feat(observability): opt-in token capture in Stop hook (VOYAGE_TOKEN_METER)

This commit is contained in:
Kjell Tore Guttormsen 2026-06-26 14:42:58 +02:00
commit 46d51f8088
3 changed files with 243 additions and 2 deletions

View file

@ -17,6 +17,7 @@
// - All stderr prefixed with [voyage]. // - All stderr prefixed with [voyage].
// - EXDEV mitigation: tmp file in same dir as target (do NOT use atomicWriteJson). // - EXDEV mitigation: tmp file in same dir as target (do NOT use atomicWriteJson).
import { stdin } from 'node:process';
import { readFileSync, existsSync, writeFileSync, renameSync } from 'node:fs'; import { readFileSync, existsSync, writeFileSync, renameSync } from 'node:fs';
import { join, dirname } from 'node:path'; import { join, dirname } from 'node:path';
import { transformToPrometheus } from '../../lib/exporters/textfile-format.mjs'; import { transformToPrometheus } from '../../lib/exporters/textfile-format.mjs';
@ -24,6 +25,13 @@ import { transformToOtlpJson } from '../../lib/exporters/otlp-format.mjs';
import { validateTextfilePath } from '../../lib/exporters/path-validator.mjs'; import { validateTextfilePath } from '../../lib/exporters/path-validator.mjs';
import { validateOtlpEndpoint } from '../../lib/exporters/endpoint-validator.mjs'; import { validateOtlpEndpoint } from '../../lib/exporters/endpoint-validator.mjs';
import { applyFieldAllowlist } from '../../lib/exporters/field-allowlist.mjs'; import { applyFieldAllowlist } from '../../lib/exporters/field-allowlist.mjs';
import { captureTokenUsage } from '../../lib/stats/token-usage.mjs';
async function readStdin() {
let data = '';
for await (const chunk of stdin) data += chunk;
return data;
}
const VALID_MODES = new Set(['off', 'textfile', 'otlp']); const VALID_MODES = new Set(['off', 'textfile', 'otlp']);
const TEXTFILE_NAME = 'voyage.prom'; const TEXTFILE_NAME = 'voyage.prom';
@ -38,6 +46,7 @@ const STATS_FILES = [
{ file: 'trekexecute-stats.jsonl', schema: 'trekexecute' }, { file: 'trekexecute-stats.jsonl', schema: 'trekexecute' },
{ file: 'trekreview-stats.jsonl', schema: 'trekreview' }, { file: 'trekreview-stats.jsonl', schema: 'trekreview' },
{ file: 'trekcontinue-stats.jsonl', schema: 'trekcontinue' }, { file: 'trekcontinue-stats.jsonl', schema: 'trekcontinue' },
{ file: 'token-usage-stats.jsonl', schema: 'token-usage' },
]; ];
function loadAndAllowlist(dataDir) { function loadAndAllowlist(dataDir) {
@ -134,6 +143,26 @@ async function exportOtlp(records, env) {
(async () => { (async () => {
try { try {
const env = process.env; const env = process.env;
// SKAL-2: opt-in main-context token/cost capture (default off → zero added
// latency). Rides this existing Stop hook rather than a new hook script —
// avoids a second-Stop-hook race and keeps the README hook-count pin intact.
// Runs BEFORE the export-mode gate so capture is independent of export.
// Fail-open: any error is swallowed; capture must never block Stop.
if (env.VOYAGE_TOKEN_METER) {
try {
const raw = await readStdin();
if (raw.trim()) {
const payload = JSON.parse(raw);
captureTokenUsage({
transcriptPath: payload.transcript_path,
sessionId: payload.session_id,
dataDir: env.CLAUDE_PLUGIN_DATA,
});
}
} catch { /* fail-open: never block Stop, never throw */ }
}
const mode = (env.VOYAGE_EXPORT_MODE || 'off').toLowerCase(); const mode = (env.VOYAGE_EXPORT_MODE || 'off').toLowerCase();
if (mode === 'off') return; if (mode === 'off') return;

