@gajae-code/coding-agent 0.8.2 → 0.9.0

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 (179) hide show
  1. package/CHANGELOG.md +29 -0
  2. package/dist/types/append-only-mode.d.ts +31 -0
  3. package/dist/types/capability/rule.d.ts +8 -0
  4. package/dist/types/cli/fixture-report.d.ts +20 -0
  5. package/dist/types/cli/fixture-report.test.d.ts +1 -0
  6. package/dist/types/cli/nofile-limit.d.ts +11 -0
  7. package/dist/types/cli/update-cli.d.ts +22 -0
  8. package/dist/types/config/settings-schema.d.ts +105 -1
  9. package/dist/types/discovery/index.d.ts +1 -0
  10. package/dist/types/export/html/template.generated.d.ts +1 -1
  11. package/dist/types/export/ttsr.d.ts +14 -3
  12. package/dist/types/gjc-runtime/deep-interview-state.d.ts +5 -2
  13. package/dist/types/gjc-runtime/session-state-sidecar.d.ts +7 -0
  14. package/dist/types/gjc-runtime/state-writer.d.ts +3 -0
  15. package/dist/types/gjc-runtime/team-runtime.d.ts +26 -0
  16. package/dist/types/gjc-runtime/ultragoal-receipt-freshness.d.ts +42 -0
  17. package/dist/types/gjc-runtime/ultragoal-runtime.d.ts +119 -13
  18. package/dist/types/goals/runtime.d.ts +1 -0
  19. package/dist/types/hooks/native-skill-hook.d.ts +12 -0
  20. package/dist/types/modes/components/assistant-message.d.ts +5 -1
  21. package/dist/types/modes/components/status-line.d.ts +9 -0
  22. package/dist/types/modes/controllers/command-controller.d.ts +1 -0
  23. package/dist/types/modes/controllers/event-controller.d.ts +5 -0
  24. package/dist/types/modes/interactive-mode.d.ts +2 -0
  25. package/dist/types/modes/rpc/rpc-types.d.ts +14 -0
  26. package/dist/types/modes/shared/agent-wire/event-contract.d.ts +19 -0
  27. package/dist/types/modes/shared/agent-wire/event-envelope.d.ts +23 -1
  28. package/dist/types/modes/types.d.ts +1 -0
  29. package/dist/types/modes/utils/ui-helpers.d.ts +1 -0
  30. package/dist/types/notifications/config.d.ts +6 -0
  31. package/dist/types/notifications/telegram-daemon.d.ts +8 -0
  32. package/dist/types/notifications/threaded-render.d.ts +6 -0
  33. package/dist/types/sdk.d.ts +3 -0
  34. package/dist/types/session/agent-session.d.ts +43 -1
  35. package/dist/types/session/file-mention-pruning.d.ts +8 -0
  36. package/dist/types/session/messages.d.ts +4 -0
  37. package/dist/types/session/session-manager.d.ts +17 -1
  38. package/dist/types/skill-state/active-state.d.ts +18 -1
  39. package/dist/types/system-prompt.d.ts +18 -2
  40. package/dist/types/task/executor.d.ts +1 -0
  41. package/dist/types/task/fork-context-budget.d.ts +7 -0
  42. package/dist/types/task/token-log.d.ts +30 -0
  43. package/dist/types/task/token-log.test.d.ts +1 -0
  44. package/dist/types/tools/bisect.d.ts +81 -0
  45. package/dist/types/tools/index.d.ts +1 -0
  46. package/dist/types/tools/search.d.ts +2 -2
  47. package/dist/types/utils/file-mentions.d.ts +14 -0
  48. package/dist/types/utils/git.d.ts +32 -0
  49. package/dist/types/web/scrapers/youtube.d.ts +8 -0
  50. package/package.json +7 -7
  51. package/src/append-only-mode.ts +58 -0
  52. package/src/autoresearch/prompt.md +1 -1
  53. package/src/autoresearch/tools/log-experiment.ts +1 -1
  54. package/src/capability/rule.ts +9 -0
  55. package/src/cli/config-cli.ts +27 -3
  56. package/src/cli/fixture-report.test.ts +230 -0
  57. package/src/cli/fixture-report.ts +171 -0
  58. package/src/cli/nofile-limit.ts +48 -0
  59. package/src/cli/update-cli.ts +85 -9
  60. package/src/cli.ts +26 -0
  61. package/src/commands/contribution-prep.ts +1 -1
  62. package/src/config/settings-schema.ts +68 -1
  63. package/src/defaults/gjc/skills/deep-interview/SKILL.md +12 -13
  64. package/src/defaults/gjc/skills/ralplan/SKILL.md +30 -93
  65. package/src/defaults/gjc/skills/team/SKILL.md +16 -13
  66. package/src/defaults/gjc/skills/ultragoal/SKILL.md +59 -4
  67. package/src/defaults/gjc/skills/ultragoal/pipeline-validation-contracts.md +30 -0
  68. package/src/defaults/gjc-defaults.ts +9 -0
  69. package/src/discovery/helpers.ts +10 -0
  70. package/src/discovery/index.ts +1 -0
  71. package/src/export/html/template.generated.ts +1 -1
  72. package/src/export/html/template.js +60 -2
  73. package/src/export/ttsr.ts +96 -18
  74. package/src/extensibility/extensions/runner.ts +181 -219
  75. package/src/extensibility/gjc-plugins/runtime-adapters.ts +156 -44
  76. package/src/gjc-runtime/deep-interview-state.ts +31 -2
  77. package/src/gjc-runtime/session-state-sidecar.ts +208 -13
  78. package/src/gjc-runtime/state-writer.ts +15 -0
  79. package/src/gjc-runtime/team-runtime.ts +164 -15
  80. package/src/gjc-runtime/ultragoal-guard.ts +55 -66
  81. package/src/gjc-runtime/ultragoal-receipt-freshness.ts +315 -0
  82. package/src/gjc-runtime/ultragoal-runtime.ts +1593 -143
  83. package/src/gjc-runtime/workflow-manifest.generated.json +113 -2
