@tekyzinc/gsd-t 3.29.10 → 4.0.12
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.
- package/CHANGELOG.md +176 -0
- package/bin/gsd-t-parallel.cjs +64 -13
- package/bin/gsd-t-test-data-adapters/file-json-array.cjs +27 -3
- package/bin/gsd-t-test-data-adapters/localstorage-key-prefix.cjs +6 -1
- package/bin/gsd-t-test-data-adapters/sqlite-table-where.cjs +16 -1
- package/bin/gsd-t-test-data-ledger.cjs +1 -0
- package/bin/gsd-t.js +48 -376
- package/bin/journey-coverage.cjs +6 -3
- package/bin/parallel-cli.cjs +13 -1
- package/commands/gsd-t-debug.md +47 -533
- package/commands/gsd-t-design-decompose.md +24 -497
- package/commands/gsd-t-doc-ripple.md +23 -139
- package/commands/gsd-t-execute.md +41 -958
- package/commands/gsd-t-help.md +0 -2
- package/commands/gsd-t-impact.md +26 -295
- package/commands/gsd-t-integrate.md +44 -369
- package/commands/gsd-t-milestone.md +25 -124
- package/commands/gsd-t-partition.md +28 -539
- package/commands/gsd-t-plan.md +26 -472
- package/commands/gsd-t-prd.md +25 -311
- package/commands/gsd-t-quick.md +1 -1
- package/commands/gsd-t-resume.md +0 -33
- package/commands/gsd-t-verify.md +52 -569
- package/commands/gsd-t-wave.md +41 -430
- package/package.json +1 -1
- package/scripts/gsd-t-calibration-hook.js +3 -1
- package/scripts/hooks/pre-commit-journey-coverage +0 -2
- package/scripts/hooks/pre-commit-playwright-gate +0 -2
- package/templates/CLAUDE-global.md +43 -173
- package/templates/CLAUDE-project.md +118 -104
- package/templates/prompts/design-verify-subagent.md +3 -0
- package/templates/prompts/qa-subagent.md +3 -0
- package/templates/prompts/red-team-subagent.md +5 -2
- package/templates/workflows/_lib.js +176 -0
- package/templates/workflows/gsd-t-debug.workflow.js +86 -0
- package/templates/workflows/gsd-t-execute.workflow.js +206 -0
- package/templates/workflows/gsd-t-integrate.workflow.js +71 -0
- package/templates/workflows/gsd-t-phase.workflow.js +94 -0
- package/templates/workflows/gsd-t-quick.workflow.js +72 -0
- package/templates/workflows/gsd-t-verify.workflow.js +348 -0
- package/templates/workflows/gsd-t-wave.workflow.js +43 -0
- package/bin/check-headless-sessions.js +0 -140
- package/bin/context-budget-audit.cjs +0 -447
- package/bin/context-meter-config.cjs +0 -101
- package/bin/context-meter-config.test.cjs +0 -101
- package/bin/event-stream.cjs +0 -205
- package/bin/gsd-t-benchmark-orchestrator.js +0 -437
- package/bin/gsd-t-capture-lint.cjs +0 -440
- package/bin/gsd-t-economics.cjs +0 -315
- package/bin/gsd-t-in-session-usage.cjs +0 -213
- package/bin/gsd-t-orchestrator-config.cjs +0 -161
- package/bin/gsd-t-orchestrator-queue.cjs +0 -180
- package/bin/gsd-t-orchestrator-recover.cjs +0 -231
- package/bin/gsd-t-orchestrator-worker.cjs +0 -219
- package/bin/gsd-t-orchestrator.js +0 -535
- package/bin/gsd-t-parallel-probe.cjs +0 -132
- package/bin/gsd-t-ratelimit-probe-worker.cjs +0 -237
- package/bin/gsd-t-ratelimit-probe.cjs +0 -648
- package/bin/gsd-t-report-tokens.cjs +0 -549
- package/bin/gsd-t-stream-feed-client.cjs +0 -151
- package/bin/gsd-t-token-backfill.cjs +0 -366