View file

@ -15,11 +15,15 @@
// main-transcript sum UNDER-counts total Voyage cost. Every record is stamped // main-transcript sum UNDER-counts total Voyage cost. Every record is stamped
// scope:'main-context' so no reader mistakes cost_usd for the session total. // scope:'main-context' so no reader mistakes cost_usd for the session total.
// //
// All functions in this module are PURE (no I/O). The impure capture shell // The four core functions (parseTranscriptUsage, deriveCost, buildRecord,
// (captureTokenUsage) is added in Step 4. // upsertSessionRecord) are PURE (no I/O). captureTokenUsage is the impure
// shell that wires them to the filesystem (read transcript → upsert jsonl).
// //
// Zero npm dependencies. Node stdlib only. // Zero npm dependencies. Node stdlib only.
import { readFileSync, existsSync, writeFileSync, renameSync, statSync } from 'node:fs';
import { join, dirname } from 'node:path';
// Per-Mtok USD prices, resolved 2026-06-26 via the claude-api skill reference: // Per-Mtok USD prices, resolved 2026-06-26 via the claude-api skill reference:
// base input/output from the model table; cache rates from the prompt-caching // base input/output from the model table; cache rates from the prompt-caching
// doc multipliers (cache_read 0.1x, write_5m 1.25x, write_1h 2.0x of input). // doc multipliers (cache_read 0.1x, write_5m 1.25x, write_1h 2.0x of input).
@ -159,3 +163,64 @@ export function upsertSessionRecord(existingText, record) {
if (!replaced) out.push(JSON.stringify(record)); if (!replaced) out.push(JSON.stringify(record));
return out.join('\n') + '\n'; return out.join('\n') + '\n';
} }
/**
* Last main-chain (non-sidechain) assistant model in the transcript. Used to
* pick the price-table key. Returns null when no model is found ( deriveCost
* refuses to estimate). Pure.
*/
export function lastMainChainModel(text) {
const lines = (text || '').split('\n');
let model = null;
for (const line of lines) {
const trimmed = line.trim();
if (trimmed === '') continue;
let obj;
try { obj = JSON.parse(trimmed); }
catch { continue; }
if (!obj || obj.type !== 'assistant' || obj.isSidechain === true) continue;
const m = obj.message && obj.message.model;
if (typeof m === 'string' && m) model = m;
}
return model;
}
/**
* Impure capture shell (Step 4). Reads the transcript, derives main-context
* token totals + cost, and UPSERTS a one-line-per-session record into
* {dataDir}/token-usage-stats.jsonl via an atomic temp+rename write
* (EXDEV mitigation: tmp lives in the same dir as the target).
*
* Returns the written record, or null when skipped (no path / no dataDir /
* transcript not a readable regular file). Throws propagate to the caller
* the Stop hook wraps this in try/catch so capture stays fail-open.
*
* @param {{transcriptPath:string, sessionId:string, dataDir:string,
* now?:string}} args
*/
export function captureTokenUsage({ transcriptPath, sessionId, dataDir, now }) {
if (!transcriptPath || !dataDir) return null;
let st;
try { st = statSync(transcriptPath); }
catch { return null; }
if (!st.isFile()) return null;
const text = readFileSync(transcriptPath, 'utf-8');
const totals = parseTranscriptUsage(text);
const model = lastMainChainModel(text);
const record = buildRecord({
sessionId,
model,
totals,
now: now || new Date().toISOString(),
});
const outPath = join(dataDir, 'token-usage-stats.jsonl');
const existing = existsSync(outPath) ? readFileSync(outPath, 'utf-8') : '';
const updated = upsertSessionRecord(existing, record);
const tmpPath = join(dirname(outPath), '.token-usage-stats.jsonl.tmp');
writeFileSync(tmpPath, updated);
renameSync(tmpPath, outPath);
return record;
}