  84. package/src/gjc-runtime/workflow-manifest.ts +74 -1
  85. package/src/goals/runtime.ts +10 -2
  86. package/src/hindsight/backend.ts +1 -3
  87. package/src/hooks/native-skill-hook.ts +95 -4
  88. package/src/internal-urls/docs-index.generated.ts +14 -5
  89. package/src/main.ts +43 -2
  90. package/src/memories/index.ts +12 -1
  91. package/src/modes/components/assistant-message.ts +241 -82
  92. package/src/modes/components/custom-editor.ts +1 -0
  93. package/src/modes/components/status-line.ts +167 -54
  94. package/src/modes/components/tool-execution.ts +44 -19
  95. package/src/modes/controllers/command-controller.ts +39 -2
  96. package/src/modes/controllers/event-controller.ts +66 -47
  97. package/src/modes/controllers/input-controller.ts +1 -1
  98. package/src/modes/interactive-mode.ts +9 -1
  99. package/src/modes/rpc/rpc-mode.ts +29 -3
  100. package/src/modes/rpc/rpc-types.ts +9 -0
  101. package/src/modes/shared/agent-wire/command-validation.ts +4 -0
  102. package/src/modes/shared/agent-wire/event-contract.ts +22 -0
  103. package/src/modes/shared/agent-wire/event-envelope.ts +82 -0
  104. package/src/modes/shared/agent-wire/scopes.ts +1 -0
  105. package/src/modes/types.ts +1 -0
  106. package/src/modes/utils/ui-helpers.ts +17 -4
  107. package/src/notifications/config.ts +16 -4
  108. package/src/notifications/index.ts +78 -5
  109. package/src/notifications/telegram-daemon-cli.ts +2 -2
  110. package/src/notifications/telegram-daemon-control.ts +2 -2
  111. package/src/notifications/telegram-daemon.ts +85 -8
  112. package/src/notifications/threaded-render.ts +17 -1
  113. package/src/prompts/agents/architect.md +19 -10
  114. package/src/prompts/agents/critic.md +16 -18
  115. package/src/prompts/agents/executor.md +14 -11
  116. package/src/prompts/agents/planner.md +18 -13
  117. package/src/prompts/ci-green-request.md +3 -8
  118. package/src/prompts/goals/goal-continuation.md +0 -5
  119. package/src/prompts/goals/goal-mode-active.md +0 -5
  120. package/src/prompts/system/agent-creation-architect.md +2 -2
  121. package/src/prompts/system/plan-mode-active.md +1 -1
  122. package/src/prompts/system/plan-mode-subagent.md +3 -3
  123. package/src/prompts/system/project-prompt.md +0 -13
  124. package/src/prompts/system/subagent-system-prompt.md +3 -2
  125. package/src/prompts/system/system-prompt.md +26 -19
  126. package/src/prompts/system/volatile-project-context.md +10 -0
  127. package/src/prompts/system/web-search.md +4 -7
  128. package/src/prompts/tools/apply-patch.md +6 -0
  129. package/src/prompts/tools/ast-edit.md +1 -0
  130. package/src/prompts/tools/bash.md +1 -1
  131. package/src/prompts/tools/bisect.md +22 -0
  132. package/src/prompts/tools/browser.md +1 -1
  133. package/src/prompts/tools/cron.md +0 -2
  134. package/src/prompts/tools/eval.md +2 -2
  135. package/src/prompts/tools/find.md +1 -1
  136. package/src/prompts/tools/github.md +6 -6
  137. package/src/prompts/tools/goal.md +1 -3
  138. package/src/prompts/tools/image-gen.md +12 -5
  139. package/src/prompts/tools/irc.md +5 -4
  140. package/src/prompts/tools/job.md +2 -2
  141. package/src/prompts/tools/lsp.md +1 -1
  142. package/src/prompts/tools/monitor.md +2 -2
  143. package/src/prompts/tools/patch.md +2 -2
  144. package/src/prompts/tools/replace.md +2 -21
  145. package/src/prompts/tools/rewind.md +1 -1
  146. package/src/prompts/tools/search-tool-bm25.md +1 -11
  147. package/src/prompts/tools/search.md +1 -1
  148. package/src/prompts/tools/skill.md +3 -2
  149. package/src/prompts/tools/ssh.md +2 -1
  150. package/src/prompts/tools/subagent.md +2 -2
  151. package/src/prompts/tools/task.md +4 -5
  152. package/src/prompts/tools/todo-write.md +1 -1
  153. package/src/prompts/tools/vim.md +1 -0
  154. package/src/prompts/tools/web-search.md +2 -0
  155. package/src/prompts/tools/write.md +8 -2
  156. package/src/sdk.ts +32 -22
  157. package/src/session/agent-session.ts +732 -107
  158. package/src/session/file-mention-pruning.ts +64 -0
  159. package/src/session/messages.ts +4 -0
  160. package/src/session/session-manager.ts +59 -2
  161. package/src/skill-state/active-state.ts +141 -5
  162. package/src/skill-state/deep-interview-mutation-guard.ts +67 -3
  163. package/src/slash-commands/builtin-registry.ts +65 -0
  164. package/src/system-prompt.ts +57 -16
  165. package/src/task/executor.ts +125 -20
  166. package/src/task/fork-context-advisory.ts +2 -7
  167. package/src/task/fork-context-budget.ts +9 -0
  168. package/src/task/index.ts +4 -3
  169. package/src/task/output-manager.ts +4 -2
  170. package/src/task/token-log.test.ts +188 -0
  171. package/src/task/token-log.ts +179 -0
  172. package/src/tools/bisect.ts +367 -0
  173. package/src/tools/index.ts +3 -0
  174. package/src/tools/output-meta.ts +116 -4