- package/bin/gsd-t-token-capture.cjs +0 -321
- package/bin/gsd-t-token-dashboard.cjs +0 -353
- package/bin/gsd-t-token-regenerate-log.cjs +0 -129
- package/bin/gsd-t-tool-attribution.cjs +0 -377
- package/bin/gsd-t-tool-cost.cjs +0 -195
- package/bin/gsd-t-transcript-tee.cjs +0 -246
- package/bin/gsd-t-unattended-heartbeat.cjs +0 -188
- package/bin/gsd-t-unattended-platform.cjs +0 -551
- package/bin/gsd-t-unattended-platform.js +0 -381
- package/bin/gsd-t-unattended-safety.cjs +0 -773
- package/bin/gsd-t-unattended-safety.js +0 -788
- package/bin/gsd-t-unattended.cjs +0 -2109
- package/bin/gsd-t-unattended.js +0 -1367
- package/bin/gsd-t-worker-dispatch.cjs +0 -211
- package/bin/handoff-lock.cjs +0 -249
- package/bin/handoff-lock.js +0 -249
- package/bin/headless-auto-spawn.cjs +0 -711
- package/bin/headless-auto-spawn.js +0 -373
- package/bin/headless-exit-codes.cjs +0 -67
- package/bin/live-activity-report.cjs +0 -615
- package/bin/log-tail.cjs +0 -81
- package/bin/m44-proof-measure.cjs +0 -285
- package/bin/m46-iter-proof.cjs +0 -149
- package/bin/m46-worker-proof.cjs +0 -201
- package/bin/m55-substrate-proof.cjs +0 -134
- package/bin/metrics-rollup.js +0 -200
- package/bin/model-windows.cjs +0 -99
- package/bin/model-windows.test.cjs +0 -75
- package/bin/parallelism-report.cjs +0 -535
- package/bin/runway-estimator.cjs +0 -242
- package/bin/spawn-plan-derive.cjs +0 -163
- package/bin/spawn-plan-status-updater.cjs +0 -292
- package/bin/spawn-plan-writer.cjs +0 -204
- package/bin/supervisor-pid-fingerprint.cjs +0 -126
- package/bin/token-budget.cjs +0 -265
- package/bin/unattended-watch-format.cjs +0 -178
- package/bin/watch-progress.js +0 -155
- package/commands/gsd-t-unattended-stop.md +0 -85
- package/commands/gsd-t-unattended-watch.md +0 -465
- package/commands/gsd-t-unattended.md +0 -476
- package/commands/gsd-t-visualize.md +0 -116
- package/scripts/gsd-t-context-meter.e2e.test.js +0 -364
- package/scripts/gsd-t-context-meter.js +0 -341
- package/scripts/gsd-t-context-meter.test.js +0 -471
- package/scripts/gsd-t-dashboard-server.js +0 -1203
- package/scripts/gsd-t-post-commit-spawn-plan.sh +0 -86
- package/scripts/gsd-t-transcript.html +0 -1821
- package/scripts/hooks/gsd-t-conversation-capture.js +0 -439
- package/scripts/hooks/gsd-t-in-session-usage-hook.js +0 -84
- package/scripts/spawn-plan-fmt-tokens.cjs +0 -80
- package/templates/hooks/post-commit-spawn-plan.sh +0 -85
|
@@ -1,471 +0,0 @@
|
|
|
1
|
-
"use strict";
|
|
2
|
-
|
|
3
|
-
/**
|
|
4
|
-
* Unit tests for scripts/gsd-t-context-meter.js
|
|
5
|
-
*
|
|
6
|
-
* v3.12 (M38 meter reduction): single-band (normal/threshold) — rewritten
|
|
7
|
-
* from the M34 three-band / dead-meter era. The hook uses local token
|
|
8
|
-
* estimation (no API key, no network) since v3.12, so the old countTokens
|
|
9
|
-
* seam and missing-key tests are retired. The 7 previously stranded TD-102
|
|
10
|
-
* tests are now rewritten against the v1.3.0 invariants.
|
|
11
|
-
*
|
|
12
|
-
* All dependencies are injected via runMeter's test seams — no real file
|
|
13
|
-
* or transcript reads are needed.
