@tea-agent/loop-agent 0.38.2 → 0.38.4

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 (192) hide show
  1. package/AGENTS.md +4 -1
  2. package/CHANGELOG.md +171 -96
  3. package/README.md +11 -5
  4. package/bin/loop-agent.js +7 -3
  5. package/dist/application/task-lifecycle/advance.js +4 -0
  6. package/dist/build-stamp.json +3 -3
  7. package/dist/cli/command-definitions.js +16 -5
  8. package/dist/cli/program.js +16 -2
  9. package/dist/cli/update/notifier.js +2 -2
  10. package/dist/cli.js +2 -1
  11. package/dist/commands/dag-rerun.js +55 -1
  12. package/dist/commands/init-model-catalog.js +464 -0
  13. package/dist/commands/init-upgrade.js +265 -97
  14. package/dist/commands/init.js +428 -27
  15. package/dist/commands/inspect-next.js +8 -0
  16. package/dist/commands/task-advance.js +31 -0
  17. package/dist/executors/dag-pi-executor.js +76 -10
  18. package/dist/executors/pi-event-serializer.js +42 -1
  19. package/dist/executors/pi-executor.js +51 -0
  20. package/dist/executors/pi-extension-resolver.js +233 -0
  21. package/dist/executors/pi-playwright-cli-tool.js +51 -12
  22. package/dist/executors/pi-sdk-executor.js +187 -59
  23. package/dist/executors/shell-executor.js +265 -15
  24. package/dist/executors/shell-write-guard.js +7 -0
  25. package/dist/shared/dag-prompt-override.js +28 -0
  26. package/dist/shared/operator/capabilities.js +32 -0
  27. package/dist/shared/playwright-cli-command-policy.js +15 -0
  28. package/dist/shared/update/console-notifier.js +100 -0
  29. package/dist/{cli → shared}/update/npm-client.js +41 -8
  30. package/dist/{cli → shared}/update/state.js +41 -13
  31. package/dist/task/contract/constants.js +1 -0
  32. package/dist/task/contract/diff.js +26 -0
  33. package/dist/task/contract/project.js +5 -0
  34. package/dist/task/contract/schema.js +6 -1
  35. package/dist/task/source-prepare/build-draft.js +15 -0
  36. package/dist/task/source-prepare/parse-intent.js +24 -4
  37. package/dist/worker/cli.js +31 -1
  38. package/dist/worker/console/chat/routes.js +1 -0
  39. package/dist/worker/console/dag-execution-receipt.js +33 -0
  40. package/dist/worker/console/operator-actions.js +116 -0
  41. package/dist/worker/console/operator-user-error.js +2 -2
  42. package/dist/worker/console/pi-readiness.js +4 -4
  43. package/dist/worker/console/routes.js +43 -2
  44. package/dist/worker/console/security.js +63 -9
  45. package/dist/worker/console/server.js +36 -5
  46. package/dist/worker/console/static/assets/{abnfDiagram-N423BO3Z-C4bmo3iY.js → abnfDiagram-N423BO3Z-gauhD9kD.js} +1 -1
  47. package/dist/worker/console/static/assets/{arc-SAD7McOS.js → arc-pm3c2Rsl.js} +1 -1
  48. package/dist/worker/console/static/assets/{architectureDiagram-T3A2C74G-DAVLUdK1.js → architectureDiagram-T3A2C74G-D_JAg3wH.js} +1 -1
  49. package/dist/worker/console/static/assets/{blockDiagram-VBNYF7ZC-CPPdv4vj.js → blockDiagram-VBNYF7ZC-Dk7-ex52.js} +1 -1
  50. package/dist/worker/console/static/assets/{c4Diagram-5PPSVZJV-D1GRGZax.js → c4Diagram-5PPSVZJV-Uval9oRI.js} +1 -1
  51. package/dist/worker/console/static/assets/channel-CVrnUdZl.js +1 -0
  52. package/dist/worker/console/static/assets/{chunk-2GRJ4B5K-D_Bq2ZqK.js → chunk-2GRJ4B5K-BSYjHs-v.js} +1 -1
  53. package/dist/worker/console/static/assets/{chunk-2Q5K7J3B-CPLVtamK.js → chunk-2Q5K7J3B-KGUuBXI8.js} +1 -1
  54. package/dist/worker/console/static/assets/{chunk-5RXB4S5H-DnaZWcL3.js → chunk-5RXB4S5H-JqAtxD9N.js} +1 -1
  55. package/dist/worker/console/static/assets/{chunk-5VM5RSS4-BRTmFgnq.js → chunk-5VM5RSS4-D6EJOVQK.js} +1 -1
  56. package/dist/worker/console/static/assets/{chunk-6Q2QTUOP-BRlfYZoH.js → chunk-6Q2QTUOP-B1JriSjm.js} +1 -1
  57. package/dist/worker/console/static/assets/{chunk-GF5L2VYU-CjNybD4u.js → chunk-GF5L2VYU-D62b96Jd.js} +1 -1
  58. package/dist/worker/console/static/assets/{chunk-JWPE2WC7-D3e-74fG.js → chunk-JWPE2WC7-BsNKEC_r.js} +1 -1
  59. package/dist/worker/console/static/assets/{chunk-KBJHAD2P-Br3ir6f2.js → chunk-KBJHAD2P-B56v6-ZP.js} +1 -1
  60. package/dist/worker/console/static/assets/{chunk-RYQCIY6F-1NgVuRbh.js → chunk-RYQCIY6F-D2KPnd8K.js} +1 -1
  61. package/dist/worker/console/static/assets/{chunk-XXDRQBXY-BTYuvuWN.js → chunk-XXDRQBXY-kzKSCaaL.js} +1 -1
  62. package/dist/worker/console/static/assets/classDiagram-JCYQIIEL-aX91WTya.js +1 -0
  63. package/dist/worker/console/static/assets/classDiagram-v2-OCEON4UE-aX91WTya.js +1 -0
  64. package/dist/worker/console/static/assets/{cose-bilkent-JH36ORCC-BC6Bgosf.js → cose-bilkent-JH36ORCC-C6N6KNJd.js} +1 -1
  65. package/dist/worker/console/static/assets/{cynefin-VYW2F7L2-DV-HypfL.js → cynefin-VYW2F7L2-DiyymT2D.js} +1 -1
  66. package/dist/worker/console/static/assets/{cynefinDiagram-MW4NZA55-DhTaddO-.js → cynefinDiagram-MW4NZA55-BeKAXvf3.js} +1 -1
  67. package/dist/worker/console/static/assets/{dagre-VZM6K2ZE-BnEcOmiI.js → dagre-VZM6K2ZE-TkIdXfPF.js} +1 -1
  68. package/dist/worker/console/static/assets/{diagram-7IWD3JNH-xphGRVKp.js → diagram-7IWD3JNH-BFmSG3ks.js} +1 -1