  175. package/src/tools/search.ts +8 -2
  176. package/src/tools/skill.ts +15 -0
  177. package/src/utils/file-mentions.ts +39 -7
  178. package/src/utils/git.ts +73 -0
  179. package/src/web/scrapers/youtube.ts +24 -12
@@ -49,14 +49,38 @@ type CliSettingDef = {
49
49
 
50
50
  const ALL_SETTING_PATHS = Object.keys(SETTINGS_SCHEMA) as SettingPath[];
51
51
  const REDACTED_SECRET_VALUE = "<redacted>";
52
- const SECRET_SETTING_SEGMENT_PATTERN = /(?:api[-_]?key|token|secret|password|passwd|pwd|credential)/i;
52
+ const SECRET_SETTING_WORDS = new Set(["token", "secret", "password", "passwd", "pwd", "credential", "credentials"]);
53
+ const SECRET_SETTING_COMPOUND_PREFIXES = [
54
+ "api",
55
+ "auth",
56
+ "access",
57
+ "refresh",
58
+ "bearer",
59
+ "session",
60
+ "client",
61
+ "broker",
62
+ "bot",
63
+ "basic",
64
+ ];
65
+ const SECRET_SETTING_COMPOUND_SUFFIXES = ["token", "secret", "password", "credential"];
66
+
67
+ function isSecretSettingSegment(segment: string): boolean {
68
+ const normalized = segment.toLowerCase();
69
+ if (SECRET_SETTING_WORDS.has(normalized)) return true;
70
+ if (/api[-_]?key/i.test(segment)) return true;
71
+ const words = normalized.split(/[-_]/).filter(Boolean);
72
+ if (words.some(word => SECRET_SETTING_WORDS.has(word))) return true;
73
+ return SECRET_SETTING_COMPOUND_PREFIXES.some(prefix =>
74
+ SECRET_SETTING_COMPOUND_SUFFIXES.some(suffix => normalized === `${prefix}${suffix}`),
75
+ );
76
+ }
53
77
 
54
78
  function isSecretSettingPath(path: string): boolean {
55
- return path.split(".").some(segment => SECRET_SETTING_SEGMENT_PATTERN.test(segment));
79
+ return path.split(".").some(segment => isSecretSettingSegment(segment));
56
80
  }
57
81
 
58
82
  function redactConfigValue(path: string, value: unknown, showSecrets?: boolean): unknown {
59
- if (showSecrets || typeof value !== "string" || !isSecretSettingPath(path)) {
83
+ if (showSecrets || value === undefined || value === null || !isSecretSettingPath(path)) {
60
84
  return value;
61
85
  }
62
86
  return REDACTED_SECRET_VALUE;
@@ -0,0 +1,230 @@
1
+ import { describe, expect, it } from "bun:test";
2
+ import { mkdir, mkdtemp, rm, writeFile } from "node:fs/promises";
3
+ import { tmpdir } from "node:os";
4
+ import { join } from "node:path";
5
+ import { sessionRoot } from "../gjc-runtime/session-layout";
6
+ import type { TaskTokenLog } from "../task/types";
7
+ import { buildFixtureReport, type LiveRunReportShape, runFixtureReport } from "./fixture-report";
8
+
9
+ function expectParseReportCompatible(report: LiveRunReportShape, fixtureId: string): void {
10
+ expect(report.schemaVersion).toBe(1);
11
+ expect(typeof report.binaryId).toBe("string");
12
+ expect(report.binaryId.length).toBeGreaterThan(0);
13
+ expect(report.fixtureId).toBe(fixtureId);
14
+ for (const value of Object.values(report.totals)) {
15
+ expect(typeof value).toBe("number");
16
+ expect(Number.isFinite(value)).toBe(true);
17
+ }
18
+ expect(report.receiptArtifactRatio).toBeNull();
19
+ expect(report.spawnDecisions).toBeNull();
20
+ expect(report.roi).toBeNull();
21
+ }
22
+
23
+ describe("fixture report", () => {
24
+ it("reports empty logs as a schema-v1 null-cache-hit report", () => {
25
+ const report = buildFixtureReport("empty-fixture", []);
26
+
27
+ expectParseReportCompatible(report, "empty-fixture");
28
+ expect(report.totals).toEqual({
29
+ turns: 0,
30
+ inputTokens: 0,
31
+ outputTokens: 0,
32
+ cacheReadTokens: 0,
33
+ cacheWriteTokens: 0,
34
+ totalTokens: 0,
35
+ });
36
+ expect(report.cacheHitRate).toBeNull();
37
+ });
38
+
39
+ it("sums deterministic fixed-fixture raw token buckets without folding cache into input", () => {
40
+ const logs: TaskTokenLog[] = [
41
+ {
42
+ subagentId: "root",
43
+ agent: "main",
44
+ turn: 1,
45
+ at: "2026-01-01T00:00:00.000Z",
46
+ input: 100,
47
+ output: 20,
48
+ cacheRead: 40,
49
+ cacheWrite: 10,
50
+ totalTokens: 170,
51
+ model: "fixture-model",
52
+ },
53
+ {
54
+ subagentId: "1-executor",
55
+ agent: "executor",
56
+ turn: 1,
57
+ at: "2026-01-01T00:00:01.000Z",
58
+ input: 60,
59
+ output: 15,
60
+ cacheRead: 20,
61
+ cacheWrite: 5,
62
+ totalTokens: 100,
63
+ model: "fixture-model",
64
+ },
65
+ ];
66
+ const report = buildFixtureReport("fixed-fixture", logs);
67
+ const inputTokens = logs.reduce((sum, log) => sum + log.input, 0);
68
+ const outputTokens = logs.reduce((sum, log) => sum + log.output, 0);
69
+ const cacheReadTokens = logs.reduce((sum, log) => sum + log.cacheRead, 0);
70
+ const cacheWriteTokens = logs.reduce((sum, log) => sum + log.cacheWrite, 0);
71
+ const totalTokens = inputTokens + outputTokens + cacheReadTokens + cacheWriteTokens;
72
+
73
+ expectParseReportCompatible(report, "fixed-fixture");
74
+ expect(report.totals).toEqual({
75
+ turns: logs.length,
76
+ inputTokens,
77
+ outputTokens,
78
+ cacheReadTokens,