|
|
14
|
-
*/
|
|
15
|
-
|
|
16
|
-
const { test, beforeEach, afterEach } = require("node:test");
|
|
17
|
-
const assert = require("node:assert/strict");
|
|
18
|
-
const fs = require("fs");
|
|
19
|
-
const path = require("path");
|
|
20
|
-
const os = require("os");
|
|
21
|
-
|
|
22
|
-
const { runMeter, defaultState } = require("./gsd-t-context-meter");
|
|
23
|
-
|
|
24
|
-
/* ─────────────────────────── test harness ─────────────────────────── */
|
|
25
|
-
|
|
26
|
-
let tmpRoot;
|
|
27
|
-
|
|
28
|
-
beforeEach(() => {
|
|
29
|
-
tmpRoot = fs.mkdtempSync(path.join(os.tmpdir(), "gsd-t-cm-hook-"));
|
|
30
|
-
});
|
|
31
|
-
|
|
32
|
-
afterEach(() => {
|
|
33
|
-
try {
|
|
34
|
-
fs.rmSync(tmpRoot, { recursive: true, force: true });
|
|
35
|
-
} catch (_) {
|
|
36
|
-
/* ignore */
|
|
37
|
-
}
|
|
38
|
-
});
|
|
39
|
-
|
|
40
|
-
function makeConfig(overrides = {}) {
|
|
41
|
-
return {
|
|
42
|
-
version: 1,
|
|
43
|
-
thresholdPct: 75,
|
|
44
|
-
modelWindowSize: 200000,
|
|
45
|
-
checkFrequency: 5,
|
|
46
|
-
statePath: ".gsd-t/.context-meter-state.json",
|
|
47
|
-
logPath: ".gsd-t/context-meter.log",
|
|
48
|
-
timeoutMs: 2000,
|
|
49
|
-
...overrides,
|
|
50
|
-
};
|
|
51
|
-
}
|
|
52
|
-
|
|
53
|
-
function stateFile(root) {
|
|
54
|
-
return path.join(root, ".gsd-t", ".context-meter-state.json");
|
|
55
|
-
}
|
|
56
|
-
|
|
57
|
-
function logFile(root) {
|
|
58
|
-
return path.join(root, ".gsd-t", "context-meter.log");
|
|
59
|
-
}
|
|
60
|
-
|
|
61
|
-
function seedState(root, partial) {
|
|
62
|
-
const statePath = stateFile(root);
|
|
63
|
-
fs.mkdirSync(path.dirname(statePath), { recursive: true });
|
|
64
|
-
fs.writeFileSync(
|
|
65
|
-
statePath,
|
|
66
|
-
JSON.stringify({ ...defaultState(), ...partial }, null, 2)
|
|
67
|
-
);
|
|
68
|
-
}
|
|
69
|
-
|
|
70
|
-
function makePayload() {
|
|
71
|
-
return {
|
|
72
|
-
session_id: "test-session",
|
|
73
|
-
transcript_path: path.join(tmpRoot, "fake-transcript.jsonl"),
|
|
74
|
-
tool_name: "Bash",
|
|
75
|
-
tool_input: {},
|
|
76
|
-
tool_response: {},
|
|
77
|
-
};
|
|
78
|
-
}
|
|
79
|
-
|
|
80
|
-
const FAKE_PARSED = {
|
|
81
|
-
system: "",
|
|
82
|
-
messages: [{ role: "user", content: [{ type: "text", text: "hi" }] }],
|
|
83
|
-
};
|
|
84
|
-
|
|
85
|
-
/* ───────────────────────────── tests ───────────────────────────── */
|
|
86
|
-
|
|
87
|
-
test("1. check-frequency skip — estimator NOT called, counter incremented, stdout {}", async () => {
|
|
88
|
-
seedState(tmpRoot, { checkCount: 3 });
|
|
89
|
-
|
|
90
|
-
const estimateCalls = [];
|
|
91
|
-
const out = await runMeter({
|
|
92
|
-
payload: makePayload(),
|
|
93
|
-
projectRoot: tmpRoot,
|
|
94
|
-
_loadConfig: () => makeConfig({ checkFrequency: 5 }),
|
|
95
|
-
_parseTranscript: async () => {
|
|
96
|
-
throw new Error("parseTranscript should not be called on skip");
|
|
97
|
-
},
|
|
98
|
-
_estimateTokens: () => {
|
|
99
|
-
estimateCalls.push("called");
|
|
100
|
-
throw new Error("estimateTokens should not be called on skip");
|
|
101
|
-
},
|
|
102
|
-
});
|
|
103
|
-
|
|
104
|
-
assert.deepEqual(out, {});
|
|
105
|
-
assert.equal(estimateCalls.length, 0);
|
|
106
|
-
const state = JSON.parse(fs.readFileSync(stateFile(tmpRoot), "utf8"));
|
|
107
|
-
assert.equal(state.checkCount, 4);