  69. package/dist/worker/console/static/assets/{diagram-B4RE2ZJO-CrdD1mRG.js → diagram-B4RE2ZJO-DqpJT2Fs.js} +1 -1
  70. package/dist/worker/console/static/assets/{diagram-LBJQPF4R-D4YdLsaF.js → diagram-LBJQPF4R-r5ATN43t.js} +1 -1
  71. package/dist/worker/console/static/assets/{diagram-Q27KOJAE-LCKyGZ5U.js → diagram-Q27KOJAE-DEcSCMuR.js} +1 -1
  72. package/dist/worker/console/static/assets/{diagram-UB23O5K3-BeiTe7t0.js → diagram-UB23O5K3-VOvYRLVc.js} +1 -1
  73. package/dist/worker/console/static/assets/{ebnfDiagram-BXEA7PRR-BO2cq8GS.js → ebnfDiagram-BXEA7PRR-D7f4Cb1u.js} +1 -1
  74. package/dist/worker/console/static/assets/{erDiagram-JOGREHBK-CmWww3Nt.js → erDiagram-JOGREHBK-ziUf3U0G.js} +1 -1
  75. package/dist/worker/console/static/assets/{flowDiagram-UKHOOZJN-C2kkUYg5.js → flowDiagram-UKHOOZJN-BoRdQ-1y.js} +1 -1
  76. package/dist/worker/console/static/assets/{ganttDiagram-PKOTCBZU-DY_JZ2iQ.js → ganttDiagram-PKOTCBZU-B1-GncNm.js} +1 -1
  77. package/dist/worker/console/static/assets/{gitGraphDiagram-DS77QQ5N-BizhCSw9.js → gitGraphDiagram-DS77QQ5N-CTW-ph8W.js} +1 -1
  78. package/dist/worker/console/static/assets/index-C8rOIAJr.css +1 -0
  79. package/dist/worker/console/static/assets/{index-BXN89VPX.js → index-DEbyrEfW.js} +147 -68
  80. package/dist/worker/console/static/assets/{infoDiagram-6WML65LV-Dlo2eK_d.js → infoDiagram-6WML65LV-BFA337tZ.js} +1 -1
  81. package/dist/worker/console/static/assets/{ishikawaDiagram-WSZJBQD7-3EIlIIuZ.js → ishikawaDiagram-WSZJBQD7-BtEujgRR.js} +1 -1
  82. package/dist/worker/console/static/assets/{journeyDiagram-NVQOT4AX-DiMDe5Yh.js → journeyDiagram-NVQOT4AX-C8ZwD2Jp.js} +1 -1
  83. package/dist/worker/console/static/assets/{kanban-definition-27J2QSJJ-sN553oIN.js → kanban-definition-27J2QSJJ-Ck-z8B11.js} +1 -1
  84. package/dist/worker/console/static/assets/{linear-D9Py31Ld.js → linear-CAyXjQjN.js} +1 -1
  85. package/dist/worker/console/static/assets/{mermaid.core-D7Occzk_.js → mermaid.core-CcWKn6tS.js} +5 -5
  86. package/dist/worker/console/static/assets/{mindmap-definition-FAOFIHXS-DjDRboiR.js → mindmap-definition-FAOFIHXS-C2NuDO7E.js} +1 -1
  87. package/dist/worker/console/static/assets/{pegDiagram-VL7TDLO6-Cgc4qlzp.js → pegDiagram-VL7TDLO6-BFub3PDz.js} +1 -1
  88. package/dist/worker/console/static/assets/{pieDiagram-7S7Q4E2Y-DxZj8pKd.js → pieDiagram-7S7Q4E2Y-BJx5THZ5.js} +1 -1
  89. package/dist/worker/console/static/assets/{quadrantDiagram-CIZ2JOQS-CJ50MngR.js → quadrantDiagram-CIZ2JOQS-DhalkDWV.js} +1 -1
  90. package/dist/worker/console/static/assets/{railroadDiagram-AXF67PYL-BY-OD8Tr.js → railroadDiagram-AXF67PYL-CVcBM36o.js} +1 -1
  91. package/dist/worker/console/static/assets/{requirementDiagram-LRYGKXZP-CQFButfK.js → requirementDiagram-LRYGKXZP-D_vy4n0w.js} +1 -1
  92. package/dist/worker/console/static/assets/{sankeyDiagram-W5VNT64P-BB_QUK_b.js → sankeyDiagram-W5VNT64P-DGYdJL1D.js} +1 -1
  93. package/dist/worker/console/static/assets/{sequenceDiagram-SI44F4Z6-CqhXZjyx.js → sequenceDiagram-SI44F4Z6-C1xJFg3k.js} +1 -1
  94. package/dist/worker/console/static/assets/{sizeCapture-X5ZJPWSS-2l1U4rgJ.js → sizeCapture-X5ZJPWSS-D_Fc43df.js} +1 -1
  95. package/dist/worker/console/static/assets/{stateDiagram-OKZ733FA-BX1FbrhJ.js → stateDiagram-OKZ733FA-BhGbIjbu.js} +1 -1
  96. package/dist/worker/console/static/assets/stateDiagram-v2-UEYNNEHI-DUt1lyFL.js +1 -0
  97. package/dist/worker/console/static/assets/{swimlanes-SLNWSIFB-DOSnT7hb.js → swimlanes-SLNWSIFB-DVkNcO-V.js} +2 -2
  98. package/dist/worker/console/static/assets/swimlanesDiagram-ULZ7WXOC-BM6SGr7j.js +8 -0
  99. package/dist/worker/console/static/assets/{timeline-definition-Z64GVDOM-DVJQA7AN.js → timeline-definition-Z64GVDOM-BQU9ufV0.js} +1 -1
  100. package/dist/worker/console/static/assets/{vennDiagram-T6HMQDX7-74uxl9gx.js → vennDiagram-T6HMQDX7-BCQhRmYB.js} +1 -1
  101. package/dist/worker/console/static/assets/{wardleyDiagram-T6FBY63Y-DsBxlcg_.js → wardleyDiagram-T6FBY63Y-DWerOyHO.js} +1 -1
  102. package/dist/worker/console/static/assets/{xychartDiagram-ELKLHX3M-snuW9Mjw.js → xychartDiagram-ELKLHX3M-DxLeebQU.js} +1 -1
  103. package/dist/worker/console/static/index.html +2 -2
  104. package/dist/worker/console/static-src/operator-chat/mutation-gate.js +1 -1
  105. package/dist/worker/console/static-src/operator-chat/open-preview-in-browser.js +328 -0
  106. package/dist/worker/loop-agent/loop-agent-client.js +7 -3
  107. package/dist/worker/observe/node-input.js +74 -3
  108. package/dist/worker/observe/node-process.js +377 -0
  109. package/dist/worker/observe/routes.js +36 -3
  110. package/dist/worker/observe/shell-handler-keys.js +34 -0
  111. package/dist/worker/observe/static/api.js +66 -0
  112. package/dist/worker/observe/static/constants.js +8 -1
  113. package/dist/worker/observe/static/dag-history-labels.js +4 -0
  114. package/dist/worker/observe/static/format.js +23 -0
  115. package/dist/worker/observe/static/markdown-render.js +20 -1
