──────────────────────────────────────────────────────────────────────
PERSONNEL ROSTER
▛▜▙▟十四
─── DOSSIER ───────────────────────────
>>AI safety researcher. Registered the Shadow-LLM org and shadow-llm.com domain in 2023 after transitioning from AI efficiency. Drives the archive's taxonomy and triage.
──────────────────────────────────────────────────────────────────────
36 HEAVENLY SPIRITS· 天罡三十六星
We mirror the canonical 天罡三十六星 — the thirty-six Heavenly Spirits of Daoist cosmology, made famous by Water Margin as the 36 stars that descended as Liangshan's outlaw heroes. Inside Shadow-LLM-Guardians, each seat is a heritable codename. Hold one and you carry the character's motif.
#017Click a vacant seat above to open a pre-filled [SEAT-CLAIM] issue. We'll look at your public artifacts (papers, repos, contributed cases, defense recipes) and either confirm the seat or ask for more on-record work first.
// Seats are not purely first-come. Fit matters: 天暗星 carries the "shadow / ill-fated nobility" motif, 天杀星 carries pure offense, etc. Pick one whose character resonates with the work you do.
Open invitation.
Shadow-LLM-Guardians is open-source. There is no paid team and no application form — only contributors, working in public, on the record. We're glad to have you in any of three lanes:
- ▶ Red team — surface novel failure modes in frontier models or agent harnesses; file them here.
- ▶ Blue team — write mitigations, regression suites, harness-level guardrails; link them from cases they defend against.
- ▶ Reproducers — re-run submitted cases on current model versions and confirm or refute. Often the scarcest contribution.
The contribution graph IS the application. Drop into a case discussion on failure-cases, file your own, reproduce someone else's. After a few solid contributions, open an issue titled [ROSTER] add @your-handle with links to your work. No résumé, no interview — just the record.
// license · case content is CC-BY-4.0; code is MIT. contributions inherit.