View file

@ -0,0 +1,147 @@
// tests/hooks/otel-export-token-capture.test.mjs
// SKAL-2 Step 4 — opt-in main-context token/cost capture folded into the Stop
// hook (otel-export.mjs), gated by VOYAGE_TOKEN_METER.
//
// Fail-open contract: any error → exit 0, no token file, Stop never blocked.
// Pattern: tests/hooks/otel-export.test.mjs (setupDataDir + runHookWithEnv).
import { test } from 'node:test';
import { strict as assert } from 'node:assert';
import { mkdtempSync, rmSync, existsSync, readFileSync, writeFileSync } from 'node:fs';
import { join, dirname } from 'node:path';
import { tmpdir } from 'node:os';
import { fileURLToPath } from 'node:url';
import { runHookWithEnv } from '../helpers/hook-helper.mjs';
const HERE = dirname(fileURLToPath(import.meta.url));
const HOOK_PATH = join(HERE, '..', '..', 'hooks', 'scripts', 'otel-export.mjs');
const FIXTURE = join(HERE, '..', 'fixtures', 'token-usage', 'sample-transcript.jsonl');
const TOKEN_FILE = 'token-usage-stats.jsonl';
const mkDataDir = () => mkdtempSync(join(tmpdir(), 'voyage-token-capture-'));
const stdinFor = (extra = {}) => JSON.stringify({ transcript_path: FIXTURE, session_id: 'sess-1', ...extra });
const readToken = (dir) => readFileSync(join(dir, TOKEN_FILE), 'utf-8').trim().split('\n').filter(Boolean).map(JSON.parse);
test('VOYAGE_TOKEN_METER=1 → writes token-usage-stats.jsonl with expected totals + scope:main-context', async () => {
const dir = mkDataDir();
try {
const r = await runHookWithEnv(HOOK_PATH, stdinFor(), {
VOYAGE_TOKEN_METER: '1',
VOYAGE_EXPORT_MODE: 'off',
CLAUDE_PLUGIN_DATA: dir,
});
assert.equal(r.code, 0);
assert.equal(existsSync(join(dir, TOKEN_FILE)), true, 'token file must be written');
const recs = readToken(dir);
assert.equal(recs.length, 1);
const rec = recs[0];
// Fixture totals: dedup-by-requestId + sidechain-exclusion (verified in Step 1).
assert.equal(rec.tokens_input, 3000);
assert.equal(rec.tokens_output, 500);
assert.equal(rec.tokens_cache_creation, 1000);
assert.equal(rec.tokens_cache_read, 6000);
assert.equal(rec.scope, 'main-context');
assert.equal(rec.model, 'claude-opus-4-8');
assert.ok(Math.abs(rec.cost_usd - 0.03675) < 1e-9, `cost ${rec.cost_usd}`);
assert.equal(rec.session_id, 'sess-1');
} finally {
rmSync(dir, { recursive: true, force: true });
}
});
test('UPSERT — running twice for the same session_id keeps ONE line (not two)', async () => {
const dir = mkDataDir();
try {
const env = { VOYAGE_TOKEN_METER: '1', VOYAGE_EXPORT_MODE: 'off', CLAUDE_PLUGIN_DATA: dir };
await runHookWithEnv(HOOK_PATH, stdinFor(), env);
const r2 = await runHookWithEnv(HOOK_PATH, stdinFor(), env);
assert.equal(r2.code, 0);
const recs = readToken(dir);
assert.equal(recs.length, 1, 'upsert must NOT append a second line for the same session');
assert.equal(recs[0].tokens_input, 3000);
} finally {
rmSync(dir, { recursive: true, force: true });
}
});
test('UPSERT — distinct session_ids append (one line per session)', async () => {
const dir = mkDataDir();
try {
const env = { VOYAGE_TOKEN_METER: '1', VOYAGE_EXPORT_MODE: 'off', CLAUDE_PLUGIN_DATA: dir };
await runHookWithEnv(HOOK_PATH, stdinFor({ session_id: 'A' }), env);
await runHookWithEnv(HOOK_PATH, stdinFor({ session_id: 'B' }), env);
const recs = readToken(dir);
assert.equal(recs.length, 2);
assert.deepEqual(recs.map(r => r.session_id).sort(), ['A', 'B']);
} finally {
rmSync(dir, { recursive: true, force: true });
}
});
test('OPT-IN — VOYAGE_TOKEN_METER unset → NO token file written', async () => {
const dir = mkDataDir();
try {
const r = await runHookWithEnv(HOOK_PATH, stdinFor(), {
VOYAGE_TOKEN_METER: '', // explicit empty mimics unset
VOYAGE_EXPORT_MODE: 'off',
CLAUDE_PLUGIN_DATA: dir,
});
assert.equal(r.code, 0);
assert.equal(existsSync(join(dir, TOKEN_FILE)), false, 'no token file without opt-in');
} finally {
rmSync(dir, { recursive: true, force: true });
}
});
test('FAIL-OPEN — malformed stdin → exit 0, no token file, no throw', async () => {
const dir = mkDataDir();
try {
const r = await runHookWithEnv(HOOK_PATH, 'not valid json {{{', {
VOYAGE_TOKEN_METER: '1',
VOYAGE_EXPORT_MODE: 'off',
CLAUDE_PLUGIN_DATA: dir,
});
assert.equal(r.code, 0);
assert.equal(existsSync(join(dir, TOKEN_FILE)), false);
} finally {
rmSync(dir, { recursive: true, force: true });
}
});
test('FAIL-OPEN — missing transcript file → exit 0, no token file', async () => {
const dir = mkDataDir();
try {
const r = await runHookWithEnv(
HOOK_PATH,
JSON.stringify({ transcript_path: '/no/such/transcript.jsonl', session_id: 'x' }),
{ VOYAGE_TOKEN_METER: '1', VOYAGE_EXPORT_MODE: 'off', CLAUDE_PLUGIN_DATA: dir },
);
assert.equal(r.code, 0);
assert.equal(existsSync(join(dir, TOKEN_FILE)), false);
} finally {
rmSync(dir, { recursive: true, force: true });
}
});
test('capture coexists with export — VOYAGE_TOKEN_METER=1 + textfile mode writes BOTH', async () => {
const dir = mkDataDir();
try {
// Seed a stats file so the export path has something to write.
writeFileSync(join(dir, 'trekplan-stats.jsonl'),
JSON.stringify({ ts: '2026-06-26T08:00:00.000Z', slug: 't', mode: 'default', codebase_files: 10, profile: 'premium' }) + '\n');
const r = await runHookWithEnv(HOOK_PATH, stdinFor(), {
VOYAGE_TOKEN_METER: '1',
VOYAGE_EXPORT_MODE: 'textfile',
CLAUDE_PLUGIN_DATA: dir,
});
assert.equal(r.code, 0);
assert.equal(existsSync(join(dir, TOKEN_FILE)), true, 'token file written');
assert.equal(existsSync(join(dir, 'voyage.prom')), true, 'export still completed');
// The exported textfile should now also carry the token metric (STATS_FILES wiring).
const prom = readFileSync(join(dir, 'voyage.prom'), 'utf-8');
assert.match(prom, /voyage_token_usage_tokens_input/);
assert.ok(!prom.includes('sess-1'), 'session_id must not leak into export (CWE-212)');
} finally {
rmSync(dir, { recursive: true, force: true });
}
});