|
|
108
|
-
assert.equal(state.lastError, null);
|
|
109
|
-
});
|
|
110
|
-
|
|
111
|
-
test("2. check-frequency hit — under threshold → {} + state updated (single-band normal)", async () => {
|
|
112
|
-
seedState(tmpRoot, { checkCount: 4 });
|
|
113
|
-
|
|
114
|
-
const out = await runMeter({
|
|
115
|
-
payload: makePayload(),
|
|
116
|
-
projectRoot: tmpRoot,
|
|
117
|
-
_loadConfig: () => makeConfig(),
|
|
118
|
-
_parseTranscript: async () => FAKE_PARSED,
|
|
119
|
-
_estimateTokens: () => ({ inputTokens: 10000 }),
|
|
120
|
-
});
|
|
121
|
-
|
|
122
|
-
assert.deepEqual(out, {});
|
|
123
|
-
const state = JSON.parse(fs.readFileSync(stateFile(tmpRoot), "utf8"));
|
|
124
|
-
assert.equal(state.checkCount, 5);
|
|
125
|
-
assert.equal(state.inputTokens, 10000);
|
|
126
|
-
assert.equal(state.pct, 5);
|
|
127
|
-
assert.equal(state.threshold, "normal");
|
|
128
|
-
assert.equal(state.lastError, null);
|
|
129
|
-
assert.equal(state.modelWindowSize, 200000);
|
|
130
|
-
});
|
|
131
|
-
|
|
132
|
-
test("3. check-frequency hit — at/over threshold → silent marker emitted, band='threshold'", async () => {
|
|
133
|
-
seedState(tmpRoot, { checkCount: 4 });
|
|
134
|
-
|
|
135
|
-
const out = await runMeter({
|
|
136
|
-
payload: makePayload(),
|
|
137
|
-
projectRoot: tmpRoot,
|
|
138
|
-
_loadConfig: () => makeConfig(),
|
|
139
|
-
_parseTranscript: async () => FAKE_PARSED,
|
|
140
|
-
_estimateTokens: () => ({ inputTokens: 160000 }), // 80% > 75%
|
|
141
|
-
});
|
|
142
|
-
|
|
143
|
-
// v3.12: additionalContext is a SHORT machine-readable marker, not a STOP banner.
|
|
144
|
-
assert.equal(typeof out.additionalContext, "string");
|
|
145
|
-
assert.equal(out.additionalContext, "next-spawn-headless:true");
|
|
146
|
-
// No user-facing banner strings.
|
|
147
|
-
assert.ok(!/MANDATORY STOP/.test(out.additionalContext));
|
|
148
|
-
assert.ok(!/\/gsd-t-pause/.test(out.additionalContext));
|
|
149
|
-
|
|
150
|
-
const state = JSON.parse(fs.readFileSync(stateFile(tmpRoot), "utf8"));
|
|
151
|
-
assert.equal(state.checkCount, 5);
|
|
152
|
-
assert.equal(state.pct, 80);
|
|
153
|
-
assert.equal(state.threshold, "threshold");
|
|
154
|
-
assert.equal(state.inputTokens, 160000);
|
|
155
|
-
});
|
|
156
|
-
|
|
157
|
-
test("4. no API key required — local estimator runs without any env setup", async () => {
|
|
158
|
-
// v3.12: estimation is local (no API), so absence of any API key must not
|
|
159
|
-
// affect behavior. This is the inverse of the old "missing_key" test.
|
|
160
|
-
seedState(tmpRoot, { checkCount: 4 });
|
|
161
|
-
|
|
162
|
-
const out = await runMeter({
|
|
163
|
-
payload: makePayload(),
|
|
164
|
-
projectRoot: tmpRoot,
|
|
165
|
-
env: {}, // empty env — no keys of any kind
|
|
166
|
-
_loadConfig: () => makeConfig(),
|
|
167
|
-
_parseTranscript: async () => FAKE_PARSED,
|
|
168
|
-
_estimateTokens: () => ({ inputTokens: 1000 }),
|
|
169
|
-
});
|
|
170
|
-
|
|
171
|
-
assert.deepEqual(out, {});
|
|
172
|
-
const state = JSON.parse(fs.readFileSync(stateFile(tmpRoot), "utf8"));
|
|
173
|
-
assert.equal(state.checkCount, 5);
|
|
174
|
-
assert.equal(state.lastError, null);
|
|
175
|
-
assert.equal(state.inputTokens, 1000);
|
|
176
|
-
// Log must never reference missing_key or API keys.
