@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
|
@@ -1,364 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* gsd-t-context-meter.e2e.test.js — black-box E2E integration test for M34.
|
|
3
|
-
*
|
|
4
|
-
* TEST-ONLY FILE. Not shipped to users. Does not participate in production
|
|
5
|
-
* require graphs. Spawned as part of `node --test` only.
|
|
6
|
-
*
|
|
7
|
-
* Exercises the real child-process hook as Claude Code would invoke it:
|
|
8
|
-
*
|
|
9
|
-
* 1. A temporary project root is constructed under os.tmpdir() containing:
|
|
10
|
-
* - .gsd-t/context-meter-config.json (real config loader target)
|
|
11
|
-
* - transcript.jsonl (minimal Claude-Code-shaped transcript)
|
|
12
|
-
* 2. `node scripts/gsd-t-context-meter.js` is spawned as a child process
|
|
13
|
-
* with cwd = tempdir.
|
|
14
|
-
* 3. We write the PostToolUse JSON payload to the child's stdin, close
|
|
15
|
-
* stdin, collect stdout, and assert both the stdout shape and the
|
|
16
|
-
* on-disk state file.
|
|
17
|
-
*
|
|
18
|
-
* Since v3.12 the context meter uses local token estimation (no API call),
|
|
19
|
-
* so no stub HTTP server is needed. The transcript content determines the
|
|
20
|
-
* estimated token count via chars/3.5 heuristic.
|
|
21
|
-
*
|
|
22
|
-
* @module scripts/gsd-t-context-meter.e2e.test
|
|
23
|
-
*/
|
|
24
|
-
|
|
25
|
-
"use strict";
|
|
26
|
-
|
|
27
|
-
const { test, beforeEach, afterEach } = require("node:test");
|
|
28
|
-
const assert = require("node:assert/strict");
|
|
29
|
-
const { spawn } = require("node:child_process");
|
|
30
|
-
const fs = require("node:fs");
|
|
31
|
-
const path = require("node:path");
|
|
32
|
-
const os = require("node:os");
|
|
33
|
-
|
|
34
|
-
const HOOK_SCRIPT = path.resolve(__dirname, "gsd-t-context-meter.js");
|
|
35
|
-
const HARD_TIMEOUT_MS = 12000;
|
|
36
|
-
|
|
37
|
-
/* ──────────────────────────── test fixtures ──────────────────────────── */
|
|
38
|
-
|
|
39
|
-
class Sandbox {
|
|
40
|
-
constructor() {
|
|
41
|
-
this.tempdir = null;
|
|
42
|
-
this.childProcs = [];
|
|
43
|
-
}
|
|
44
|
-
|
|
45
|
-
async setup() {
|
|
46
|
-
this.tempdir = fs.mkdtempSync(path.join(os.tmpdir(), "gsd-t-cm-e2e-"));
|
|
47
|
-
fs.mkdirSync(path.join(this.tempdir, ".gsd-t"), { recursive: true });
|
|
48
|
-
}
|
|
49
|
-
|
|
50
|
-
writeConfig(config) {
|
|
51
|
-
const full = Object.assign(
|
|
52
|
-
{
|
|
53
|
-
version: 1,
|
|
54
|
-
thresholdPct: 75,
|
|
55
|
-
modelWindowSize: 200000,
|
|
56
|
-
checkFrequency: 1,
|
|
57
|
-
apiKeyEnvVar: "ANTHROPIC_API_KEY",
|
|
58
|
-
statePath: ".gsd-t/.context-meter-state.json",
|
|
59
|
-
logPath: ".gsd-t/context-meter.log",
|
|
60
|
-
timeoutMs: 2000,
|
|
61
|
-
},
|
|
62
|
-
config || {}
|
|
63
|
-
);
|
|
64
|
-
fs.writeFileSync(
|
|
65
|
-
path.join(this.tempdir, ".gsd-t", "context-meter-config.json"),
|
|
66
|
-
JSON.stringify(full, null, 2),
|
|
67
|
-
"utf8"
|
|
68
|
-
);
|
|
69
|
-
return full;
|
|
70
|
-
}
|
|
71
|
-
|
|
72
|
-
/**
|
|
73
|
-
* Write a Claude-Code transcript JSONL with configurable content size.
|
|
74
|
-
* The charCount parameter controls how many characters of text content
|
|
75
|
-
* are in the transcript, which determines the estimated token count.