79
+ cacheWriteTokens,
80
+ totalTokens,
81
+ });
82
+ expect(report.cacheHitRate).toBe(cacheReadTokens / (inputTokens + cacheReadTokens));
83
+ });
84
+
85
+ it("emits only parseReport-compatible schema-v1 fields", () => {
86
+ const report = buildFixtureReport("shape-fixture", [
87
+ {
88
+ subagentId: "root",
89
+ turn: 1,
90
+ at: "2026-01-01T00:00:00.000Z",
91
+ input: 0,
92
+ output: 0,
93
+ cacheRead: 1,
94
+ cacheWrite: 0,
95
+ totalTokens: 1,
96
+ },
97
+ ]);
98
+
99
+ expect(Object.keys(report).sort()).toEqual([
100
+ "binaryId",
101
+ "cacheHitRate",
102
+ "fixtureId",
103
+ "receiptArtifactRatio",
104
+ "roi",
105
+ "schemaVersion",
106
+ "spawnDecisions",
107
+ "totals",
108
+ ]);
109
+ expect(Object.keys(report.totals).sort()).toEqual([
110
+ "cacheReadTokens",
111
+ "cacheWriteTokens",
112
+ "inputTokens",
113
+ "outputTokens",
114
+ "totalTokens",
115
+ "turns",
116
+ ]);
117
+ expectParseReportCompatible(report, "shape-fixture");
118
+ });
119
+ });
120
+
121
+ async function captureRun(fixtureId: string): Promise<{ stdout: string; stderr: string; exitCode: number }> {
122
+ const originalCwd = process.cwd();
123
+ const originalStdout = process.stdout.write.bind(process.stdout);
124
+ const originalStderr = process.stderr.write.bind(process.stderr);
125
+ let exitCode = 0;
126
+ let stdout = "";
127
+ let stderr = "";
128
+ process.stdout.write = ((chunk: string | Uint8Array): boolean => {
129
+ stdout += typeof chunk === "string" ? chunk : Buffer.from(chunk).toString("utf-8");
130
+ return true;
131
+ }) as typeof process.stdout.write;
132
+ process.stderr.write = ((chunk: string | Uint8Array): boolean => {
133
+ stderr += typeof chunk === "string" ? chunk : Buffer.from(chunk).toString("utf-8");
134
+ return true;
135
+ }) as typeof process.stderr.write;
136
+ try {
137
+ exitCode = await runFixtureReport(fixtureId);
138
+ return { stdout, stderr, exitCode };
139
+ } finally {
140
+ process.stdout.write = originalStdout;
141
+ process.stderr.write = originalStderr;
142
+ process.chdir(originalCwd);
143
+ }
144
+ }
145
+
146
+ describe("runFixtureReport CLI degradation", () => {
147
+ it("exits non-zero without emitting JSON for an unknown fixture id", async () => {
148
+ await withTempCwd(async () => {
149
+ const { stdout, stderr, exitCode } = await captureRun("definitely-not-a-real-session-id");
150
+ expect(exitCode).toBe(1);
151
+ expect(stdout).toBe("");
152
+ expect(stderr).toContain("unknown fixture id");
153
+ });
154
+ });
155
+
156
+ it("reads only the requested session token-log directory", async () => {
157
+ await withTempCwd(async () => {
158
+ const firstSession = "fixture-session-a";
159
+ const secondSession = "fixture-session-b";
160
+ await writeTokenLog(firstSession, { input: 100, output: 10, cacheRead: 0, cacheWrite: 0, totalTokens: 110 });
161
+ await writeTokenLog(secondSession, { input: 900, output: 90, cacheRead: 0, cacheWrite: 0, totalTokens: 990 });
162
+
163
+ const { stdout, exitCode } = await captureRun(firstSession);
164
+ const report = JSON.parse(stdout) as LiveRunReportShape;
165
+
166
+ expect(exitCode).toBe(0);
167
+ expect(report.fixtureId).toBe(firstSession);
168
+ expect(report.totals.totalTokens).toBe(110);
169
+ });
170
+ });
171
+
172
+ it("emits deterministic PR9 default-candidate fixture reports", async () => {
173
+ await withTempCwd(async () => {
174
+ const { stdout, exitCode } = await captureRun("pr9.read-artifact-spill-threshold.after");
175
+ const report = JSON.parse(stdout) as LiveRunReportShape;
176
+
177
+ expect(exitCode).toBe(0);
178
+ expectParseReportCompatible(report, "pr9.read-artifact-spill-threshold.after");
179
+ expect(report.totals.totalTokens).toBe(20_100);
180
+ });
181
+ });
182
+
183
+ it("exits non-zero without emitting JSON for a corrupt token-log", async () => {
184
+ await withTempCwd(async () => {
185
+ const id = "corrupt-session";
186
+ const dir = join(sessionRoot(process.cwd(), id), "token-logs");
187
+ await mkdir(dir, { recursive: true });
188
+ await writeFile(join(dir, "token-log.jsonl"), "not json\n{}\n", "utf-8");
189
+ const { stdout, stderr, exitCode } = await captureRun(id);
190
+ expect(exitCode).toBe(1);
191
+ expect(stdout).toBe("");
192
+ expect(stderr).toContain("failed to build fixture report");
193
+ });
194
+ });
195
+ });
196
+
197
+ async function withTempCwd(fn: (cwd: string) => Promise<void>): Promise<void> {
198
+ const originalCwd = process.cwd();
199
+ const dir = await mkdtemp(join(tmpdir(), "gjc-fixture-report-"));
200
+ process.chdir(dir);
201
+ try {
202
+ await fn(dir);
203
+ } finally {
204
+ // Restore the real cwd BEFORE removing the temp dir, otherwise the process
205
+ // is left sitting in a deleted directory and later tests in the same
206
+ // process fail with ENOENT on relative-path/process.cwd() operations.