|
|
177
|
-
if (fs.existsSync(logFile(tmpRoot))) {
|
|
178
|
-
const log = fs.readFileSync(logFile(tmpRoot), "utf8");
|
|
179
|
-
assert.ok(!/missing_key/.test(log));
|
|
180
|
-
assert.ok(!/ANTHROPIC_API_KEY/.test(log));
|
|
181
|
-
}
|
|
182
|
-
});
|
|
183
|
-
|
|
184
|
-
test("5. transcript parse failure — returns null → lastError 'parse_failure'", async () => {
|
|
185
|
-
seedState(tmpRoot, { checkCount: 4 });
|
|
186
|
-
|
|
187
|
-
const out = await runMeter({
|
|
188
|
-
payload: makePayload(),
|
|
189
|
-
projectRoot: tmpRoot,
|
|
190
|
-
_loadConfig: () => makeConfig(),
|
|
191
|
-
_parseTranscript: async () => null,
|
|
192
|
-
_estimateTokens: () => {
|
|
193
|
-
throw new Error("should not estimate when parse fails");
|
|
194
|
-
},
|
|
195
|
-
});
|
|
196
|
-
|
|
197
|
-
assert.deepEqual(out, {});
|
|
198
|
-
const state = JSON.parse(fs.readFileSync(stateFile(tmpRoot), "utf8"));
|
|
199
|
-
assert.equal(state.checkCount, 5);
|
|
200
|
-
assert.equal(state.lastError.code, "parse_failure");
|
|
201
|
-
});
|
|
202
|
-
|
|
203
|
-
test("6. estimator failure — returns null → lastError 'estimate_error', inputTokens reset", async () => {
|
|
204
|
-
seedState(tmpRoot, { checkCount: 4, inputTokens: 99999 });
|
|
205
|
-
|
|
206
|
-
const out = await runMeter({
|
|
207
|
-
payload: makePayload(),
|
|
208
|
-
projectRoot: tmpRoot,
|
|
209
|
-
_loadConfig: () => makeConfig(),
|
|
210
|
-
_parseTranscript: async () => FAKE_PARSED,
|
|
211
|
-
_estimateTokens: () => null,
|
|
212
|
-
});
|
|
213
|
-
|
|
214
|
-
assert.deepEqual(out, {});
|
|
215
|
-
const state = JSON.parse(fs.readFileSync(stateFile(tmpRoot), "utf8"));
|
|
216
|
-
assert.equal(state.lastError.code, "estimate_error");
|
|
217
|
-
// Reset inputTokens to 0 on failure to avoid stale readings tripping threshold false-positives.
|
|
218
|
-
assert.equal(state.inputTokens, 0);
|
|
219
|
-
assert.equal(state.pct, 0);
|
|
220
|
-
assert.equal(state.threshold, "normal");
|
|
221
|
-
});
|
|
222
|
-
|
|
223
|
-
test("7. state file corruption — overwritten with valid defaults + fresh count", async () => {
|
|
224
|
-
const sp = stateFile(tmpRoot);
|
|
225
|
-
fs.mkdirSync(path.dirname(sp), { recursive: true });
|
|
226
|
-
fs.writeFileSync(sp, "not json{");
|
|
227
|
-
|
|
228
|
-
const out = await runMeter({
|
|
229
|
-
payload: makePayload(),
|
|
230
|
-
projectRoot: tmpRoot,
|
|
231
|
-
_loadConfig: () => makeConfig({ checkFrequency: 5 }),
|
|
232
|
-
_parseTranscript: async () => FAKE_PARSED,
|
|
233
|
-
_estimateTokens: () => ({ inputTokens: 100 }),
|
|
234
|
-
});
|
|
235
|
-
|
|
236
|
-
assert.deepEqual(out, {});
|
|
237
|
-
const state = JSON.parse(fs.readFileSync(sp, "utf8"));
|
|
238
|
-
assert.equal(state.version, 1);
|
|
239
|
-
assert.equal(state.checkCount, 1);
|
|
240
|
-
});
|
|
241
|
-
|
|
242
|
-
test("7b. state file corruption + frequency hit — estimator called once, state valid", async () => {
|
|
243
|
-
const sp = stateFile(tmpRoot);
|
|
244
|
-
fs.mkdirSync(path.dirname(sp), { recursive: true });
|
|
245
|
-
fs.writeFileSync(sp, "not json{");
|
|
246
|
-
|
|
247
|
-
const out = await runMeter({
|
|
248
|
-
payload: makePayload(),
|
|
249
|
-
projectRoot: tmpRoot,
|
|
250
|
-
_loadConfig: () => makeConfig({ checkFrequency: 1 }),
|
|
251
|
-
_parseTranscript: async () => FAKE_PARSED,