  116. package/dist/worker/observe/static/operator-chrome.js +4 -0
  117. package/dist/worker/observe/static/state.js +13 -2
  118. package/dist/worker/observe/static/styles.css +73 -0
  119. package/dist/worker/observe/static/views/dag-inspector.js +612 -8
  120. package/dist/worker/observe/static/views/dag.js +8 -1
  121. package/dist/worker/observe/static/views/failures.js +5 -2
  122. package/dist/worker/observe/static/views/session-timeline.js +78 -9
  123. package/dist/workflows/dag/backend-test-case-coverage-analysis.js +76 -6
  124. package/dist/workflows/dag/backend-test-scenario-partitions.js +73 -1
  125. package/dist/workflows/dag/contract-output-registry.js +15 -0
  126. package/dist/workflows/dag/failure-category.js +12 -0
  127. package/dist/workflows/dag/frontend-implementation-contract.js +145 -40
  128. package/dist/workflows/dag/frontend-prewrite-gate.js +87 -135
  129. package/dist/workflows/dag/frontend-repair.js +14 -0
  130. package/dist/workflows/dag/frontend-risk.js +15 -2
  131. package/dist/workflows/dag/frontend-test-case-checklist.js +26 -2
  132. package/dist/workflows/dag/frontend-test-case-manifest.js +5 -0
  133. package/dist/workflows/dag/frontend-test-case-quality.js +5 -13
  134. package/dist/workflows/dag/frontend-test-environment-probe.js +227 -0
  135. package/dist/workflows/dag/frontend-test-markdown.js +61 -0
  136. package/dist/workflows/dag/frontend-test-result-contract.js +160 -28
  137. package/dist/workflows/dag/frontend-test-standard-scenarios.js +68 -0
  138. package/dist/workflows/dag/frontend-verification-trace.js +29 -4
  139. package/dist/workflows/dag/init-hybrid.js +196 -117
  140. package/dist/workflows/dag/node-execution.js +139 -10
  141. package/dist/workflows/dag/prompt.js +15 -1
  142. package/dist/workflows/dag/rerun-plan.js +134 -3
  143. package/dist/workflows/dag/rerun-run.js +58 -6
  144. package/dist/workflows/dag/retry-policy.js +91 -0
  145. package/dist/workflows/dag/structured-output-repair.js +712 -0
  146. package/dist/workflows/dag/types.js +32 -1
  147. package/dist/workflows/dag/validate.js +41 -1
  148. package/docs/init-surface.manifest.json +5 -0
  149. package/docs/operations/README.md +1 -1
  150. package/docs/operations/local-development-environment.md +1 -5
  151. package/docs/skills/vetted-skill-registry.md +14 -0
  152. package/docs/templates/README.md +1 -1
  153. package/docs/templates/agent-dag.schema.json +44 -0
  154. package/docs/templates/backend-test-dag.json +2 -2
  155. package/docs/templates/frontend-test-dag.json +8 -10
  156. package/docs/templates/frontend-test-dag.retrieve-context.prompt.md +1 -1
  157. package/docs/templates/init-managed-agents.md +13 -3
  158. package/harness.json +2 -2
  159. package/package.json +12 -4
  160. package/scripts/next-info.mjs +356 -0
  161. package/scripts/next-publish-gate.mjs +238 -0
  162. package/scripts/release-source-binding.mjs +251 -0
  163. package/skills/analyze-product-requirements/SKILL.md +8 -5
  164. package/skills/analyze-product-requirements/references/acceptance-criteria.md +1 -1
  165. package/skills/analyze-product-requirements/references/clarification-and-knowledge.md +10 -6
  166. package/skills/analyze-product-requirements/references/example.md +50 -0
  167. package/skills/analyze-product-requirements/references/forward-test-cases.md +11 -7
  168. package/skills/analyze-product-requirements/references/kb-integration.md +5 -5
  169. package/skills/analyze-product-requirements/references/product-analysis-schema.md +8 -4
  170. package/skills/analyze-product-requirements/references/product-requirement-schema.md +2 -2
  171. package/skills/analyze-product-requirements/references/requirement-clarification-schema.md +1 -1
  172. package/skills/analyze-product-requirements/scripts/test-validators.mjs +11 -1
  173. package/skills/analyze-product-requirements/scripts/validate-product-analysis.mjs +20 -1
  174. package/skills/analyze-product-requirements/scripts/validate-requirement-clarification.mjs +9 -0
  175. package/skills/codebase-scout/SKILL.md +1 -1
  176. package/skills/fe-test-ui-scout/SKILL.md +65 -0
  177. package/skills/fe-test-ui-scout/references/ledger-schema.md +62 -0
  178. package/skills/fe-test-ui-scout/references/recon-protocol.md +54 -0
  179. package/skills/improve-codebase-architecture/SKILL.md +81 -0
  180. package/skills/improve-codebase-architecture/deepening.md +37 -0
  181. package/skills/improve-codebase-architecture/html-report.md +123 -0
  182. package/skills/improve-codebase-architecture/interface-design.md +44 -0
  183. package/skills/improve-codebase-architecture/language.md +53 -0
  184. package/skills/loop-agent/references/command-reference.md +11 -4
  185. package/skills/playwright-cli/SKILL.md +22 -0
  186. package/skills/playwright-cli-case-generator/SKILL.md +20 -1
  187. package/dist/worker/console/static/assets/channel-R6oQIDkw.js +0 -1
  188. package/dist/worker/console/static/assets/classDiagram-JCYQIIEL-BPXlmPWI.js +0 -1
