
-
555
- const result = await structured_parse_decomposition.execute(
556
- { response: JSON.stringify(decompObj) },
557
- mockCtx,
558
- );
559
-
560
- const parsed = JSON.parse(result);
561
- expect(parsed.success).toBe(true);
562
- expect(parsed.summary.dependency_count).toBe(2);
563
- });
564
-
565
- it("returns error for invalid decomposition", async () => {
566
- const result = await structured_parse_decomposition.execute(
567
- { response: '{"task": "Test"}' }, // Missing required subtasks
568
- mockCtx,
569
- );
570
-
571
- const parsed = JSON.parse(result);
572
- expect(parsed.success).toBe(false);
573
- expect(parsed.error).toContain("does not match schema");
574
- expect(parsed.expected_shape).toBeDefined();
575
- });
576
- });
577
-
578
- // ============================================================================
579
- // structured_parse_cell_tree - Typed Cell Tree Parsing
580
- // ============================================================================
581
-
582
- describe("structured_parse_cell_tree integration", () => {
583
- it("parses valid cell tree with summary", async () => {
584
- const treeObj = {
585
- epic: {
586
- title: "Add authentication system",
587
- description: "OAuth + session management + permissions",
588
- },
589
- subtasks: [
590
- {
591
- title: "OAuth integration",
592
- description: "Connect to OAuth provider",
593
- files: ["src/auth/oauth.ts", "src/auth/config.ts"],
594
- dependencies: [],
595
- estimated_complexity: 4,
596
- },
597
- {
598
- title: "Session store",
599
- description: "Redis-backed session management",
600
- files: ["src/auth/sessions.ts"],
601
- dependencies: [0],
602
- estimated_complexity: 3,
603
- },
604
- {
605
- title: "Permissions system",
606
- description: "Role-based access control",
607
- files: ["src/auth/permissions.ts", "src/auth/roles.ts"],
608
- dependencies: [0, 1],
609
- estimated_complexity: 5,
610
- },
611
- ],
612
- };
613
-
614
- const result = await structured_parse_cell_tree.execute(
615
- { response: JSON.stringify(treeObj) },
616
- mockCtx,
617
- );
618
-
619
- const parsed = JSON.parse(result);
620
- expect(parsed.success).toBe(true);
621
- expect(parsed.data.epic.title).toBe("Add authentication system");
622
- expect(parsed.summary.subtask_count).toBe(3);
623
- expect(parsed.summary.total_files).toBe(5);
624
- expect(parsed.summary.complexity_total).toBe(12);
625
- });
626
-
627
- it("lists unique files in summary", async () => {
628
- const treeObj = {
629
- epic: { title: "Test epic" },
630
- subtasks: [
631
- {
632
- title: "T1",
633
- files: ["shared.ts", "a.ts"],
634
- dependencies: [],
635
- estimated_complexity: 2,
636
- },
637
- {
638
- title: "T2",
639
- files: ["shared.ts", "b.ts"],
640
- dependencies: [],
641
- estimated_complexity: 2,
642
- },
643
- ],
644
- };
645
-
646
- const result = await structured_parse_cell_tree.execute(
647
- { response: JSON.stringify(treeObj) },
648
- mockCtx,
649
- );
650
-
651
- const parsed = JSON.parse(result);
652
- expect(parsed.success).toBe(true);
653
- expect(parsed.summary.total_files).toBe(3);
654
- expect(parsed.summary.files).toEqual(["shared.ts", "a.ts", "b.ts"]);
655
- });
656
-
657
- it("calculates complexity total correctly", async () => {
658
- const treeObj = {
659
- epic: { title: "Test epic" },
660
- subtasks: [
661
- {
662
- title: "T1",
663
- files: ["a.ts"],
664
- dependencies: [],
665
- estimated_complexity: 2,
666
- },
667
- {
668
- title: "T2",
669
- files: ["b.ts"],
670
- dependencies: [],
671
- estimated_complexity: 3,
672
- },
673
- {
674
- title: "T3",
675
- files: ["c.ts"],
676
- dependencies: [],
677
- estimated_complexity: 1,
678
- },
679
- ],
680
- };
681
-
682
- const result = await structured_parse_cell_tree.execute(
683
- { response: JSON.stringify(treeObj) },
684
- mockCtx,
685
- );
686
-
687
- const parsed = JSON.parse(result);
688
- expect(parsed.success).toBe(true);
689
- expect(parsed.summary.complexity_total).toBe(6);