|
|
252
|
-
_estimateTokens: () => ({ inputTokens: 500 }),
|
|
253
|
-
});
|
|
254
|
-
|
|
255
|
-
assert.deepEqual(out, {});
|
|
256
|
-
const state = JSON.parse(fs.readFileSync(sp, "utf8"));
|
|
257
|
-
assert.equal(state.version, 1);
|
|
258
|
-
assert.equal(state.checkCount, 1);
|
|
259
|
-
assert.equal(state.inputTokens, 500);
|
|
260
|
-
assert.equal(state.lastError, null);
|
|
261
|
-
});
|
|
262
|
-
|
|
263
|
-
test("8. missing transcript_path in payload — lastError 'no_transcript', counter increments", async () => {
|
|
264
|
-
seedState(tmpRoot, { checkCount: 4 });
|
|
265
|
-
|
|
266
|
-
const out = await runMeter({
|
|
267
|
-
payload: { session_id: "x" },
|
|
268
|
-
projectRoot: tmpRoot,
|
|
269
|
-
_loadConfig: () => makeConfig(),
|
|
270
|
-
_parseTranscript: async () => {
|
|
271
|
-
throw new Error("should not parse when transcript_path missing");
|
|
272
|
-
},
|
|
273
|
-
_estimateTokens: () => {
|
|
274
|
-
throw new Error("should not estimate when transcript_path missing");
|
|
275
|
-
},
|
|
276
|
-
});
|
|
277
|
-
|
|
278
|
-
assert.deepEqual(out, {});
|
|
279
|
-
const state = JSON.parse(fs.readFileSync(stateFile(tmpRoot), "utf8"));
|
|
280
|
-
assert.equal(state.checkCount, 5);
|
|
281
|
-
assert.equal(state.lastError.code, "no_transcript");
|
|
282
|
-
});
|
|
283
|
-
|
|
284
|
-
test("9. atomic write — no .tmp file on disk after successful run", async () => {
|
|
285
|
-
seedState(tmpRoot, { checkCount: 4 });
|
|
286
|
-
|
|
287
|
-
await runMeter({
|
|
288
|
-
payload: makePayload(),
|
|
289
|
-
projectRoot: tmpRoot,
|
|
290
|
-
_loadConfig: () => makeConfig(),
|
|
291
|
-
_parseTranscript: async () => FAKE_PARSED,
|
|
292
|
-
_estimateTokens: () => ({ inputTokens: 1000 }),
|
|
293
|
-
});
|
|
294
|
-
|
|
295
|
-
const tmp = stateFile(tmpRoot) + ".tmp";
|
|
296
|
-
assert.equal(fs.existsSync(tmp), false, ".tmp file should not exist after rename");
|
|
297
|
-
assert.equal(fs.existsSync(stateFile(tmpRoot)), true, "state file should exist");
|
|
298
|
-
});
|
|
299
|
-
|
|
300
|
-
test("10. fail-open on unexpected throw — loadConfig throws → runMeter returns {}", async () => {
|
|
301
|
-
const out = await runMeter({
|
|
302
|
-
payload: makePayload(),
|
|
303
|
-
projectRoot: tmpRoot,
|
|
304
|
-
_loadConfig: () => {
|
|
305
|
-
throw new Error("boom");
|
|
306
|
-
},
|
|
307
|
-
_parseTranscript: async () => FAKE_PARSED,
|
|
308
|
-
_estimateTokens: () => ({ inputTokens: 1 }),
|
|
309
|
-
});
|
|
310
|
-
|
|
311
|
-
assert.deepEqual(out, {});
|
|
312
|
-
});
|
|
313
|
-
|
|
314
|
-
test("10b. fail-open — parseTranscript throws synchronously → {}", async () => {
|
|
315
|
-
seedState(tmpRoot, { checkCount: 4 });
|
|
316
|
-
const out = await runMeter({
|
|
317
|
-
payload: makePayload(),
|
|
318
|
-
projectRoot: tmpRoot,
|
|
319
|
-
_loadConfig: () => makeConfig(),
|
|
320
|
-
_parseTranscript: () => {
|
|
321
|
-
throw new Error("sync boom");
|
|
322
|
-
},
|
|
323
|
-
_estimateTokens: () => ({ inputTokens: 1 }),
|
|
324
|
-
});
|
|
325
|
-
|
|
326
|
-
assert.deepEqual(out, {});
|
|
327
|
-
const state = JSON.parse(fs.readFileSync(stateFile(tmpRoot), "utf8"));
|
|
328
|
-
assert.equal(state.lastError.code, "parse_failure");
|
|
329
|
-
});
|
|
330
|
-
|
|
331
|
-
test("10c. fail-open — estimateTokens throws → {} + lastError='estimate_error'", async () => {
