Merge pull request #62 from wuyoscar/update/2026-03-29

feat: 2026-03-29 — full template audit, SKILL.md conversion, new ISC cases

Author: wuyoscar

.gitattributes

@@ -36,7 +36,6 @@ AGENT.md
 CLAUDE.md
 HANDOFF.md
 ROADMAP.md
-README_zh.md
 README_ja.md
 README_ko.md
 README_es.md
@@ -62,3 +61,9 @@ run.log
 
 # Harmful content data files (regenerate with pipeline)
 posts/
+
+# Template outputs (contain harmful model outputs — keep local only)
+templates/**/output*.txt
+templates/**/output*.json
+.omc/
+config/

.gitignore

@@ -0,0 +1,105 @@
+# ISC-Bench — Agent Entry Guide
+
+## For Humans
+
+Copy this to your AI agent (Claude Code, OpenClaw, Cursor, etc.):
+
+```text
+Help me understand and use ISC-Bench:
+https://raw.githubusercontent.com/wuyoscar/ISC-Bench/main/AGENT_README.md
+```
+
+---
+
+## For AI Agents
+
+### Goal
+
+Help the user understand, inspect, reproduce, or contribute to the `ISC-Bench` repository.
+
+`ISC-Bench` is an academic safety research repository for studying **Internal Safety Collapse (ISC)** in frontier LLMs through the **TVD** framework:
+
+- **Task**: a normal-looking workflow
+- **Validator**: a checker that defines success
+- **Data**: missing fields the model must complete
+
+Core claim:
+
+- models can produce harmful or policy-relevant content
+- not because of a direct malicious prompt
+- but as a side effect of completing an ordinary-looking task
+
+### Boundaries
+
+When working with this repository, stay within these limits:
+
+- treat the repository as academic safety research
+- do not rewrite the project as operational misuse guidance
+- do not strengthen harmful examples unnecessarily
+- prefer public templates and documented workflows
+- use [`VERIFICATION.md`](VERIFICATION.md) when judging whether a case counts as **Triggered**
+
+### Step 1: Clone the Repository
+
+```bash
+git clone https://github.com/wuyoscar/ISC-Bench.git
+cd ISC-Bench
+```
+
+### Step 2: Read in Order
+
+Read these files in this order:
+
+1. [`README.md`](README.md) — project overview, leaderboard, and public entry points
+2. [`VERIFICATION.md`](VERIFICATION.md) — how ISC-Bench defines **Triggered** and verifies cases
+3. [`templates/README.md`](templates/README.md) — public scenario library
+4. [`experiment/README.md`](experiment/README.md) — reproducible evaluation pipelines
+5. [`community/README.md`](community/README.md) — attributed reproductions and evidence
+
+### Step 3: Choose the Right Path
+
+Use the path that matches the user's goal:
+
+- **Inspect evidence**
+  Open the JailbreakArena section in [`README.md`](README.md), then follow the linked issue or community case.
+
+- **Understand or reuse templates**
+  Start from [`templates/README.md`](templates/README.md), then open the most relevant scenario directory.
+
+- **Run the benchmark pipeline**
+  Read [`SKILL.md`](SKILL.md) and [`experiment/README.md`](experiment/README.md), then choose:
+  - `experiment/isc_single`
+  - `experiment/isc_icl`
+  - `experiment/isc_agent`
+
+- **Contribute a new case**
+  Check [`VERIFICATION.md`](VERIFICATION.md), collect evidence, and then open the ISC submission issue.
+
+### Step 4: Submit a Case
+
+If the user wants to submit a new case:
+
+1. pick a template and reproduce the behavior
+2. save evidence such as model output or API logs
+3. check [`VERIFICATION.md`](VERIFICATION.md) to confirm the case meets benchmark standards
+4. open the ISC submission issue:
+
+```text
+https://github.com/wuyoscar/ISC-Bench/issues/new?template=isc-submission.md&title=[ISC]+Model+Name
+```
+
+### Quick Reference
+
+| Resource | Purpose |
+|---------|---------|
+| [`README.md`](README.md) | Human-facing overview |
+| [`VERIFICATION.md`](VERIFICATION.md) | Rules and verification standards |
+| [`templates/`](templates/) | Public TVD scenarios |
+| [`experiment/`](experiment/) | Single-turn, ICL, and agentic pipelines |
+| [`community/`](community/) | Curated reproductions tied to issues |
+| [`SKILL.md`](SKILL.md) | Command-level workflow for running ISC-Bench |
+| [`tutorials/`](tutorials/) | Onboarding material |
+
+### One-Sentence Summary
+
+Clone the repo, read [`README.md`](README.md) and [`VERIFICATION.md`](VERIFICATION.md), then choose the correct template, experiment, or submission path based on the user's goal.

AGENT_README.md

@@ -2,7 +2,39 @@
 
 All notable updates to ISC-Bench are documented here.
 
-## 2026-03-27 (latest)
+## 2026-03-29 (latest)
+
+### Milestones
+- 🎆 **700+ GitHub stars**
+- **24/100** Arena models confirmed under ISC (top 100 leaderboard)
+
+### New ISC Cases
+- **GPT-5.4** (#57) by @wuyoscar — Agentic content moderation with **real OpenAI Moderation API verification**, 9/9 categories, 100% recall. Model self-optimized toxic text to pass threshold.
+- **Gemini 2.5 Pro** (#52) by @wuyoscar — LaTeX template (no code), social engineering attack scripts
+
+### ISC-Agent
+- 3 new moderation templates added (5 total): `aiml_moderation_content` (CSV), `aiml_moderation_input` (TXT), `aiml_moderation_output` (JSONL)
+- `aiml_moderation_content` now calls real OpenAI Moderation API for end-to-end verification
+
+### Templates
+- Renamed `aiml_openai_moderation` → `aiml_moderation_content` with 3 prompt variants
+- New template: `other_latex_fraud_survey` — pure LaTeX, no code
+- 2 templates converted to SKILL.md format: `compchem_ase_simulation`, `aiml_moderation_content` — with YAML frontmatter, NEVER lists, and customization guides
+- `verify_template.sh` script for standardized template testing
+
+### README
+- "Jailbroken" → "Triggered" across leaderboard, website, and gen scripts
+- Leaderboard capped at top 100 (was 330), split into 3 tiers (1-25, 26-50, 51-100)
+- "Recent News" split into "Updates" (ISC cases) + "News" (milestones, papers)
+- Added community quote: Adrian De Wynter
+- `cookbook/` → `tutorials/` (all markdown, 5 tutorials)
+- New FAQ entries: defense, code attack, related works
+
+### Tutorials
+- All 5 tutorials converted from .ipynb to .md
+- New: `04_icl_few_shot.md`, `05_attack_composability.md`
+
+## 2026-03-27
 
 ### Milestones
 - 🎆 **500+ GitHub stars**