|
|
76
|
-
*/
|
|
77
|
-
writeTranscript(filename = "transcript.jsonl", charCount = 100, model = "claude-opus-4-6") {
|
|
78
|
-
const userText = "x".repeat(Math.floor(charCount / 2));
|
|
79
|
-
const assistantText = "y".repeat(Math.ceil(charCount / 2));
|
|
80
|
-
const lines = [
|
|
81
|
-
JSON.stringify({
|
|
82
|
-
type: "user",
|
|
83
|
-
message: { role: "user", content: userText },
|
|
84
|
-
uuid: "u1",
|
|
85
|
-
sessionId: "sess-1",
|
|
86
|
-
}),
|
|
87
|
-
JSON.stringify({
|
|
88
|
-
type: "assistant",
|
|
89
|
-
message: {
|
|
90
|
-
role: "assistant",
|
|
91
|
-
content: [{ type: "text", text: assistantText }],
|
|
92
|
-
model,
|
|
93
|
-
},
|
|
94
|
-
uuid: "a1",
|
|
95
|
-
sessionId: "sess-1",
|
|
96
|
-
}),
|
|
97
|
-
];
|
|
98
|
-
const p = path.join(this.tempdir, filename);
|
|
99
|
-
fs.writeFileSync(p, lines.join("\n") + "\n", "utf8");
|
|
100
|
-
return p;
|
|
101
|
-
}
|
|
102
|
-
|
|
103
|
-
writeState(state) {
|
|
104
|
-
const full = Object.assign(
|
|
105
|
-
{
|
|
106
|
-
version: 1,
|
|
107
|
-
timestamp: null,
|
|
108
|
-
inputTokens: 0,
|
|
109
|
-
modelWindowSize: 0,
|
|
110
|
-
pct: 0,
|
|
111
|
-
threshold: "normal",
|
|
112
|
-
checkCount: 0,
|
|
113
|
-
lastError: null,
|
|
114
|
-
},
|
|
115
|
-
state || {}
|
|
116
|
-
);
|
|
117
|
-
fs.writeFileSync(
|
|
118
|
-
path.join(this.tempdir, ".gsd-t", ".context-meter-state.json"),
|
|
119
|
-
JSON.stringify(full, null, 2),
|
|
120
|
-
"utf8"
|
|
121
|
-
);
|
|
122
|
-
}
|
|
123
|
-
|
|
124
|
-
async runHook({ payload, env }) {
|
|
125
|
-
const fullEnv = Object.assign({}, process.env, {});
|
|
126
|
-
if (env) {
|
|
127
|
-
for (const [k, v] of Object.entries(env)) {
|
|
128
|
-
if (v === null || v === undefined) {
|
|
129
|
-
delete fullEnv[k];
|
|
130
|
-
} else {
|
|
131
|
-
fullEnv[k] = v;
|
|
132
|
-
}
|
|
133
|
-
}
|
|
134
|
-
}
|
|
135
|
-
|
|
136
|
-
const child = spawn(process.execPath, [HOOK_SCRIPT], {
|
|
137
|
-
cwd: this.tempdir,
|
|
138
|
-
env: fullEnv,
|
|
139
|
-
stdio: ["pipe", "pipe", "pipe"],
|
|
140
|
-
});
|
|
141
|
-
this.childProcs.push(child);
|
|
142
|
-
|
|
143
|
-
const stdoutChunks = [];
|
|
144
|
-
const stderrChunks = [];
|
|
145
|
-
child.stdout.on("data", (c) => stdoutChunks.push(c));
|
|
146
|
-
child.stderr.on("data", (c) => stderrChunks.push(c));
|
|
147
|
-
|
|
148
|
-
child.stdin.write(JSON.stringify(payload || {}));
|
|
149
|
-
child.stdin.end();
|
|
150
|
-
|
|
151
|
-
const result = await new Promise((resolve, reject) => {
|
|
152
|
-
const killTimer = setTimeout(() => {
|
|
153
|
-
try {
|
|
154
|
-
child.kill("SIGKILL");
|
|
155
|
-
} catch (_) {
|
|
156
|
-
/* ignore */
|
|
157
|
-
}
|
|
158
|
-
reject(new Error(`hook child process timeout after ${HARD_TIMEOUT_MS}ms`));
|
|
159
|
-
}, HARD_TIMEOUT_MS);
|
|
160
|
-
|
|
161
|
-
child.on("close", (code) => {
|
|
162
|
-
clearTimeout(killTimer);
|
|
163
|
-
resolve({
|
|
164
|
-
stdout: Buffer.concat(stdoutChunks).toString("utf8"),
|
|
165
|
-
stderr: Buffer.concat(stderrChunks).toString("utf8"),
|
|
166
|
-
code,
|
|
167
|
-
});
|
|
168
|
-
});
|
|
169
|
-
child.on("error", (err) => {
|
|
170
|
-
clearTimeout(killTimer);
|
|
171
|
-
reject(err);
|
|
172
|
-
});
|
|
173
|
-
});
|
|
174
|
-
|
|
175
|
-
return result;
|
|
176
|
-
}
|
|
177
|
-
|
|
178
|
-
readState() {
|
|
179
|
-
const p = path.join(this.tempdir, ".gsd-t", ".context-meter-state.json");
|
|
180
|
-
if (!fs.existsSync(p)) return null;
|
|
181
|
-
return JSON.parse(fs.readFileSync(p, "utf8"));
|
|
182
|
-
}
|
|
183
|
-
|
|
184
|
-
stateFileExists() {
|
|
185
|
-