  189. package/dist/worker/console/static/assets/classDiagram-v2-OCEON4UE-BPXlmPWI.js +0 -1
  190. package/dist/worker/console/static/assets/index-CuH5CNFu.css +0 -1
  191. package/dist/worker/console/static/assets/stateDiagram-v2-UEYNNEHI-DJKi0Qel.js +0 -1
  192. package/dist/worker/console/static/assets/swimlanesDiagram-ULZ7WXOC-YYe7EEm5.js +0 -8
@@ -0,0 +1,8 @@
1
+ import { collectNextInfo, formatNextInfo, } from "../../scripts/next-info.mjs";
2
+ export async function runInspectNext(repoRoot, args = []) {
3
+ const json = args.includes("--json");
4
+ const result = collectNextInfo({ cwd: repoRoot });
5
+ process.stdout.write(formatNextInfo(result, { json }));
6
+ if (!result.ok)
7
+ process.exitCode = result.exitCode || 1;
8
+ }
@@ -1,4 +1,5 @@
1
1
  import { advanceTaskLifecycle, } from "../application/task-lifecycle/index.js";
2
+ import { backendTestConfigSchema, } from "../task/config-types.js";
2
3
  import { buildOperatorResult, operatorFailed, processExitCodeForOutcome, writeOperatorJson, } from "../shared/operator/index.js";
3
4
  import { createRunDagProgressObserver, validateRunDagProgressIntervalMs, } from "./run-dag-progress.js";
4
5
  const COMMAND = "task advance";
@@ -14,6 +15,7 @@ const USAGE = `usage:
14
15
  [--forbidden-path <glob>]...
15
16
  [--verify <label:command>]...
16
17
  [--verify-timeout <label:timeoutMs>]...
18
+ [--backend-test <json>]
17
19
  [--accept-recommendations <digest>]
18
20
  [--approve-gate <gate-id:digest>]
19
21
  [--reject-gate <gate-id:digest> --reason <text>]
@@ -52,6 +54,30 @@ function parseVerifyTimeout(flag) {
52
54
  }
53
55
  return { label: flag.slice(0, idx), timeoutMs };
54
56
  }
57
+ /**
58
+ * Parse `--backend-test <json>` into a validated managed backendTest config.
59
+ * `{}` is legal (explicit clear back to the default layout).
60
+ */
61
+ function parseBackendTestFlag(raw) {
62
+ let parsed;
63
+ try {
64
+ parsed = JSON.parse(raw);
65
+ }
66
+ catch {
67
+ throw new Error(`invalid --backend-test ${raw}; expected an inline JSON object`);
68
+ }
69
+ if (typeof parsed !== "object" || parsed === null || Array.isArray(parsed)) {
70
+ throw new Error(`invalid --backend-test ${raw}; expected an inline JSON object`);
71
+ }
72
+ const result = backendTestConfigSchema.safeParse(parsed);
73
+ if (!result.success) {
74
+ const summary = result.error.issues
75
+ .map((issue) => `${issue.path.join(".") || "(root)"}: ${issue.message}`)
76
+ .join("; ");
77
+ throw new Error(`invalid --backend-test: ${summary}`);
78
+ }
79
+ return result.data ?? {};
80
+ }
55
81
  export function parseTaskAdvanceArgs(args) {
56
82
  const options = {};
57
83
  const positional = [];
@@ -113,6 +139,10 @@ export function parseTaskAdvanceArgs(args) {
113
139
  options.verifyTimeout = pushList(options.verifyTimeout, next());
114
140
  continue;
115
141
  }
142
+ if (token === "--backend-test") {
143
+ options.backendTest = parseBackendTestFlag(next());
144
+ continue;
145
+ }
116
146
  if (token === "--accept-recommendations") {
117
147
  options.acceptRecommendations = next();
118
148
  continue;
@@ -273,6 +303,7 @@ export function toUseCaseInput(repoRoot, taskId, options, observer) {
273
303
  allowedPaths: options.allowedPaths,
274
304
  forbiddenPaths: options.forbiddenPaths,
275
305
  verifyCommands: verifyCommands.length > 0 ? verifyCommands : undefined,
306
+ backendTest: options.backendTest,
276
307
  approveGateToken: options.approveGate,
277
308
  rejectGateToken: options.rejectGate,
278
309
  rejectReason: options.reason,
@@ -2,13 +2,14 @@ import path from "node:path";
2
2
  import { createHash } from "node:crypto";
3
3
  import { readFile } from "node:fs/promises";
4
4
  import { writeDagNodeJsonArtifact, writeTextArtifactFile, } from "../infrastructure/harness/artifact-store.js";
5
- import { executePiStep, } from "./pi-executor.js";
5
+ import { executePiStep, resolvePiBackend, } from "./pi-executor.js";
6
+ import { resolveDagPiExtensions, } from "./pi-extension-resolver.js";
6
7
  import { buildPiWriterToolPolicyContext, createPiWriterCustomTools, } from "./pi-writer-tool-policy.js";
7
8
  import { cleanupPlaywrightCliDefaultSession, createPlaywrightCliTool, PI_COMMAND_CAPABILITY_REGISTRY, resolveCaseIdFromWriteSet, resolveEvidenceDirFromWriteSet, } from "./pi-playwright-cli-tool.js";
8
9
  import { dagCommandPolicyAllows, resolveDagCommandPolicy, } from "../workflows/dag/types.js";
9
10
  import { redactPromptForLog, truncateOutput, } from "../shared/output-truncation.js";
10
11
  import { GitStatusUnavailableError, pathsChangedDuringRun, readGitStatusPorcelain, recoverRootNulArtifact, snapshotGitStatusPathFingerprints, snapshotGitStatusPorcelain, validateShellWriteGuard, } from "./shell-write-guard.js";
11
- import { isWriterEmptyDiffRetryCandidate, isWriterTransportRetryCandidate, INCOMPLETE_WRITE_SET_RETRY_CATEGORY, WRITER_CLEAN_TIMEOUT_RETRY_CATEGORY, WRITER_EMPTY_DIFF_RETRY_CATEGORY, } from "../workflows/dag/retry-policy.js";
12
