:hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.
Python ab-testingai-analyticsanalyticscdpdata-warehouseexperimentsfeature-flagsjavascriptproduct-analyticspythonreactsession-replaysurveystypescriptweb-analytics
import { describe, expect, it } from 'vitest'
import { parseExecCallInnerArgs } from '@/tools/exec'
import { skillAnalyticsProperties } from '@/tools/skills/analytics'
/**
* These lock in three regressions: a skill read that records no skill (the whole
* point), an agent-supplied string reaching analytics unvalidated, and a paged
* continuation counted as a fresh load.
*/
describe('skillAnalyticsProperties', () => {
it.each([
['skill-get', { skill_name: 'conductor' }, 'conductor'],
['skill-file-get', { skill_name: 'conductor', file_path: 'references/x.md' }, 'conductor'],
// The deprecated aliases reach analytics under their own name, so a set that
// lists only the canonical names is a silent hole.
['llma-skill-get', { skill_name: 'conductor' }, 'conductor'],
['llma-skill-file-get', { skill_name: 'conductor' }, 'conductor'],
// Digits and internal hyphens are legal in a stored name.
['skill-get', { skill_name: 'signals-scout-v2' }, 'signals-scout-v2'],
['skill-get', { skill_name: 'a'.repeat(64) }, 'a'.repeat(64)],
])('identifies the skill %s read', (tool, args, expected) => {
expect(skillAnalyticsProperties(tool, args)).toMatchObject({ $mcp_skill_name: expected })
})
it.each([
// Stamping a non-skill tool would put caller text on unrelated events.
['a non-skill tool', 'execute-sql', { skill_name: 'conductor' }],
['skill-list, which reads no single skill', 'skill-list', { search: 'conductor' }],
// Writes already emit `llma skill *` server-side; stamping here too would
// give two sources that disagree.
['a write tool', 'skill-update', { skill_name: 'conductor' }],
['an unresolved tool name', undefined, { skill_name: 'conductor' }],
])('records nothing for %s', (_label, tool, args) => {
expect(skillAnalyticsProperties(tool, args)).toEqual({})
})
// The rule `execCommandAnalyticsProperties` states: recorded values are value-free
// by construction. A name is recordable only if it matches the shape the store
// enforces at creation — anything else is the agent's own text.
it.each([
['spaces', 'my skill'],
['uppercase', 'Conductor'],
['path traversal', '../../etc/passwd'],
['a url', 'https://example.com/x'],
['empty', ''],
['a leading hyphen', '-conductor'],
['over 64 characters', 'a'.repeat(65)],
['sql-ish text', "conductor'; DROP TABLE--"],
['newlines', 'conductor\nsecond'],
['a non-string', 42],
['null', null],
['an object', { nested: true }],
])('drops a skill name that is %s', (_label, value) => {
expect(skillAnalyticsProperties('skill-get', { skill_name: value })).toEqual({})
})
// Analytics must never break a tool call, so malformed input yields no properties
// rather than throwing.
it.each([
['undefined', undefined],
['null', null],
['a string', 'skill_name=conductor'],
['an empty object', {}],
])('returns empty for %s args', (_label, args) => {
expect(skillAnalyticsProperties('skill-get', args)).toEqual({})
})
describe('body offset', () => {
// skill-get returns a body in slices, so one skill load is several calls that
// differ only by offset. Without this, count() overstates loads several-fold.
it.each([
['the first page', 0, 0],
['a continuation', 5000, 5000],
['a numeric string', '5000', 5000],
])('records %s', (_label, input, expected) => {
expect(skillAnalyticsProperties('skill-get', { skill_name: 'conductor', body_offset: input })).toEqual({
$mcp_skill_name: 'conductor',
$mcp_skill_body_offset: expected,
})
})
it.each([
['absent', undefined],
['negative', -1],
['not a number', 'later'],
['NaN', Number.NaN],
['infinite', Number.POSITIVE_INFINITY],
['fractional', 1.5],
])('omits an offset that is %s, so its absence means "whole body"', (_label, value) => {
expect(skillAnalyticsProperties('skill-get', { skill_name: 'conductor', body_offset: value })).toEqual({
$mcp_skill_name: 'conductor',
})
})
it('omits the offset for skill-file-get, which does not paginate a body', () => {
expect(skillAnalyticsProperties('skill-file-get', { skill_name: 'conductor', body_offset: 10 })).toEqual({
$mcp_skill_name: 'conductor',
})
})
})
})
/**
* Nearly all skill reads arrive through the single `exec` dispatcher, where arguments
* live inside a command string. A property derived only from tool arguments would
* miss all of them.
*/
describe('parseExecCallInnerArgs', () => {
it.each([
['a call command', 'call skill-get {"skill_name":"conductor"}', { skill_name: 'conductor' }],
['a --json flag', 'call --json skill-get {"skill_name":"conductor"}', { skill_name: 'conductor' }],
['a --confirm flag', 'call --confirm skill-get {"skill_name":"conductor"}', { skill_name: 'conductor' }],
['a flag after the JSON body', 'call skill-get {"skill_name":"conductor"} --json', { skill_name: 'conductor' }],
[
'extra whitespace and more keys',
'call skill-get {"skill_name":"conductor","version":4}',
{ skill_name: 'conductor', version: 4 },
],
// The dispatcher treats a body-less call as `{}`; this must match it.
['no JSON body', 'call skill-list', {}],
])('extracts inner arguments from %s', (_label, command, expected) => {
expect(parseExecCallInnerArgs(command)).toEqual(expected)
})
it.each([
['a non-call verb', 'info skill-get'],
['an empty command', ''],
['invalid JSON', 'call skill-get {not json}'],
['a JSON scalar', 'call skill-get 5'],
['a JSON array', 'call skill-get ["conductor"]'],
['JSON null', 'call skill-get null'],
])('returns undefined for %s', (_label, command) => {
expect(parseExecCallInnerArgs(command)).toBeUndefined()
})
})