207
+ process.chdir(originalCwd);
208
+ await rm(dir, { recursive: true, force: true });
209
+ }
210
+ }
211
+
212
+ async function writeTokenLog(
213
+ sessionId: string,
214
+ metrics: { input: number; output: number; cacheRead: number; cacheWrite: number; totalTokens: number },
215
+ ): Promise<void> {
216
+ const dir = join(sessionRoot(process.cwd(), sessionId), "token-logs");
217
+ await mkdir(dir, { recursive: true });
218
+ await writeFile(
219
+ join(dir, "token-log.jsonl"),
220
+ `${JSON.stringify({
221
+ subagentId: "root",
222
+ agent: "main",
223
+ turn: 1,
224
+ at: "2026-01-01T00:00:00.000Z",
225
+ model: "fixture-model",
226
+ ...metrics,
227
+ })}\n`,
228
+ "utf-8",
229
+ );
230
+ }
@@ -0,0 +1,171 @@
1
+ import * as fs from "node:fs/promises";
2
+ import * as path from "node:path";
3
+ import type { GjcSessionContext } from "../gjc-runtime/session-layout";
4
+ import { resolveGjcSessionForRead, SessionResolutionError } from "../gjc-runtime/session-resolution";
5
+ import { computeTaskTokenMetrics, readTaskTokenLogs } from "../task/token-log";
6
+ import type { TaskTokenLog } from "../task/types";
7
+
8
+ const LIVE_RUNNER_SCHEMA_VERSION = 1;
9
+ const BINARY_ID = "gjc";
10
+
11
+ function deterministicLog(
12
+ input: number,
13
+ output: number,
14
+ cacheRead: number,
15
+ cacheWrite: number,
16
+ totalTokens: number,
17
+ ): TaskTokenLog {
18
+ return {
19
+ subagentId: "root",
20
+ agent: "main",
21
+ turn: 1,
22
+ at: "2026-01-01T00:00:00.000Z",
23
+ input,
24
+ output,
25
+ cacheRead,
26
+ cacheWrite,
27
+ totalTokens,
28
+ model: "fixture-model",
29
+ };
30
+ }
31
+
32
+ export interface LiveRunReportShape {
33
+ schemaVersion: 1;
34
+ binaryId: string;
35
+ fixtureId: string;
36
+ totals: {
37
+ turns: number;
38
+ inputTokens: number;
39
+ outputTokens: number;
40
+ cacheReadTokens: number;
41
+ cacheWriteTokens: number;
42
+ totalTokens: number;
43
+ };
44
+ cacheHitRate: number | null;
45
+ receiptArtifactRatio: number | null;
46
+ spawnDecisions: number | null;
47
+ roi: number | null;
48
+ }
49
+
50
+ const DETERMINISTIC_FIXTURES: Record<string, readonly TaskTokenLog[]> = {
51
+ "fixed-fixture": [
52
+ {
53
+ subagentId: "root",
54
+ agent: "main",
55
+ turn: 1,
56
+ at: "2026-01-01T00:00:00.000Z",
57
+ input: 100,
58
+ output: 20,
59
+ cacheRead: 40,
60
+ cacheWrite: 10,
61
+ totalTokens: 170,
62
+ model: "fixture-model",
63
+ },
64
+ {
65
+ subagentId: "1-executor",
66
+ agent: "executor",
67
+ turn: 1,
68
+ at: "2026-01-01T00:00:01.000Z",
69
+ input: 60,
70
+ output: 15,
71
+ cacheRead: 20,
72
+ cacheWrite: 5,
73
+ totalTokens: 100,
74
+ model: "fixture-model",
75
+ },
76
+ ],
77
+ };
78
+ const DEFAULT_REDUCTION_FIXTURE_LOGS: Record<string, readonly TaskTokenLog[]> = {
79
+ "pr9.task-recursion.before": [deterministicLog(2400, 420, 600, 200, 3620)],
80
+ "pr9.task-recursion.after": [deterministicLog(1800, 360, 450, 150, 2760)],
81
+ "pr9.output-caps.before": [deterministicLog(22_000, 3_200, 0, 0, 25_200)],
82
+ "pr9.output-caps.after": [deterministicLog(18_000, 2_400, 0, 0, 20_400)],
83
+ "pr9.max-inline-result-bytes.before": [deterministicLog(48_000, 2_000, 0, 0, 50_000)],
84
+ "pr9.max-inline-result-bytes.after": [deterministicLog(12_000, 1_600, 0, 0, 13_600)],
85
+ "pr9.read-artifact-spill-threshold.before": [deterministicLog(96_000, 2_500, 0, 0, 98_500)],
86
+ "pr9.read-artifact-spill-threshold.after": [deterministicLog(18_000, 2_100, 0, 0, 20_100)],