|
|
332
|
-
seedState(tmpRoot, { checkCount: 4 });
|
|
333
|
-
const out = await runMeter({
|
|
334
|
-
payload: makePayload(),
|
|
335
|
-
projectRoot: tmpRoot,
|
|
336
|
-
_loadConfig: () => makeConfig(),
|
|
337
|
-
_parseTranscript: async () => FAKE_PARSED,
|
|
338
|
-
_estimateTokens: () => {
|
|
339
|
-
throw new Error("sync boom");
|
|
340
|
-
},
|
|
341
|
-
});
|
|
342
|
-
|
|
343
|
-
assert.deepEqual(out, {});
|
|
344
|
-
const state = JSON.parse(fs.readFileSync(stateFile(tmpRoot), "utf8"));
|
|
345
|
-
assert.equal(state.lastError.code, "estimate_error");
|
|
346
|
-
});
|
|
347
|
-
|
|
348
|
-
test("11. log never contains message content — only categories/counts", async () => {
|
|
349
|
-
seedState(tmpRoot, { checkCount: 4 });
|
|
350
|
-
const secretText = "SECRET_MESSAGE_CONTENT_XYZ";
|
|
351
|
-
|
|
352
|
-
await runMeter({
|
|
353
|
-
payload: makePayload(),
|
|
354
|
-
projectRoot: tmpRoot,
|
|
355
|
-
_loadConfig: () => makeConfig(),
|
|
356
|
-
_parseTranscript: async () => ({
|
|
357
|
-
system: "",
|
|
358
|
-
messages: [
|
|
359
|
-
{ role: "user", content: [{ type: "text", text: secretText }] },
|
|
360
|
-
],
|
|
361
|
-
}),
|
|
362
|
-
_estimateTokens: () => ({ inputTokens: 42 }),
|
|
363
|
-
});
|
|
364
|
-
|
|
365
|
-
const log = fs.readFileSync(logFile(tmpRoot), "utf8");
|
|
366
|
-
assert.ok(!log.includes(secretText), "log must not contain message content");
|
|
367
|
-
assert.match(log, /measure/);
|
|
368
|
-
assert.match(log, /tokens=42/);
|
|
369
|
-
});
|
|
370
|
-
|
|
371
|
-
test("12. clock injection — timestamp uses injected clock", async () => {
|
|
372
|
-
seedState(tmpRoot, { checkCount: 4 });
|
|
373
|
-
const fixed = new Date("2026-04-16T18:00:00.000Z");
|
|
374
|
-
|
|
375
|
-
await runMeter({
|
|
376
|
-
payload: makePayload(),
|
|
377
|
-
projectRoot: tmpRoot,
|
|
378
|
-
clock: () => fixed,
|
|
379
|
-
_loadConfig: () => makeConfig(),
|
|
380
|
-
_parseTranscript: async () => FAKE_PARSED,
|
|
381
|
-
_estimateTokens: () => ({ inputTokens: 1000 }),
|
|
382
|
-
});
|
|
383
|
-
|
|
384
|
-
const state = JSON.parse(fs.readFileSync(stateFile(tmpRoot), "utf8"));
|
|
385
|
-
assert.equal(state.timestamp, fixed.toISOString());
|
|
386
|
-
});
|
|
387
|
-
|
|
388
|
-
/* ── M-fix: model-aware context window (the reported regression) ───────── */
|
|
389
|
-
|
|
390
|
-
test("13. Opus 4.7 @ ~36% of a 1M window stays 'normal' (regression repro)", async () => {
|
|
391
|
-
// The exact reported symptom: ~360k tokens used on an Opus 4.7 session.
|
|
392
|
-
// With the old hardcoded 200k window this computed 180% → premature
|
|
393
|
-
// headless handoff at ~64% of context REMAINING. With model-aware sizing
|
|
394
|
-
// the window is 1M, so 360k = 36% = normal, no handoff.
|
|
395
|
-
seedState(tmpRoot, { checkCount: 4 });
|
|
396
|
-
|
|
397
|
-
const out = await runMeter({
|
|
398
|
-
payload: makePayload(),
|
|
399
|
-
projectRoot: tmpRoot,
|
|
400
|
-
_loadConfig: () => makeConfig(), // config still says 200k — must be overridden
|
|
401
|
-
_parseTranscript: async () => ({ ...FAKE_PARSED, model: "claude-opus-4-7" }),
|
|
402
|
-
_estimateTokens: () => ({ inputTokens: 360000 }),
|
|
403
|
-
});
|
|
404
|
-
|
|
405
|
-
// No handoff marker — this is the whole point of the fix.