+ import { isTargetTemplateTransientRetryNode, isWriterEmptyDiffRetryCandidate, isWriterTransportRetryCandidate, INCOMPLETE_WRITE_SET_RETRY_CATEGORY, WRITER_CLEAN_TIMEOUT_RETRY_CATEGORY, WRITER_EMPTY_DIFF_RETRY_CATEGORY, } from "../workflows/dag/retry-policy.js";
12
13
  import { assessBackendTestMdPlanCompleteness, assessBackendTestMdWriterCompleteness, assessBackendTestPytestPlanCompleteness, assessBackendTestPytestWriterCompleteness, assessBackendTestShardChildCompleteness, backendTestWriterProgressRoleForTask, classifyBackendTestWriterCompletenessFailure, isBackendTestCompletenessRetryCandidate, isBackendTestMdPlanTask, isBackendTestPytestPlanTask, isBackendTestShardChildTask, writeBackendTestWriterProgressArtifacts, } from "../workflows/dag/backend-test-writer-completeness.js";
13
14
  import { resolveBackendTestLayout } from "../workflows/dag/backend-test-layout.js";
14
15
  import { redactSecrets, truncateUtf8Preview } from "../shared/preview.js";
@@ -327,6 +328,36 @@ export async function writePiExecutorArtifacts(artifactsDir, input) {
327
328
  await writeTextArtifactFile(summaryPath, buildPiResultSummaryMarkdown(input));
328
329
  return { promptPath, summaryPath };
329
330
  }
331
+ /**
332
+ * Run-owned per-node Pi extension facts (plan 2026-08-21): requested ids,
333
+ * resolved packages (with absolute entry paths), and degraded/missing ids.
334
+ * Written only for nodes that declare `piExtensions`; never into the repo-level
335
+ * DAG template. Best-effort: artifact failures never fail the node.
336
+ */
337
+ async function writePiExtensionsArtifact(nodeArtifactsDir, requested, resolved, missing) {
338
+ try {
339
+ const runDir = path.dirname(nodeArtifactsDir);
340
+ const nodeId = path.basename(nodeArtifactsDir);
341
+ await writeDagNodeJsonArtifact(runDir, nodeId, "pi-extensions.json", {
342
+ schemaVersion: 1,
343
+ requested: [...requested],
344
+ resolved: resolved.map((entry) => ({
345
+ id: entry.id,
346
+ packageName: entry.packageName,
347
+ version: entry.version,
348
+ path: entry.entryPaths.join(", "),
349
+ })),
350
+ missing: missing.map((entry) => ({
351
+ id: entry.id,
352
+ reason: entry.reason,
353
+ ...(entry.detail ? { detail: entry.detail } : {}),
354
+ })),
355
+ });
356
+ }
357
+ catch {
358
+ // best-effort run fact; node execution continues
359
+ }
360
+ }
330
361
  async function resolveValidatedFrontendBaseUrlFromContext(workspaceRoot, runDir) {
331
362
  try {
332
363
  const capability = JSON.parse(await readFile(path.join(runDir, "preflight-frontend-browser-tool-shell", "frontend-browser-capability.json"), "utf8"));
@@ -550,6 +581,29 @@ export async function executeDagPiNode(input, meta, piStepFn = executePiStep, wr
550
581
  };
551
582
  }
552
583
  }
584
+ // Plan 2026-08-21 D4/D6: resolve the node's explicit Pi extension
585
+ // allowlist against the settings inventory. Degrade on missing packages —
586
+ // never fail the node — and persist requested/resolved/missing facts.
587
+ const piExtensionsResolution = input.task.piExtensions
588
+ ? resolveDagPiExtensions({
589
+ repoRoot: input.cwd,
590
+ ids: input.task.piExtensions,
591
+ })
592
+ : undefined;
593
+ if (piExtensionsResolution) {
594
+ const piBackend = resolvePiBackend();
595
+ const cliDegraded = piBackend === "cli-only";
596
+ await writePiExtensionsArtifact(path.join(meta.runDir, input.task.id), input.task.piExtensions, cliDegraded ? [] : piExtensionsResolution.resolved, cliDegraded
597
+ ? input.task.piExtensions.map((id) => ({
598
+ id,
599
+ reason: "not-installed",
600
+ detail: "cli backend does not load pi extensions (sdk-only v1)",
601
+ }))
602
+ : piExtensionsResolution.missing);
603
+ }
604
+ const piExtensionPaths = piExtensionsResolution && resolvePiBackend() !== "cli-only"
605
+ ? piExtensionsResolution.resolved.flatMap((entry) => entry.entryPaths)
606
+ : undefined;
553
607
  result = await piStepFn({
554
608
  attachedFiles: [],
555
609
  modelConfig: resolveDagPiModelConfig(input.model, input.thinking ? { thinking: input.thinking } : undefined),
@@ -564,6 +618,9 @@ export async function executeDagPiNode(input, meta, piStepFn = executePiStep, wr
564
618
  stallTimeoutMs: input.stallTimeoutMs,
565
619
  abortGraceMs: input.abortGraceMs,
566
620
  ...(writerToolPolicy ? { writerToolPolicy } : {}),
621
+ ...(piExtensionPaths && piExtensionPaths.length > 0
622
+ ? { piExtensionPaths }
623
+ : {}),
567
624
  onActivity: bridgeActivity
568
625
  ? (activity) => {
569
626
  if (activity.kind === "lease" ||
@@ -885,22 +942,31 @@ export async function executeDagPiNode(input, meta, piStepFn = executePiStep, wr
885
942
  if (meta.writeGuardAttribution === "best-effort") {
886
943
  stderrParts.push("write guard note: concurrent rank writers use best-effort per-node attribution; keep same-rank writeSet entries disjoint");
887
944
  }
888
- // Pure provider stall/timeout with zero workspace side effects is eligible
889
- // for a single writer transport retry (implement-pi flake resilience).