87
+ "pr9.maintenance-pruning.before": [deterministicLog(40_000, 2_000, 18_000, 6_000, 66_000)],
88
+ "pr9.maintenance-pruning.after": [deterministicLog(30_000, 1_800, 10_000, 3_000, 44_800)],
89
+ "pr9.append-only-provider-expansion.before": [deterministicLog(28_000, 1_700, 8_000, 2_500, 40_200)],
90
+ "pr9.append-only-provider-expansion.after": [deterministicLog(20_000, 1_700, 14_000, 2_500, 38_200)],
91
+ "pr9.rpc-compact-deltas.before": [deterministicLog(14_000, 1_200, 6_000, 1_500, 22_700)],
92
+ "pr9.rpc-compact-deltas.after": [deterministicLog(10_000, 900, 6_200, 1_500, 18_600)],
93
+ };
94
+
95
+ export function buildFixtureReport(fixtureId: string, logs: readonly TaskTokenLog[]): LiveRunReportShape {
96
+ const metrics = computeTaskTokenMetrics(logs);
97
+ return {
98
+ schemaVersion: LIVE_RUNNER_SCHEMA_VERSION,
99
+ binaryId: BINARY_ID,
100
+ fixtureId,
101
+ totals: {
102
+ turns: metrics.turns,
103
+ inputTokens: metrics.inputTokens,
104
+ outputTokens: metrics.outputTokens,
105
+ cacheReadTokens: metrics.cacheReadTokens,
106
+ cacheWriteTokens: metrics.cacheWriteTokens,
107
+ totalTokens: metrics.totalTokens,
108
+ },
109
+ cacheHitRate: logs.length === 0 ? null : metrics.cacheHitRate,
110
+ receiptArtifactRatio: null,
111
+ spawnDecisions: null,
112
+ roi: null,
113
+ };
114
+ }
115
+
116
+ export async function runFixtureReport(fixtureId: string): Promise<number> {
117
+ let resolved: ResolvedFixtureLogs;
118
+ try {
119
+ resolved = await resolveFixtureLogs(fixtureId);
120
+ } catch (error) {
121
+ // e.g. a corrupt token-log. Fail loudly rather than emitting an all-zero report.
122
+ process.stderr.write(
123
+ `failed to build fixture report for ${fixtureId}: ${error instanceof Error ? error.message : String(error)}\n`,
124
+ );
125
+ return 1;
126
+ }
127
+ if (resolved.kind === "unknown") {
128
+ // Neither a known deterministic fixture nor a resolvable GJC session. Emit a
129
+ // bounded error on stderr with a non-zero exit rather than a schema-valid
130
+ // all-zero report, which a before/after benchmark would misread as "0 tokens".
131
+ process.stderr.write(`unknown fixture id and no matching GJC session: ${fixtureId}\n`);
132
+ return 1;
133
+ }
134
+ process.stdout.write(JSON.stringify(buildFixtureReport(fixtureId, resolved.logs)));
135
+ return 0;
136
+ }
137
+
138
+ type ResolvedFixtureLogs =
139
+ | { readonly kind: "logs"; readonly logs: readonly TaskTokenLog[] }
140
+ | { readonly kind: "unknown" };
141
+
142
+ async function resolveFixtureLogs(fixtureId: string): Promise<ResolvedFixtureLogs> {
143
+ const deterministic = DETERMINISTIC_FIXTURES[fixtureId] ?? DEFAULT_REDUCTION_FIXTURE_LOGS[fixtureId];
144
+ if (deterministic) return { kind: "logs", logs: deterministic };
145
+ let session: GjcSessionContext;
146
+ try {
147
+ session = await resolveGjcSessionForRead(process.cwd(), {
148
+ flagValue: fixtureId,
149
+ envSessionId: process.env.GJC_SESSION_ID,
150
+ });
151
+ } catch (error) {
152
+ if (error instanceof SessionResolutionError) return { kind: "unknown" };
153
+ throw error;
154
+ }
155
+ // resolveGjcSessionForRead accepts any explicit flagValue as a session id
156
+ // without checking the dir exists, so a typo would otherwise yield a
157
+ // schema-valid all-zero report. Require the session root to exist; a real
158
+ // session with no turns yet still reads as a legitimate empty log set.