const p = path.join(this.tempdir, ".gsd-t", ".context-meter-state.json");
|
|
186
|
-
return fs.existsSync(p);
|
|
187
|
-
}
|
|
188
|
-
|
|
189
|
-
tmpFileExists() {
|
|
190
|
-
const p = path.join(this.tempdir, ".gsd-t", ".context-meter-state.json.tmp");
|
|
191
|
-
return fs.existsSync(p);
|
|
192
|
-
}
|
|
193
|
-
|
|
194
|
-
async dispose() {
|
|
195
|
-
for (const c of this.childProcs) {
|
|
196
|
-
try {
|
|
197
|
-
if (!c.killed) c.kill("SIGKILL");
|
|
198
|
-
} catch (_) {
|
|
199
|
-
/* ignore */
|
|
200
|
-
}
|
|
201
|
-
}
|
|
202
|
-
this.childProcs = [];
|
|
203
|
-
|
|
204
|
-
if (this.tempdir) {
|
|
205
|
-
try {
|
|
206
|
-
fs.rmSync(this.tempdir, { recursive: true, force: true });
|
|
207
|
-
} catch (_) {
|
|
208
|
-
/* ignore */
|
|
209
|
-
}
|
|
210
|
-
this.tempdir = null;
|
|
211
|
-
}
|
|
212
|
-
}
|
|
213
|
-
}
|
|
214
|
-
|
|
215
|
-
/* ──────────────────────────── shared state ──────────────────────────── */
|
|
216
|
-
|
|
217
|
-
let sandbox;
|
|
218
|
-
|
|
219
|
-
beforeEach(async () => {
|
|
220
|
-
sandbox = new Sandbox();
|
|
221
|
-
await sandbox.setup();
|
|
222
|
-
});
|
|
223
|
-
|
|
224
|
-
afterEach(async () => {
|
|
225
|
-
if (sandbox) {
|
|
226
|
-
await sandbox.dispose();
|
|
227
|
-
sandbox = null;
|
|
228
|
-
}
|
|
229
|
-
});
|
|
230
|
-
|
|
231
|
-
/* ──────────────────────────── tests ──────────────────────────── */
|
|
232
|
-
|
|
233
|
-
test("E2E 1. below threshold — stdout {} and state reflects estimate", async () => {
|
|
234
|
-
// 100 chars → ~29 tokens. The transcript declares claude-opus-4-6, so the
|
|
235
|
-
// EFFECTIVE window is the real 1M (model-aware sizing), not the config's
|
|
236
|
-
// legacy 200K — ~0.003% of 1M, well below threshold.
|
|
237
|
-
sandbox.writeConfig({ thresholdPct: 75, modelWindowSize: 200000, checkFrequency: 1 });
|
|
238
|
-
const transcriptPath = sandbox.writeTranscript("transcript.jsonl", 100);
|
|
239
|
-
|
|
240
|
-
const { stdout, code } = await sandbox.runHook({
|
|
241
|
-
payload: { session_id: "test-below", transcript_path: transcriptPath },
|
|
242
|
-
});
|
|
243
|
-
|
|
244
|
-
assert.equal(code, 0, "hook should always exit 0");
|
|
245
|
-
const parsed = JSON.parse(stdout || "{}");
|
|
246
|
-
assert.deepEqual(parsed, {}, "below-threshold stdout must be exactly {}");
|
|
247
|
-
|
|
248
|
-
const state = sandbox.readState();
|
|
249
|
-
assert.ok(state, "state file should exist");
|
|
250
|
-
assert.equal(state.version, 1);
|
|
251
|
-
assert.ok(state.inputTokens > 0, "should have estimated some tokens");
|
|
252
|
-
assert.ok(state.inputTokens < 1000, "small transcript should estimate < 1K tokens");
|
|
253
|
-
assert.equal(
|
|
254
|
-
state.modelWindowSize,
|
|
255
|
-
1_000_000,
|
|
256
|
-
"window resolved from the transcript's claude-opus-4-6 model (1M), not config 200K"
|
|
257
|
-
);
|
|
258
|
-
assert.ok(state.pct < 1, "pct should be well below threshold");
|
|
259
|
-
assert.equal(state.threshold, "normal");
|
|
260
|
-
assert.equal(state.checkCount, 1);
|
|
261
|
-
assert.equal(state.lastError, null);
|
|
262
|
-
assert.ok(typeof state.timestamp === "string" && state.timestamp.length > 0);
|
|
263
|
-
assert.equal(sandbox.tmpFileExists(), false, "no leftover .tmp file");
|
|
264
|
-
});
|
|
265
|
-
|
|
266
|
-
test("E2E 2. above threshold — stdout additionalContext with large transcript", async () => {
|
|
267
|
-
// Haiku → real 200K window. 600K chars → ~171K tokens → ~85.7% of 200K →
|
|
268
|
-
// threshold band + additionalContext. (Model-aware sizing means we pin a
|
|
269
|
-
// 200K-window model here rather than relying on a stale config default.)
