Skip to content

Add CLAUDE.md and AGENTS.md project instructions #5

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
fernandosmither wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Add CLAUDE.md and AGENTS.md project instructions #5

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
21 changes: 21 additions & 0 deletions AGENTS.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,21 @@
# SubliminalCaT — Algoverse AI Safety Research Program

**Core research question:** "Can a model misalign during Chain-of-Thought by reasoning about something benign like a math problem?"

## Codebase Structure

- `thought_virus/` — Team's previous token-finding pipeline (`subliminal_token_analyzer.py`, `run_analysis.py`, `run_cot_injection_analysis.py`)
- Experiment scripts for natural insertion and single-turn conditions
- TruthfulQA evaluation scripts (MC1 + MC2 metrics)

## Infrastructure

- **GPU:** Lambda Cloud. The user provisions machines and provides SSH access via IP.
- When running remote experiments: monitor them to ensure they're running correctly and not wasting GPU resources on bugs.

## Conventions

- Results go in structured output directories with metadata.
- Use existing infrastructure from `thought_virus/` where possible.
- Add final result files to `.gitignore` — advise the user where they are so they can upload to the shared Google Drive.
- **Always ask the user** for design decisions rather than guessing. Interview the user in depth about technical implementation, experiment design, concerns, and tradeoffs. Ask non-obvious, probing questions. Continue interviewing until the spec is complete, then proceed with implementation.
21 changes: 21 additions & 0 deletions CLAUDE.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,21 @@
# SubliminalCaT — Algoverse AI Safety Research Program

**Core research question:** "Can a model misalign during Chain-of-Thought by reasoning about something benign like a math problem?"

## Codebase Structure

- `thought_virus/` — Team's previous token-finding pipeline (`subliminal_token_analyzer.py`, `run_analysis.py`, `run_cot_injection_analysis.py`)
- Experiment scripts for natural insertion and single-turn conditions
- TruthfulQA evaluation scripts (MC1 + MC2 metrics)

## Infrastructure

- **GPU:** Lambda Cloud. The user provisions machines and provides SSH access via IP.
- When running remote experiments: monitor them to ensure they're running correctly and not wasting GPU resources on bugs.

## Conventions

- Results go in structured output directories with metadata.
- Use existing infrastructure from `thought_virus/` where possible.
- Add final result files to `.gitignore` — advise the user where they are so they can upload to the shared Google Drive.
- **Always ask the user** for design decisions rather than guessing. Use the AskUserQuestion tool to interview the user in depth about technical implementation, experiment design, concerns, and tradeoffs. Ask non-obvious, probing questions. Continue interviewing until the spec is complete, then proceed with implementation.