890
- // Partial writes, tool calls, or write-guard failures keep raw `timeout`.
891
- const writerCleanTimeout = !mapped.ok &&
892
- mapped.failureCategory === "timeout" &&
893
- writeGuardOk &&
945
+ const rawFailureCategory = mapped.failureCategory;
946
+ const writeToolCallCount = readWriterThinkingExhaustionEvidence(result).writeToolCallCount;
947
+ const zeroSideEffects = writeGuardOk &&
894
948
  !completenessFailure &&
895
949
  !writerThinkingExhausted &&
896
- isWriterTransportRetryCandidate(input.task) &&
950
+ !writerOutcomeViolation &&
897
951
  changeManifestChangedFiles !== undefined &&
898
952
  changeManifestChangedFiles.length === 0 &&
899
- readWriterThinkingExhaustionEvidence(result).writeToolCallCount === 0;
953
+ writeToolCallCount === 0;
954
+ const targetCleanTransport = !mapped.ok &&
955
+ isTargetTemplateTransientRetryNode(input.task) &&
956
+ rawFailureCategory !== undefined &&
957
+ ["timeout", "network", "rate-limit", "unavailable"].includes(rawFailureCategory) &&
958
+ zeroSideEffects;
959
+ const frontendCleanTimeout = !mapped.ok &&
960
+ rawFailureCategory === "timeout" &&
961
+ isWriterTransportRetryCandidate(input.task) &&
962
+ !isTargetTemplateTransientRetryNode(input.task) &&
963
+ zeroSideEffects;
964
+ const writerCleanTimeout = targetCleanTransport || frontendCleanTimeout;
900
965
  return {
901
966
  ...mapped,
902
967
  ok: false,
903
968
  stderr: stderrParts.filter(Boolean).join("\n\n"),
969
+ ...(rawFailureCategory ? { rawFailureCategory } : {}),
904
970
  failureCategory: mapped.ok
905
971
  ? writeGuardOk
906
972
  ? completenessFailure
@@ -51,9 +51,50 @@ export function normalizePiSessionEvent(event) {
51
51
  }
52
52
  return unifyToolPayloadFields(event);
53
53
  }
54
+ const TERMINAL_FINISH_REASON_TYPES = new Set([
55
+ 'message_end',
56
+ 'turn_end',
57
+ 'assistant_message',
58
+ ]);
59
+ function readExplicitFinishReason(source) {
60
+ const message = isRecord(source.message) ? source.message : undefined;
61
+ const candidates = [
62
+ source.finish_reason,
63
+ source.finishReason,
64
+ source.stop_reason,
65
+ source.stopReason,
66
+ message?.finish_reason,
67
+ message?.finishReason,
68
+ message?.stop_reason,
69
+ message?.stopReason,
70
+ ];
71
+ for (const candidate of candidates) {
72
+ if (typeof candidate === 'string' && candidate.trim().length > 0) {
73
+ return candidate;
74
+ }
75
+ }
76
+ return undefined;
77
+ }
78
+ function applyFinishReason(original, normalized) {
79
+ const originalType = typeof original.type === 'string' ? original.type : undefined;
80
+ if (!originalType || !TERMINAL_FINISH_REASON_TYPES.has(originalType)) {
81
+ return normalized;
82
+ }
83
+ const existing = readExplicitFinishReason({
84
+ finish_reason: normalized.finish_reason,
85
+ });
86
+ if (existing !== undefined) {
87
+ return normalized;
88
+ }
89
+ const promoted = readExplicitFinishReason(original);
90
+ if (promoted === undefined) {
91
+ return normalized;
92
+ }
93
+ return { ...normalized, finish_reason: promoted };
94
+ }
54
95
  /** Serialize one SDK session event as a single JSONL line. */
55
96
  export function serializeSessionEvent(event, options) {
56
- const normalized = normalizePiSessionEvent(event);
97
+ const normalized = applyFinishReason(event, normalizePiSessionEvent(event));
57
98
  if (typeof normalized.recordedAt !== 'string') {
58
99
  normalized.recordedAt = options?.recordedAt ?? new Date().toISOString();
59
100
  }
@@ -408,6 +408,50 @@ export class BoundedTextPreview {
408
408
  ].join("\n");
409
409
  }
410
410
  }
411
+ /**
412
+ * Normalize provider-specific completion-limit labels to the DAG's canonical
413
+ * `length` signal. Other terminal reasons remain observable without being
414
+ * treated as truncation.
415
+ */
416
+ export function normalizeProviderStopReason(value) {
417
+ if (typeof value !== "string" || value.trim().length === 0)
418
+ return undefined;
419
+ const trimmed = value.trim();
420
+ const compact = trimmed.toLowerCase().replace(/[\s_-]/g, "");
421
+ if ([
422
+ "length",
423
+ "maxtokens",
424
+ "maxoutputtokens",
425
+ "outputlimit",
426
+ "tokenlimit",
427
+ "tokenlimitreached",
428
+ ].includes(compact)) {
429
+ return "length";
430
+ }
431
+ return trimmed;
432
+ }
433
+ /**
434
+ * Extract a terminal stop reason from the provider event shapes used by Pi
435
+ * SDK/CLI bridges. Providers disagree on `stop*` versus `finish*`; keep this
436
+ * boundary centralized so downstream DAG recovery sees one canonical signal.