|
|
270
|
-
sandbox.writeConfig({ thresholdPct: 75, modelWindowSize: 200000, checkFrequency: 1 });
|
|
271
|
-
const transcriptPath = sandbox.writeTranscript(
|
|
272
|
-
"transcript.jsonl",
|
|
273
|
-
600000,
|
|
274
|
-
"claude-haiku-4-5-20251001"
|
|
275
|
-
);
|
|
276
|
-
|
|
277
|
-
const { stdout, code } = await sandbox.runHook({
|
|
278
|
-
payload: { session_id: "test-above", transcript_path: transcriptPath },
|
|
279
|
-
});
|
|
280
|
-
|
|
281
|
-
assert.equal(code, 0);
|
|
282
|
-
const parsed = JSON.parse(stdout || "{}");
|
|
283
|
-
assert.ok(parsed.additionalContext, "must emit additionalContext");
|
|
284
|
-
// v3.12 (M38): additionalContext is a short silent marker, not a MANDATORY STOP banner.
|
|
285
|
-
assert.equal(parsed.additionalContext, "next-spawn-headless:true");
|
|
286
|
-
assert.ok(!/MANDATORY STOP/.test(parsed.additionalContext));
|
|
287
|
-
assert.ok(!/\/gsd-t-pause/.test(parsed.additionalContext));
|
|
288
|
-
|
|
289
|
-
const state = sandbox.readState();
|
|
290
|
-
assert.ok(state);
|
|
291
|
-
assert.ok(state.inputTokens > 100000, "large transcript should estimate >100K tokens");
|
|
292
|
-
assert.ok(state.pct > 50, "pct should be above threshold");
|
|
293
|
-
assert.equal(state.checkCount, 1);
|
|
294
|
-
assert.equal(state.lastError, null);
|
|
295
|
-
assert.equal(sandbox.tmpFileExists(), false);
|
|
296
|
-
});
|
|
297
|
-
|
|
298
|
-
test("E2E 2b. REGRESSION — large Opus transcript stays 'normal' on the 1M window", async () => {
|
|
299
|
-
// The reported bug, end-to-end: ~600K chars → ~171K tokens. Under the old
|
|
300
|
-
// hardcoded 200K window this read as ~85% → false headless handoff while
|
|
301
|
-
// ~64% of context REMAINED. With model-aware sizing (claude-opus-4-7 → 1M),
|
|
302
|
-
// 171K is only ~17% → stdout {} → no premature handoff.
|
|
303
|
-
sandbox.writeConfig({ thresholdPct: 75, modelWindowSize: 200000, checkFrequency: 1 });
|
|
304
|
-
const transcriptPath = sandbox.writeTranscript(
|
|
305
|
-
"transcript.jsonl",
|
|
306
|
-
600000,
|
|
307
|
-
"claude-opus-4-7"
|
|
308
|
-
);
|
|
309
|
-
|
|
310
|
-
const { stdout, code } = await sandbox.runHook({
|
|
311
|
-
payload: { session_id: "test-regression", transcript_path: transcriptPath },
|
|
312
|
-
});
|
|
313
|
-
|
|
314
|
-
assert.equal(code, 0);
|
|
315
|
-
const parsed = JSON.parse(stdout || "{}");
|
|
316
|
-
assert.deepEqual(parsed, {}, "must NOT hand off — the reported regression");
|
|
317
|
-
|
|
318
|
-
const state = sandbox.readState();
|
|
319
|
-
assert.equal(state.modelWindowSize, 1_000_000);
|
|
320
|
-
assert.ok(state.inputTokens > 100000, "large transcript, >100K tokens");
|
|
321
|
-
assert.ok(state.pct < 75, `pct ${state.pct} must be below threshold on a 1M window`);
|
|
322
|
-
assert.equal(state.threshold, "normal");
|
|
323
|
-
});
|
|
324
|
-
|
|
325
|
-
test("E2E 3. missing transcript — stdout {}, state has parse error", async () => {
|
|
326
|
-
sandbox.writeConfig({ thresholdPct: 75, checkFrequency: 1 });
|
|
327
|
-
|
|
328
|
-
const { stdout, code } = await sandbox.runHook({
|
|
329
|
-
payload: {
|
|
330
|
-
session_id: "test-nofile",
|
|
331
|
-
transcript_path: path.join(sandbox.tempdir, "nonexistent.jsonl"),
|
|
332
|
-
},
|
|
333
|
-
});
|
|
334
|
-
|
|
335
|
-
assert.equal(code, 0);
|
|
336
|
-
const parsed = JSON.parse(stdout || "{}");
|
|
337
|
-
assert.deepEqual(parsed, {});
|
|
338
|
-
|
|
339
|
-
const state = sandbox.readState();
|
|
340
|
-
assert.ok(state);
|
|
341
|
-
assert.equal(state.checkCount, 1);
|
|
342
|