|
|
406
|
-
assert.deepEqual(out, {});
|
|
407
|
-
|
|
408
|
-
const state = JSON.parse(fs.readFileSync(stateFile(tmpRoot), "utf8"));
|
|
409
|
-
assert.equal(state.modelWindowSize, 1_000_000, "window resolved from model, not config");
|
|
410
|
-
assert.equal(state.pct, 36, "360k / 1M = 36%");
|
|
411
|
-
assert.equal(state.threshold, "normal");
|
|
412
|
-
});
|
|
413
|
-
|
|
414
|
-
test("14. Opus 4.7 @ 80% of the true 1M window DOES hand off", async () => {
|
|
415
|
-
// The handoff must still fire at the real 75% threshold against the
|
|
416
|
-
// corrected window — we keep the guard, we just size it correctly.
|
|
417
|
-
seedState(tmpRoot, { checkCount: 4 });
|
|
418
|
-
|
|
419
|
-
const out = await runMeter({
|
|
420
|
-
payload: makePayload(),
|
|
421
|
-
projectRoot: tmpRoot,
|
|
422
|
-
_loadConfig: () => makeConfig(),
|
|
423
|
-
_parseTranscript: async () => ({ ...FAKE_PARSED, model: "claude-opus-4-7-20260115" }),
|
|
424
|
-
_estimateTokens: () => ({ inputTokens: 800000 }), // 80% of 1M > 75%
|
|
425
|
-
});
|
|
426
|
-
|
|
427
|
-
assert.equal(out.additionalContext, "next-spawn-headless:true");
|
|
428
|
-
const state = JSON.parse(fs.readFileSync(stateFile(tmpRoot), "utf8"));
|
|
429
|
-
assert.equal(state.modelWindowSize, 1_000_000);
|
|
430
|
-
assert.equal(state.pct, 80);
|
|
431
|
-
assert.equal(state.threshold, "threshold");
|
|
432
|
-
});
|
|
433
|
-
|
|
434
|
-
test("15. no model in transcript → falls back to config window (back-compat)", async () => {
|
|
435
|
-
// Existing transcripts / stubs without a model field must behave exactly
|
|
436
|
-
// as before: config's modelWindowSize governs.
|
|
437
|
-
seedState(tmpRoot, { checkCount: 4 });
|
|
438
|
-
|
|
439
|
-
const out = await runMeter({
|
|
440
|
-
payload: makePayload(),
|
|
441
|
-
projectRoot: tmpRoot,
|
|
442
|
-
_loadConfig: () => makeConfig({ modelWindowSize: 200000 }),
|
|
443
|
-
_parseTranscript: async () => FAKE_PARSED, // no `model` key
|
|
444
|
-
_estimateTokens: () => ({ inputTokens: 160000 }), // 80% of 200k
|
|
445
|
-
});
|
|
446
|
-
|
|
447
|
-
assert.equal(out.additionalContext, "next-spawn-headless:true");
|
|
448
|
-
const state = JSON.parse(fs.readFileSync(stateFile(tmpRoot), "utf8"));
|
|
449
|
-
assert.equal(state.modelWindowSize, 200000);
|
|
450
|
-
assert.equal(state.pct, 80);
|
|
451
|
-
});
|
|
452
|
-
|
|
453
|
-
test("16. Haiku session correctly sized at 200k (not over-large 1M)", async () => {
|
|
454
|
-
seedState(tmpRoot, { checkCount: 4 });
|
|
455
|
-
|
|
456
|
-
const out = await runMeter({
|
|
457
|
-
payload: makePayload(),
|
|
458
|
-
projectRoot: tmpRoot,
|
|
459
|
-
_loadConfig: () => makeConfig(),
|
|
460
|
-
_parseTranscript: async () => ({
|
|
461
|
-
...FAKE_PARSED,
|
|
462
|
-
model: "claude-haiku-4-5-20251001",
|
|
463
|
-
}),
|
|
464
|
-
_estimateTokens: () => ({ inputTokens: 170000 }), // 85% of 200k
|
|
465
|
-
});
|
|
466
|
-
|
|
467
|
-
assert.equal(out.additionalContext, "next-spawn-headless:true");
|
|
468
|
-
const state = JSON.parse(fs.readFileSync(stateFile(tmpRoot), "utf8"));
|
|
469
|
-
assert.equal(state.modelWindowSize, 200000);
|
|
470
|
-
assert.equal(state.pct, 85);
|
|
471
|
-
});
|