690
- });
691
-
692
- it("returns error for invalid cell tree", async () => {
693
- const result = await structured_parse_cell_tree.execute(
694
- { response: '{"epic": {}}' }, // Missing required fields
695
- mockCtx,
696
- );
697
-
698
- const parsed = JSON.parse(result);
699
- expect(parsed.success).toBe(false);
700
- expect(parsed.error).toContain("does not match schema");
701
- expect(parsed.expected_shape).toBeDefined();
702
- });
703
-
704
- it("includes expected shape in error feedback", async () => {
705
- const result = await structured_parse_cell_tree.execute(
706
- { response: '{"wrong": "structure"}' },
707
- mockCtx,
708
- );
709
-
710
- const parsed = JSON.parse(result);
711
- expect(parsed.success).toBe(false);
712
- expect(parsed.expected_shape.epic).toBeDefined();
713
- expect(parsed.expected_shape.subtasks).toBeDefined();
714
- });
715
- });
716
-
717
- // ============================================================================
718
- // Edge Cases and Real-World Scenarios
719
- // ============================================================================
720
-
721
- describe("Real-world integration scenarios", () => {
722
- it("handles evaluation with unicode and special characters", async () => {
723
- const evalObj = {
724
- passed: true,
725
- criteria: {
726
- emoji_support: {
727
- passed: true,
728
- feedback: "Handles 🎉 and 你好 correctly",
729
- },
730
- },
731
- overall_feedback: "All special chars work ✅",
732
- retry_suggestion: null,
733
- };
734
-
735
- const result = await structured_parse_evaluation.execute(
736
- { response: JSON.stringify(evalObj) },
737
- mockCtx,
738
- );
739
-
740
- const parsed = JSON.parse(result);
741
- expect(parsed.success).toBe(true);
742
- expect(parsed.data.criteria.emoji_support.feedback).toContain("🎉");
743
- });
744
-
745
- it("handles decomposition with deeply nested file paths", async () => {
746
- const decompObj = {
747
- task: "Reorganize project structure",
748
- subtasks: [
749
- {
750
- title: "Move auth files",
751
- description: "Restructure auth module",
752
- files: [
753
- "src/features/auth/services/oauth/providers/google.ts",
754
- "src/features/auth/services/oauth/providers/github.ts",
755
- ],
756
- estimated_effort: "small",
757
- },
758
- ],
759
- };
760
-
761
- const result = await structured_parse_decomposition.execute(
762
- { response: JSON.stringify(decompObj) },
763
- mockCtx,
764
- );
765
-
766
- const parsed = JSON.parse(result);
767
- expect(parsed.success).toBe(true);
768
- expect(parsed.summary.total_files).toBe(2);
769
- });
770
-
771
- it("handles cell tree with maximum complexity values", async () => {
772
- const treeObj = {
773
- epic: { title: "Complex refactor" },
774
- subtasks: [
775
- {
776
- title: "High complexity task",
777
- files: ["complex.ts"],
778
- dependencies: [],
779
- estimated_complexity: 5,
780
- },
781
- ],
782
- };
783
-
784
- const result = await structured_parse_cell_tree.execute(
785
- { response: JSON.stringify(treeObj) },
786
- mockCtx,
787
- );
788
-
789
- const parsed = JSON.parse(result);
790
- expect(parsed.success).toBe(true);
791
- expect(parsed.summary.complexity_total).toBe(5);
792
- });
793
-
794
- it("handles JSON with escaped characters in paths", async () => {
795
- const decompObj = {
796
- task: "Windows compatibility",
797
- subtasks: [
798
- {
799
- title: "Fix paths",
800
- description: "Handle backslashes",
801
- files: ["C:\\Users\\file.ts", "path\\to\\file.ts"],
802
- estimated_effort: "small",
803
- },
804
- ],
805
- };
806
-
807
- const result = await structured_parse_decomposition.execute(
808
- { response: JSON.stringify(decompObj) },
809
- mockCtx,
810
- );
811
-
812
- const parsed = JSON.parse(result);
813
- expect(parsed.success).toBe(true);
814
- // JSON.stringify escapes backslashes, so they become double-escaped
815
- expect(parsed.data.subtasks[0].files).toContain("C:\\Users\\file.ts");
816
- });
817
- });