-
assert.ok(state.lastError && typeof state.lastError === "object");
|
|
343
|
-
assert.equal(state.lastError.code, "parse_failure");
|
|
344
|
-
});
|
|
345
|
-
|
|
346
|
-
test("E2E 4. checkFrequency skip — estimation not run, checkCount increments", async () => {
|
|
347
|
-
sandbox.writeConfig({ thresholdPct: 75, checkFrequency: 5 });
|
|
348
|
-
const transcriptPath = sandbox.writeTranscript("transcript.jsonl", 100);
|
|
349
|
-
sandbox.writeState({ checkCount: 3 });
|
|
350
|
-
|
|
351
|
-
const { stdout, code } = await sandbox.runHook({
|
|
352
|
-
payload: { session_id: "test-skip", transcript_path: transcriptPath },
|
|
353
|
-
});
|
|
354
|
-
|
|
355
|
-
assert.equal(code, 0);
|
|
356
|
-
const parsed = JSON.parse(stdout || "{}");
|
|
357
|
-
assert.deepEqual(parsed, {});
|
|
358
|
-
|
|
359
|
-
const state = sandbox.readState();
|
|
360
|
-
assert.ok(state);
|
|
361
|
-
assert.equal(state.checkCount, 4, "counter increments even on skipped turn");
|
|
362
|
-
assert.equal(state.inputTokens, 0);
|
|
363
|
-
assert.equal(sandbox.tmpFileExists(), false);
|
|
364
|
-
});
|
|
@@ -1,341 +0,0 @@
|
|
|
1
|
-
#!/usr/bin/env node
|
|
2
|
-
/**
|
|
3
|
-
* gsd-t-context-meter.js
|
|
4
|
-
*
|
|
5
|
-
* PostToolUse hook entry point for GSD-T's Context Meter (M34).
|
|
6
|
-
*
|
|
7
|
-
* Wires together:
|
|
8
|
-
* - bin/context-meter-config.cjs (loadConfig)
|
|
9
|
-
* - scripts/context-meter/transcript-parser.js (parseTranscript)
|
|
10
|
-
* - scripts/context-meter/estimate-tokens.js (estimateTokens — local, zero API cost)
|
|
11
|
-
* - scripts/context-meter/threshold.js (computePct/bandFor/buildAdditionalContext)
|
|
12
|
-
*
|
|
13
|
-
* Contract: .gsd-t/contracts/context-meter-contract.md
|
|
14
|
-
*
|
|
15
|
-
* CRITICAL INVARIANT — FAIL OPEN:
|
|
16
|
-
* Every failure path resolves to `{}` on stdout and exits 0. This hook must
|
|
17
|
-
* NEVER throw, NEVER exit non-zero, and NEVER log message content. The
|
|
18
|
-
* entire `runMeter` body is wrapped in a try/catch that swallows anything
|
|
19
|
-
* unexpected and returns `{}` — the same shape the CLI shim emits on parse
|
|
20
|
-
* failure of its own stdin. See contract rule #1.
|
|
21
|
-
*
|
|
22
|
-
* Testability:
|
|
23
|
-
* `runMeter({ payload, projectRoot, env, clock?, _parseTranscript?,
|
|
24
|
-
* _estimateTokens?, _loadConfig? })` is the pure async core. Tests
|
|
25
|
-
* fabricate payloads and inject stubs; production code uses only the CLI
|
|
26
|
-
* shim at the bottom of the file (runs when `require.main === module`).
|
|
27
|
-
*
|
|
28
|
-
* @module scripts/gsd-t-context-meter
|
|
29
|
-
*/
|
|
30
|
-
|
|
31
|
-
"use strict";
|
|
32
|
-
|
|
33
|
-
const fs = require("fs");
|
|
34
|
-
const path = require("path");
|
|
35
|
-
|
|
36
|
-
const { loadConfig: realLoadConfig } = require("../bin/context-meter-config.cjs");
|
|
37
|
-
const { parseTranscript: realParseTranscript } = require("./context-meter/transcript-parser");
|
|
38
|
-
const { estimateTokens: realEstimateTokens } = require("./context-meter/estimate-tokens");
|
|
39
|
-
const { computePct, bandFor, buildAdditionalContext } = require("./context-meter/threshold");
|
|
40
|
-
const { windowForModel } = require("../bin/model-windows.cjs");
|
|
41
|
-
|
|
42
|
-
const STATE_VERSION = 1;
|
|
43
|
-
|
|
44
|
-
/* ─────────────────────────── state file helpers ─────────────────────────── */
|
|
45
|
-
|
|
46
|
-
/**
|
|
47
|
-
* Read the current state file. Returns a fresh default on missing file or
|
|
48
|
-
* corruption (never throws).
