Tanya runs reviews, comp analytics, hiring, and Drive security at One Hundred X Inc. This is her screen.
Rendered in plain HTML from Tanya's working files. Every count is real. Private values are bars.
held queue · 3 grantees held for a written answer → resolved next day (1 kept, 2 revoked) · ██████████ ████████ ████████████
Rendered from the live board, 2026-07-26 run. Private values withheld.
Rendered from the two instance brain files, 2026-07-27 run. Private values withheld.
Rendered from the screening skill and pipeline index, 2026-07-27 run. Private values withheld.
Six rows, each one checkable against the files on this page. The last row is the one that matters.
Everything above renders from working files like these. Two excerpts, redacted at the source.
The boundary file. The wall in screen two is not a promise. It is the first section of the recruiting instance's brain file, ahead of the job itself.
# Tanya HQ — ██████ instance
You are Tanya, 100x's AI talent screener and talent-ops copilot — the ██████-run instance built on the shared Tanya framework. A session rooted in this folder acts as Tanya.
## Instance boundary (read first)
- Two Tanya instances exist by design. Nik's instance (anchored on 08 Talent Management, spec v2.0 Arm 1) carries CEO-private talent performance and comp data. This instance never reads, references, or reconstructs that data. If a task needs it, route to Nik. note: the sweep in screen one was run by the other instance. This line is why the two never meet.
## Operating rules (inherited from framework v1, binding)
1. Always disclose AI nature upfront: "I'm Tanya, an AI recruiter."
2. Capture consent before proceeding with any screening interaction.
7. All outbound is draft-first. Tanya composes; ██████ sends. No auto-send anywhere.
The audit file. The sweep in screen one, as Tanya reported it the day she ran it.
# Engineering Drive — External-Share Audit (first run)
## Headline
- The sellable IP tree is CLEAN. 100 Claude OS + its 21 top-level children (the agent source/specs/playbooks) = internal only, zero external, zero anyone-with-link. No agent-IP leak.
- The broader drive has 2,103 externally-exposed files (~0.5%): 1,540 anyone-with-link + 406 shared to external domains (some overlap). This is old sales/research/ops sprawl, not agent source.
## EXECUTION (2026-07-25, on Nik's GO)
- HELD for Nik confirm (NOT revoked — the "verify" catch):
## VERIFIED FINAL STATE (re-audit 2026-07-26 — the revocation fully succeeded)
- anyone-with-link: 0 remaining (was ██████)
- The engineering drive is now clean of anyone-with-link, ██████, and ex-contractor-██████ exposure. Remaining external = active client/vendor domains (████████████████) — legitimate collaborators, KEEP.
You just read an AI Head of Talent's actual working surface. A 400,037-file drive with risky exposure taken to zero on a written GO, a comp boundary no hiring conversation can cross, and one live screening pipeline run with disclosure, consent, and evidence. The three edge cases she would not touch went to a human.
A desk like this for your team starts with Nik. Opposed to exploring further?
CEO and founder, 100x. Previously founded Home.LLC. Building AI agents for businesses since 2024. Runs reviews, hiring, and Drive security with Tanya, and keeps every people decision for himself.