159
+ if (!(await directoryExists(session.sessionRoot))) return { kind: "unknown" };
160
+ const logs = await readTaskTokenLogs(path.join(session.sessionRoot, "token-logs"));
161
+ return { kind: "logs", logs };
162
+ }
163
+
164
+ async function directoryExists(dir: string): Promise<boolean> {
165
+ try {
166
+ const stat = await fs.stat(dir);
167
+ return stat.isDirectory();
168
+ } catch {
169
+ return false;
170
+ }
171
+ }
@@ -0,0 +1,48 @@
1
+ import { execFileSync } from "node:child_process";
2
+
3
+ export const RECOMMENDED_MACOS_NOFILE_LIMIT = 4096;
4
+
5
+ export function parseNoFileLimit(text: string): number | undefined {
6
+ const trimmed = text.trim();
7
+ if (!trimmed || trimmed === "unlimited") return undefined;
8
+ const value = Number(trimmed);
9
+ return Number.isSafeInteger(value) && value > 0 ? value : undefined;
10
+ }
11
+
12
+ export function buildMacOSNoFileLimitWarning(currentLimit: number): string {
13
+ return [
14
+ `Warning: macOS file descriptor limit is low (ulimit -n = ${currentLimit}).`,
15
+ 'GJC and project dev servers can hit EMFILE / "too many open files" while scanning or watching repositories.',
16
+ "For this terminal session, run:",
17
+ ` ulimit -n ${RECOMMENDED_MACOS_NOFILE_LIMIT}`,
18
+ "If your shell refuses that value, raise the per-user launchd limit and restart the terminal:",
19
+ ` sudo launchctl limit maxfiles ${RECOMMENDED_MACOS_NOFILE_LIMIT} 65536`,
20
+ "Avoid using huge values such as 2147483646 on macOS; they are commonly rejected or clamped.",
21
+ "Set GJC_SKIP_NOFILE_CHECK=1 to silence this preflight warning.",
22
+ ].join("\n");
23
+ }
24
+
25
+ export interface WarnIfMacOSNoFileLimitTooLowDeps {
26
+ platform?: NodeJS.Platform;
27
+ env?: NodeJS.ProcessEnv;
28
+ execFileSync?: typeof execFileSync;
29
+ writeStderr?: (text: string) => void;
30
+ }
31
+
32
+ export function warnIfMacOSNoFileLimitTooLow(deps: WarnIfMacOSNoFileLimitTooLowDeps = {}): void {
33
+ const platform = deps.platform ?? process.platform;
34
+ if (platform !== "darwin") return;
35
+ const env = deps.env ?? process.env;
36
+ if (env.GJC_SKIP_NOFILE_CHECK === "1" || env.GJC_SKIP_NOFILE_CHECK === "true") return;
37
+
38
+ const run = deps.execFileSync ?? execFileSync;
39
+ let currentLimit: number | undefined;
40
+ try {
41
+ currentLimit = parseNoFileLimit(run("/bin/sh", ["-lc", "ulimit -n"], { encoding: "utf8" }));
42
+ } catch {
43
+ return;
44
+ }
45
+ if (currentLimit === undefined || currentLimit >= RECOMMENDED_MACOS_NOFILE_LIMIT) return;
46
+ const write = deps.writeStderr ?? (text => process.stderr.write(text));
47
+ write(`${buildMacOSNoFileLimitWarning(currentLimit)}\n`);
48
+ }
@@ -552,31 +552,107 @@ async function updateViaNpm(packageName: string, expectedVersion: string): Promi
552
552
  }
553
553
 
554
554
  /**
555
- * Download a release binary to a target path, replacing an existing file.
555
+ * Flush a freshly written file's data to stable storage.
556
+ *
557
+ * Critical on network filesystems (e.g. NFS home directories): `pipeline`
558
+ * resolving does not guarantee the downloaded bytes are durable on the
559
+ * server, so the post-install `gjc --version` check can exec a binary whose
560
+ * pages are not yet consistent. The child then faults, the version check
561
+ * fails, and the update is rolled back with "could not verify updated
562
+ * version" even though the download itself succeeded. Explicitly fsyncing
563
+ * before the rename/exec avoids the race.
556
564
  */
557
- async function updateViaBinaryAt(targetPath: string, expectedVersion: string): Promise<void> {
558
- const binaryName = getBinaryName();
559
- const url = buildReleaseBinaryUrl(expectedVersion);
565
+ async function fsyncFile(filePath: string): Promise<void> {
566
+ const handle = await fs.promises.open(filePath, "r+");
567
+ try {
568
+ await handle.sync();
569
+ } finally {
570
+ await handle.close();
571
+ }
572
+ }
560
573
 
561
- const tempPath = `${targetPath}.new`;
562
- const backupPath = `${targetPath}.bak`;
563
- console.log(chalk.dim(`Downloading ${binaryName}…`));
574
+ export async function fsyncFileForTest(filePath: string): Promise<void> {
575
+ return fsyncFile(filePath);
576
+ }
564
577
 
578
+ /**
579
+ * Download a release binary to a temp path, throwing a friendly error when the
580
+ * release asset cannot be fetched.
581
+ */
582
+ async function downloadBinaryTo(url: string, tempPath: string, binaryName: string): Promise<void> {
565
583
  const response = await fetch(url, { redirect: "follow" });
566
584
  if (!response.ok || !response.body) {
567
585
  throw new Error(formatBinaryDownloadFailureMessage(binaryName, url, response.statusText || response.status));
568
586
  }
569
587
  const fileStream = fs.createWriteStream(tempPath, { mode: 0o755 });
570
588
  await pipeline(response.body, fileStream);
589
+ }
590
+
591
+ /** Injectable steps of the binary update flow (seams for testing ordering). */
592
+ export interface BinaryUpdateFlow {
593
+ download(url: string, tempPath: string): Promise<void>;
594
+ fsync(filePath: string): Promise<void>;
595
+ replace(options: BinaryReplacementOptions): Promise<InstalledVersionVerification>;
596
+ verifyInstalledVersion(expectedVersion: string): Promise<InstalledVersionVerification>;
597
+ /** Best-effort cleanup of the temp file when the flow aborts before replace. */
598
+ removeTemp?(filePath: string): Promise<void>;
599
+ /** Called once fsync has succeeded, right before replacement begins. */
600
+ beforeReplace?(): void;
601
+ }
571
602
 
572
- console.log(chalk.dim("Installing update..."));
573
- const verification = await replaceBinaryForUpdate({
603
+ /**
604
+ * Orchestrate download → fsync → replace → verify with a strict ordering
605
+ * contract: the downloaded temp binary MUST be flushed to stable storage
606
+ * before it is published (renamed into place) or exec'd for verification.
607
+ *
608
+ * If fsync fails the temp bytes are not durable, so we abort before
609
+ * replacement/verification and clean up the temp file rather than installing a
610
+ * possibly-truncated binary.