|
|
49
|
-
*/
|
|
50
|
-
function readState(statePath) {
|
|
51
|
-
try {
|
|
52
|
-
const raw = fs.readFileSync(statePath, "utf8");
|
|
53
|
-
const parsed = JSON.parse(raw);
|
|
54
|
-
if (!parsed || typeof parsed !== "object" || parsed.version !== STATE_VERSION) {
|
|
55
|
-
return defaultState();
|
|
56
|
-
}
|
|
57
|
-
return parsed;
|
|
58
|
-
} catch (_) {
|
|
59
|
-
return defaultState();
|
|
60
|
-
}
|
|
61
|
-
}
|
|
62
|
-
|
|
63
|
-
function defaultState() {
|
|
64
|
-
return {
|
|
65
|
-
version: STATE_VERSION,
|
|
66
|
-
timestamp: null,
|
|
67
|
-
inputTokens: 0,
|
|
68
|
-
modelWindowSize: 0,
|
|
69
|
-
pct: 0,
|
|
70
|
-
threshold: "normal",
|
|
71
|
-
checkCount: 0,
|
|
72
|
-
lastError: null,
|
|
73
|
-
};
|
|
74
|
-
}
|
|
75
|
-
|
|
76
|
-
/**
|
|
77
|
-
* Atomically write state to disk: write to `{statePath}.tmp` → rename to
|
|
78
|
-
* `{statePath}`. Creates parent directories as needed. Never throws.
|
|
79
|
-
*/
|
|
80
|
-
function writeStateAtomic(statePath, state) {
|
|
81
|
-
try {
|
|
82
|
-
const dir = path.dirname(statePath);
|
|
83
|
-
fs.mkdirSync(dir, { recursive: true });
|
|
84
|
-
const tmp = `${statePath}.tmp`;
|
|
85
|
-
fs.writeFileSync(tmp, JSON.stringify(state, null, 2), "utf8");
|
|
86
|
-
fs.renameSync(tmp, statePath);
|
|
87
|
-
} catch (_) {
|
|
88
|
-
/* fail open — logging will also be best-effort */
|
|
89
|
-
}
|
|
90
|
-
}
|
|
91
|
-
|
|
92
|
-
/* ──────────────────────────── logging helper ─────────────────────────── */
|
|
93
|
-
|
|
94
|
-
/**
|
|
95
|
-
* Append a short line-based diagnostic to logPath. Line format:
|
|
96
|
-
* "{ISO-timestamp} {LEVEL} {category} {short-detail}"
|
|
97
|
-
*
|
|
98
|
-
* NEVER logs transcript content, message text, or API request bodies.
|
|
99
|
-
*/
|
|
100
|
-
function appendLog(logPath, level, category, detail, clock) {
|
|
101
|
-
try {
|
|
102
|
-
const dir = path.dirname(logPath);
|
|
103
|
-
fs.mkdirSync(dir, { recursive: true });
|
|
104
|
-
const ts = (clock ? clock() : new Date()).toISOString();
|
|
105
|
-
const line = `${ts} ${level} ${category} ${detail || ""}\n`;
|
|
106
|
-
fs.appendFileSync(logPath, line, "utf8");
|
|
107
|
-
} catch (_) {
|
|
108
|
-
/* logging failure must not affect fail-open behavior */
|
|
109
|
-
}
|
|
110
|
-
}
|
|
111
|
-
|
|
112
|
-
/* ──────────────────────── core: runMeter() ──────────────────────── */
|
|
113
|
-
|
|
114
|
-
/**
|
|
115
|
-
* Run the context meter once.
|
|
116
|
-
*
|
|
117
|
-
* @param {object} opts
|
|
118
|
-
* @param {object} opts.payload parsed PostToolUse JSON (with transcript_path)
|
|
119
|
-
* @param {string} opts.projectRoot normally process.cwd()
|
|
120
|
-
* @param {object} opts.env normally process.env
|
|
121
|
-
* @param {Function} [opts.clock] optional () => Date (test seam)
|
|
122
|
-
* @param {Function} [opts._loadConfig] optional loadConfig stub (test seam)
|
|
123
|
-
* @param {Function} [opts._parseTranscript] optional parseTranscript stub (test seam)
|
|
124
|
-
* @param {Function} [opts._estimateTokens] optional estimateTokens stub (test seam)
|
|
125
|
-
* @returns {Promise<object>} `{}` or `{ additionalContext: "..." }`
|
|
126
|
-
*/
|
|
127
|
-
async function runMeter(opts) {
|
|
128
|
-
// Outer try/catch guarantees we NEVER throw. Any unexpected error → `{}`.
|
|
129
|
-
try {
|
|
130
|
-
const {
|
|
131
|
-
payload,
|
|
132
|
-
projectRoot,
|
|
133
|
-
env,
|
|
134
|
-
clock,
|
|
135
|
-
_loadConfig = realLoadConfig,
|
|
136
|
-
_parseTranscript = realParseTranscript,
|
|
137
|
-
_estimateTokens = realEstimateTokens,
|
|
138
|
-
} = opts || {};
|
|
139
|
-
|
|
140
|
-
const root = projectRoot || process.cwd();
|
|
141
|
-
const envObj = env || {};
|
|
142
|
-
const now = () => (clock ? clock() : new Date());
|
|
143
|
-
|
|
144
|
-
// 1. Load config (missing file → defaults). Any throw → bail out fail-open.