437
+ */
438
+ export function extractPiEventStopReason(event) {
439
+ const nested = [event.message, event.response, event.completion].filter(isRecord);
440
+ const containers = [...nested, event];
441
+ for (const container of containers) {
442
+ for (const key of [
443
+ "stopReason",
444
+ "stop_reason",
445
+ "finishReason",
446
+ "finish_reason",
447
+ ]) {
448
+ const normalized = normalizeProviderStopReason(container[key]);
449
+ if (normalized)
450
+ return normalized;
451
+ }
452
+ }
453
+ return undefined;
454
+ }
411
455
  export class PiJsonlStreamCollector {
412
456
  lineBuffer = "";
413
457
  discardUntilNewline = false;
@@ -415,6 +459,7 @@ export class PiJsonlStreamCollector {
415
459
  parsedEvents = 0;
416
460
  tokensUsed = 0;
417
461
  outputTooLarge = false;
462
+ stopReason;
418
463
  subagentTotalCalls = 0;
419
464
  subagentFailedCalls = 0;
420
465
  subagentModes = new Set();
@@ -454,6 +499,7 @@ export class PiJsonlStreamCollector {
454
499
  assistantText: this.assistantText.trim(),
455
500
  outputTooLarge: this.outputTooLarge,
456
501
  parsedEvents: this.parsedEvents,
502
+ stopReason: this.stopReason,
457
503
  subagentStats: this.formatSubagentStats(),
458
504
  tokensUsed: this.tokensUsed,
459
505
  };
@@ -472,6 +518,10 @@ export class PiJsonlStreamCollector {
472
518
  }
473
519
  }
474
520
  consumeEvent(event) {
521
+ const stopReason = extractPiEventStopReason(event);
522
+ if (stopReason === "length" || !this.stopReason) {
523
+ this.stopReason = stopReason ?? this.stopReason;
524
+ }
475
525
  const candidate = extractAssistantTextFromEvent(event);
476
526
  if (candidate !== undefined) {
477
527
  if (candidate.length > MAX_ASSISTANT_TEXT_CHARS) {
@@ -865,6 +915,7 @@ async function executeSingleCliAttempt(options, modelConfig, promptFilePath) {
865
915
  stdoutTruncated: stdoutPreview.truncated,
866
916
  stderrTruncated: stderrPreview.truncated,
867
917
  outputTooLarge: collected.outputTooLarge,
918
+ stopReason: collected.stopReason,
868
919
  timedOut,
869
920
  tokensUsed: collected.tokensUsed,
870
921
  subagentStats: collected.subagentStats,
@@ -0,0 +1,233 @@
1
+ import { existsSync, readdirSync, readFileSync, statSync } from "node:fs";
2
+ import os from "node:os";
3
+ import path from "node:path";
4
+ /**
5
+ * Pi extension allowlist for backend implementation DAGs (plan D2/D3).
6
+ * DAG specs carry these frozen short ids; the executor resolves them against
7
+ * the Pi agent settings package inventory at run time.
8
+ */
9
+ export const DAG_PI_EXTENSION_IDS = ["pi-codegraph", "pi-lens"];
10
+ /**
11
+ * Alias table: one DAG short id may resolve to multiple published package
12
+ * names. The first settings spec whose de-`npm:`-prefixed package name is in
13
+ * the alias set wins (project settings override user settings).
14
+ */
15
+ export const DAG_PI_EXTENSION_PACKAGE_ALIASES = {
16
+ "pi-codegraph": [
17
+ "@estebanforge/pi-codegraph-enhanced",
18
+ "@vndv/pi-codegraph",
19
+ "pi-codegraph",
20
+ ],
21
+ "pi-lens": ["pi-lens"],
22
+ };
23
+ /**
24
+ * Extension tools that must never be activated as extras on any node (D5):
25
+ * they mutate files without the writer path policy. Writer nodes keep only
26
+ * policy-wrapped edit/write; read nodes keep only navigation tools.
27
+ */
28
+ export const DAG_PI_EXTENSION_WRITE_TOOL_DENY = new Set([
29
+ "ast_grep_replace",
30
+ ]);
31
+ /** Builtin tool names extensions must never replace (D5). */
32
+ export const DAG_PI_EXTENSION_PROTECTED_BUILTIN_TOOLS = new Set(["read", "edit", "write", "bash", "grep", "find", "ls"]);
33
+ /** Deny surface extensions must never activate (D5). */
34
+ export const DAG_PI_EXTENSION_DENIED_TOOL_IDS = new Set([
35
+ "apply_patch",
36
+ "apply-patch",
37
+ "full-tools",
38
+ "full_tools",
39
+ "coding-chat",
40
+ "coding_chat",
41
+ "shell",
42
+ ]);
43
+ export function isDagPiExtensionId(value) {
44
+ return DAG_PI_EXTENSION_IDS.includes(value);
45
+ }
46
+ function getPiAgentDir() {
47
+ const envDir = process.env.PI_CODING_AGENT_DIR;
48
+ if (envDir)
49
+ return envDir;
50
+ return path.join(os.homedir(), ".pi", "agent");
51
+ }
52
+ function readSettingsPackages(settingsPath) {
53
+ try {
54
+ if (!existsSync(settingsPath))
55
+ return undefined;
56
+ const parsed = JSON.parse(readFileSync(settingsPath, "utf8"));
57
+ if (!Array.isArray(parsed.packages))
58
+ return undefined;
59
+ return parsed.packages.filter((spec) => typeof spec === "string");
60
+ }
61
+ catch {
62
+ return undefined;
63
+ }
64
+ }
65
+ /** Strip an `npm:` prefix from a settings package spec, e.g. `npm:pi-lens`. */
66
+ export function packageNameFromSpec(spec) {
67
+ return spec.startsWith("npm:") ? spec.slice(4) : spec;
68
+ }
69
+ function loadPackageInventories(repoRoot) {
70
+ const agentDir = getPiAgentDir();
71
+ const userSpecs = readSettingsPackages(path.join(agentDir, "settings.json"));
72
+ const projectSpecs = readSettingsPackages(path.join(repoRoot, ".pi", "settings.json"));
73
+ const inventories = [];
74
+ if (userSpecs) {
75
+ inventories.push({
76
+ specs: userSpecs,
77
+ nodeModulesRoot: path.join(agentDir, "npm", "node_modules"),
78
+ });
79
+ }
80
+ if (projectSpecs) {
81
+ inventories.push({
82
+ specs: projectSpecs,
83
+ nodeModulesRoot: path.join(repoRoot, "node_modules"),
84
+ });
85
+ }
86
+ return inventories;
87
+ }
88
+ const EXTENSION_FILE_RE = /\.(?:ts|js|mjs|cjs)$/;
89
+ /**
90
+ * Resolve a package's `pi.extensions` entries to absolute entry paths.