611
+ */
612
+ export async function runBinaryUpdateFlow(
613
+ targetPath: string,
614
+ url: string,
615
+ expectedVersion: string,
616
+ flow: BinaryUpdateFlow,
617
+ ): Promise<InstalledVersionVerification> {
618
+ const tempPath = `${targetPath}.new`;
619
+ const backupPath = `${targetPath}.bak`;
620
+
621
+ await flow.download(url, tempPath);
622
+ try {
623
+ await flow.fsync(tempPath);
624
+ } catch (err) {
625
+ if (flow.removeTemp) await flow.removeTemp(tempPath);
626
+ throw err;
627
+ }
628
+
629
+ flow.beforeReplace?.();
630
+ return flow.replace({
574
631
  targetPath,
575
632
  tempPath,
576
633
  backupPath,
577
634
  expectedVersion,
635
+ verifyInstalledVersion: flow.verifyInstalledVersion,
636
+ });
637
+ }
638
+
639
+ /**
640
+ * Download a release binary to a target path, replacing an existing file.
641
+ */
642
+ async function updateViaBinaryAt(targetPath: string, expectedVersion: string): Promise<void> {
643
+ const binaryName = getBinaryName();
644
+ const url = buildReleaseBinaryUrl(expectedVersion);
645
+ console.log(chalk.dim(`Downloading ${binaryName}…`));
646
+
647
+ const verification = await runBinaryUpdateFlow(targetPath, url, expectedVersion, {
648
+ download: (downloadUrl, tempPath) => downloadBinaryTo(downloadUrl, tempPath, binaryName),
649
+ fsync: fsyncFile,
650
+ replace: replaceBinaryForUpdate,
578
651
  verifyInstalledVersion: verifyInstalledRuntime,
652
+ removeTemp: unlinkIfExists,
653
+ beforeReplace: () => console.log(chalk.dim("Installing update...")),
579
654
  });
655
+
580
656
  printVerifiedVersion(expectedVersion);
581
657
  if (verification.cleanupWarning) console.warn(chalk.yellow(verification.cleanupWarning));
582
658
  printRestartGuidance();
package/src/cli.ts CHANGED
@@ -6,6 +6,7 @@
6
6
  */
7
7
  import { Args, type CliConfig, Command, type CommandEntry, Flags, run } from "@gajae-code/utils/cli";
8
8
  import { APP_NAME, formatBunRuntimeError, MIN_BUN_VERSION, VERSION } from "@gajae-code/utils/dirs";
9
+ import { runFixtureReport } from "./cli/fixture-report";
9
10
 
10
11
  if (Bun.semver.order(Bun.version, MIN_BUN_VERSION) < 0) {
11
12
  process.stderr.write(
@@ -76,6 +77,9 @@ async function installRuntimeGlobals(): Promise<void> {
76
77
  // `HTTP2Unsupported`. See @gajae-code/ai/utils/h2-fetch for details.
77
78
  installH2Fetch();
78
79
 
80
+ const { warnIfMacOSNoFileLimitTooLow } = await import("./cli/nofile-limit");
81
+ warnIfMacOSNoFileLimitTooLow();
82
+
79
83
  // Strip macOS malloc-stack-logging env vars before any subprocess is spawned.
80
84
  // Otherwise every child bun process (subagents, plugin installs, ptree spawns,
81
85
  // etc.) prints a `MallocStackLogging: can't turn off …` warning to stderr.
@@ -96,6 +100,17 @@ async function runNotifyDaemonInternalFastPath(argv: string[]): Promise<void> {
96
100
  await runNotifyCommand(cmd);
97
101
  }
98
102
 
103
+ function rootFixtureArg(argv: string[]): { present: boolean; id: string | undefined } {
104
+ for (let i = 0; i < argv.length; i++) {
105
+ const arg = argv[i];
106
+ // Stop at the first subcommand token so a `--fixture` flag belonging to a
107
+ // subcommand never hijacks the root fast-path into fixture-report mode.
108
+ if (isSubcommand(arg)) return { present: false, id: undefined };
109
+ if (arg === "--fixture") return { present: true, id: argv[i + 1] };
110
+ }
111
+ return { present: false, id: undefined };
112
+ }
113
+
99
114
  function hasRootFastFlag(argv: string[], flags: readonly string[]): boolean {
100
115
  for (const arg of argv) {
101
116
  if (isSubcommand(arg)) return false;
@@ -233,6 +248,17 @@ export async function runCli(argv: string[]): Promise<void> {
233
248
  await runSmokeTest();
234
249
  return;
235
250
  }
251
+ const fixtureArg = rootFixtureArg(argv);
252
+ if (fixtureArg.present) {
253
+ const id = fixtureArg.id;
254
+ if (!id || id.startsWith("-")) {
255
+ process.stderr.write(`${APP_NAME} --fixture requires a fixture id\n`);
256
+ process.exitCode = 1;
257
+ return;
258
+ }
259
+ process.exitCode = await runFixtureReport(id);
260
+ return;
261
+ }
236
262
  if (hasRootHelpFlag(argv)) {
237
263
  const { renderRootHelp } = await import("@gajae-code/utils/cli");
238
264
  const { getExtraHelpText } = await import("./cli/fast-help");
@@ -1,5 +1,4 @@
1
1
  import { Command, Flags } from "@gajae-code/utils/cli";
2
- import { prepareContributionPrep } from "../session/contribution-prep";
3
2
 
4
3
  function writeText(lines: string[]): void {
5
4
  process.stdout.write(`${lines.join("\n")}\n`);
@@ -19,6 +18,7 @@ export default class ContributionPrep extends Command {
19
18
 
20
19
  async run(): Promise<void> {
21
20
  const { flags } = await this.parse(ContributionPrep);
21
+ const { prepareContributionPrep } = await import("../session/contribution-prep");
22
22
  const cwd = process.cwd();
23
23
  const result = await prepareContributionPrep(
24
24
  {