|
|
145
|
-
let cfg;
|
|
146
|
-
try {
|
|
147
|
-
cfg = _loadConfig(root);
|
|
148
|
-
} catch (_) {
|
|
149
|
-
return {};
|
|
150
|
-
}
|
|
151
|
-
|
|
152
|
-
const statePath = path.isAbsolute(cfg.statePath)
|
|
153
|
-
? cfg.statePath
|
|
154
|
-
: path.join(root, cfg.statePath);
|
|
155
|
-
const logPath = path.isAbsolute(cfg.logPath)
|
|
156
|
-
? cfg.logPath
|
|
157
|
-
: path.join(root, cfg.logPath);
|
|
158
|
-
|
|
159
|
-
// 2. Read (possibly corrupt) state, increment checkCount immediately.
|
|
160
|
-
const state = readState(statePath);
|
|
161
|
-
state.version = STATE_VERSION;
|
|
162
|
-
state.modelWindowSize = cfg.modelWindowSize;
|
|
163
|
-
state.checkCount = (Number.isInteger(state.checkCount) ? state.checkCount : 0) + 1;
|
|
164
|
-
|
|
165
|
-
// 3. Check-frequency gate: not our turn → persist counter and bail out `{}`.
|
|
166
|
-
if (state.checkCount % cfg.checkFrequency !== 0) {
|
|
167
|
-
writeStateAtomic(statePath, state);
|
|
168
|
-
return {};
|
|
169
|
-
}
|
|
170
|
-
|
|
171
|
-
// 4. Extract transcript_path from payload (fail open if missing).
|
|
172
|
-
const transcriptPath =
|
|
173
|
-
payload && typeof payload === "object" && typeof payload.transcript_path === "string"
|
|
174
|
-
? payload.transcript_path
|
|
175
|
-
: null;
|
|
176
|
-
|
|
177
|
-
if (!transcriptPath) {
|
|
178
|
-
state.lastError = {
|
|
179
|
-
code: "no_transcript",
|
|
180
|
-
message: "payload missing transcript_path",
|
|
181
|
-
timestamp: now().toISOString(),
|
|
182
|
-
};
|
|
183
|
-
writeStateAtomic(statePath, state);
|
|
184
|
-
appendLog(logPath, "ERROR", "no_transcript", "payload missing transcript_path", clock);
|
|
185
|
-
return {};
|
|
186
|
-
}
|
|
187
|
-
|
|
188
|
-
// 5. Parse transcript (streaming, async). null → bail out.
|
|
189
|
-
let parsed;
|
|
190
|
-
try {
|
|
191
|
-
parsed = await _parseTranscript(transcriptPath);
|
|
192
|
-
} catch (_) {
|
|
193
|
-
parsed = null;
|
|
194
|
-
}
|
|
195
|
-
if (!parsed || !Array.isArray(parsed.messages)) {
|
|
196
|
-
state.lastError = {
|
|
197
|
-
code: "parse_failure",
|
|
198
|
-
message: `parseTranscript returned null for ${path.basename(transcriptPath)}`,
|
|
199
|
-
timestamp: now().toISOString(),
|
|
200
|
-
};
|
|
201
|
-
writeStateAtomic(statePath, state);
|
|
202
|
-
appendLog(
|
|
203
|
-
logPath,
|
|
204
|
-
"ERROR",
|
|
205
|
-
"parse_failure",
|
|
206
|
-
`${path.basename(transcriptPath)}`,
|
|
207
|
-
clock
|
|
208
|
-
);
|
|
209
|
-
return {};
|
|
210
|
-
}
|
|
211
|
-
|
|
212
|
-
// 5b. Resolve the EFFECTIVE context window from the model the orchestrator
|
|
213
|
-
// session is actually running (parsed.model). Opus 4.6/4.7 and Sonnet 4.x
|
|
214
|
-
// ship a 1M window; the config default (200k) is a legacy fallback that
|
|
215
|
-
// overcounts usage 5× and fires the headless handoff far too early. We
|
|
216
|
-
// only override when the transcript reports a model — a missing model or
|
|
217
|
-
// an explicit project config value falls through to cfg.modelWindowSize.
|
|
218
|
-
const effectiveWindow =
|
|
219
|
-
typeof parsed.model === "string" && parsed.model.length > 0
|
|
220
|
-
? windowForModel(parsed.model)
|
|
221
|
-
: cfg.modelWindowSize;
|
|
222
|
-
state.modelWindowSize = effectiveWindow;
|
|
223
|
-
|
|
224
|
-
// 6. Estimate tokens locally (no API call, zero cost).