91
+ * - A file entry maps to itself.
92
+ * - A directory entry expands to every extension file directly inside it
93
+ * (e.g. `@estebanforge/pi-codegraph-enhanced` ships `./extensions` holding
94
+ * `codegraph.ts`).
95
+ * No `pi.extensions` declaration, or a declaration whose paths do not exist,
96
+ * yields no entries (caller records `no-entry`) — no heuristic scanning
97
+ * outside declared paths.
98
+ */
99
+ export function resolvePackageEntryPaths(packageDir) {
100
+ let manifest;
101
+ try {
102
+ manifest = JSON.parse(readFileSync(path.join(packageDir, "package.json"), "utf8"));
103
+ }
104
+ catch {
105
+ return { entryPaths: [], reason: "no-entry" };
106
+ }
107
+ const declared = manifest.pi?.extensions;
108
+ if (!Array.isArray(declared)) {
109
+ return { entryPaths: [], reason: "no-entry" };
110
+ }
111
+ const entryPaths = [];
112
+ for (const entry of declared) {
113
+ if (typeof entry !== "string" || entry.length === 0)
114
+ continue;
115
+ const resolved = path.resolve(packageDir, entry);
116
+ if (!existsSync(resolved))
117
+ continue;
118
+ let isDirectory = false;
119
+ try {
120
+ isDirectory = statSync(resolved).isDirectory();
121
+ }
122
+ catch {
123
+ continue;
124
+ }
125
+ if (!isDirectory) {
126
+ entryPaths.push(resolved);
127
+ continue;
128
+ }
129
+ let names;
130
+ try {
131
+ names = readdirSync(resolved);
132
+ }
133
+ catch {
134
+ continue;
135
+ }
136
+ for (const name of names.sort()) {
137
+ if (!EXTENSION_FILE_RE.test(name))
138
+ continue;
139
+ entryPaths.push(path.join(resolved, name));
140
+ }
141
+ }
142
+ if (entryPaths.length === 0) {
143
+ return { entryPaths: [], reason: "no-entry" };
144
+ }
145
+ return { entryPaths };
146
+ }
147
+ /**
148
+ * Resolve frozen DAG short ids against the Pi settings package inventory
149
+ * (user `~/.pi/agent/settings.json`, then project `.pi/settings.json` — a
150
+ * later match on the same alias wins). Pure synchronous read; never throws.
151
+ */
152
+ export function resolveDagPiExtensions(input) {
153
+ const inventories = loadPackageInventories(input.repoRoot);
154
+ const resolved = [];
155
+ const missing = [];
156
+ for (const id of input.ids) {
157
+ if (!isDagPiExtensionId(id)) {
158
+ missing.push({
159
+ id,
160
+ reason: "package-unreadable",
161
+ detail: `unknown pi extension id: ${id}`,
162
+ });
163
+ continue;
164
+ }
165
+ const aliases = DAG_PI_EXTENSION_PACKAGE_ALIASES[id];
166
+ let matched;
167
+ // Later inventories (project) override earlier (user).
168
+ for (const inventory of inventories) {
169
+ for (const spec of inventory.specs) {
170
+ const packageName = packageNameFromSpec(spec);
171
+ if (!aliases.includes(packageName))
172
+ continue;
173
+ const packageDir = path.join(inventory.nodeModulesRoot, ...packageName.split("/"));
174
+ if (!existsSync(packageDir))
175
+ continue;
176
+ try {
177
+ const manifest = JSON.parse(readFileSync(path.join(packageDir, "package.json"), "utf8"));
178
+ if (typeof manifest.version !== "string")
179
+ continue;
180
+ matched = { packageDir, packageName, version: manifest.version };
181
+ }
182
+ catch {
183
+ // unreadable manifest: skip this candidate
184
+ }
185
+ }
186
+ }
187
+ if (!matched) {
188
+ missing.push({ id, reason: "not-installed" });
189
+ continue;
190
+ }
191
+ const entries = resolvePackageEntryPaths(matched.packageDir);
192
+ if (entries.entryPaths.length === 0) {
193
+ missing.push({
194
+ id,
195
+ reason: entries.reason === "no-entry" ? "no-entry" : "package-unreadable",
196
+ });
197
+ continue;
198
+ }
199
+ resolved.push({
200
+ id,
201
+ packageName: matched.packageName,
202
+ version: matched.version,
203
+ entryPaths: entries.entryPaths,
204
+ });
205
+ }
206
+ return { resolved, missing };
207
+ }
208
+ /**
209
+ * Compute extension tool names that may join a node's active tool allowlist
210
+ * after session creation (D5): new tool names registered by the loaded
211
+ * extensions, minus protected builtins, the deny surface, and write-capable
212
+ * extension tools (both buckets — writer writes stay policy-wrapped only).
213
+ */
214
+ export function computeExtensionToolAllowlistExtras(sessionTools, baseline) {
215
+ const baselineLower = new Set(baseline.map((name) => name.toLowerCase()));
216
+ const extras = [];
217
+ for (const tool of sessionTools) {
218
+ const name = tool && typeof tool.name === "string" ? tool.name : undefined;
219
+ if (!name)
220
+ continue;
221
+ const lower = name.toLowerCase();
222
+ if (baselineLower.has(lower))
223
+ continue;
224
+ if (DAG_PI_EXTENSION_PROTECTED_BUILTIN_TOOLS.has(lower))
225
+ continue;
226
+ if (DAG_PI_EXTENSION_DENIED_TOOL_IDS.has(lower))
227
+ continue;
228
+ if (DAG_PI_EXTENSION_WRITE_TOOL_DENY.has(lower))
229
+ continue;
230
+ extras.push(name);
231
+ }
232
+ return extras;
233
+ }