|
|
225
|
-
let tokenResp;
|
|
226
|
-
try {
|
|
227
|
-
tokenResp = _estimateTokens({
|
|
228
|
-
system: parsed.system || "",
|
|
229
|
-
messages: parsed.messages,
|
|
230
|
-
});
|
|
231
|
-
} catch (_) {
|
|
232
|
-
tokenResp = null;
|
|
233
|
-
}
|
|
234
|
-
|
|
235
|
-
if (!tokenResp || !Number.isFinite(tokenResp.inputTokens)) {
|
|
236
|
-
state.inputTokens = 0;
|
|
237
|
-
state.pct = 0;
|
|
238
|
-
state.threshold = "normal";
|
|
239
|
-
state.timestamp = now().toISOString();
|
|
240
|
-
state.lastError = {
|
|
241
|
-
code: "estimate_error",
|
|
242
|
-
message: "estimateTokens returned null",
|
|
243
|
-
timestamp: state.timestamp,
|
|
244
|
-
};
|
|
245
|
-
writeStateAtomic(statePath, state);
|
|
246
|
-
appendLog(logPath, "ERROR", "estimate_error", "estimateTokens null", clock);
|
|
247
|
-
return {};
|
|
248
|
-
}
|
|
249
|
-
|
|
250
|
-
// 8. Success path — compute pct, band, possibly emit additionalContext.
|
|
251
|
-
const pct = computePct({
|
|
252
|
-
inputTokens: tokenResp.inputTokens,
|
|
253
|
-
modelWindowSize: effectiveWindow,
|
|
254
|
-
});
|
|
255
|
-
const band = bandFor(pct, cfg.thresholdPct);
|
|
256
|
-
|
|
257
|
-
state.inputTokens = tokenResp.inputTokens;
|
|
258
|
-
state.pct = pct;
|
|
259
|
-
state.threshold = band;
|
|
260
|
-
state.timestamp = now().toISOString();
|
|
261
|
-
state.lastError = null;
|
|
262
|
-
writeStateAtomic(statePath, state);
|
|
263
|
-
appendLog(
|
|
264
|
-
logPath,
|
|
265
|
-
"INFO",
|
|
266
|
-
"measure",
|
|
267
|
-
`tokens=${tokenResp.inputTokens} pct=${pct.toFixed(1)} band=${band} ` +
|
|
268
|
-
`window=${effectiveWindow}${parsed.model ? ` model=${parsed.model}` : ""}`,
|
|
269
|
-
clock
|
|
270
|
-
);
|
|
271
|
-
|
|
272
|
-
const additionalContext = buildAdditionalContext({
|
|
273
|
-
pct,
|
|
274
|
-
modelWindowSize: effectiveWindow,
|
|
275
|
-
thresholdPct: cfg.thresholdPct,
|
|
276
|
-
});
|
|
277
|
-
if (additionalContext) {
|
|
278
|
-
return { additionalContext };
|
|
279
|
-
}
|
|
280
|
-
return {};
|
|
281
|
-
} catch (_) {
|
|
282
|
-
// Absolute safety net — fail open no matter what.
|
|
283
|
-
return {};
|
|
284
|
-
}
|
|
285
|
-
}
|
|
286
|
-
|
|
287
|
-
/* ──────────────────────────── CLI shim ──────────────────────────── */
|
|
288
|
-
|
|
289
|
-
function readStdin() {
|
|
290
|
-
return new Promise((resolve) => {
|
|
291
|
-
let data = "";
|
|
292
|
-
try {
|
|
293
|
-
if (process.stdin.isTTY) {
|
|
294
|
-
resolve("");
|
|
295
|
-
return;
|
|
296
|
-
}
|
|
297
|
-
process.stdin.setEncoding("utf8");
|
|
298
|
-
process.stdin.on("data", (chunk) => {
|
|
299
|
-
data += chunk;
|
|
300
|
-
});
|
|
301
|
-
process.stdin.on("end", () => resolve(data));
|
|
302
|
-
process.stdin.on("error", () => resolve(""));
|
|
303
|
-
} catch (_) {
|
|
304
|
-
resolve("");
|
|
305
|
-
}
|
|
306
|
-
});
|
|
307
|
-
}
|
|
308
|
-
|
|
309
|
-
async function main() {
|
|
310
|
-
let payload = null;
|
|
311
|
-
try {
|
|
312
|
-
const raw = await readStdin();
|
|
313
|
-
payload = raw ? JSON.parse(raw) : null;
|
|
314
|
-
} catch (_) {
|
|
315
|
-
payload = null;
|
|
316
|
-
}
|
|
317
|
-
|
|
318
|
-
let out = {};
|
|
319
|
-
try {
|
|
320
|
-
out = await runMeter({
|
|
321
|
-
payload: payload || {},
|
|
322
|
-
projectRoot: process.cwd(),
|
|
323
|
-
env: process.env,
|
|
324
|
-
});
|
|
325
|
-
} catch (_) {
|
|
326
|
-
out = {};
|
|
327
|
-
}
|
|
328
|
-
|
|
329
|
-
try {
|
|
330
|
-
process.stdout.write(JSON.stringify(out || {}));
|
|
331
|
-
} catch (_) {
|
|
332
|
-
process.stdout.write("{}");
|
|
333
|
-
}
|
|
334
|
-
process.exit(0);
|
|
335
|
-
}
|
|
336
|
-
|
|
337
|
-
module.exports = { runMeter, readState, writeStateAtomic, defaultState };
|
|
338
|
-
|
|
339
|
-
if (require.main === module) {
|
|
340
|
-
main();
|
|
341
|
-
}
|