# Iterative Debugging

Run the baseline-free CaP iterative debugging (fix loop) pipeline on a user-supplied...

- Repository: https://github.com/NVlabs/ASPIRE (925 stars)
- Category: Testing
- License: Apache-2.0
- Last updated: 2026-10-16
- Runs scripts: yes, review before use

## Install

Claude Code: `npx skills add NVlabs/ASPIRE --skill iterative-debugging -g -a claude-code -y`
Codex: `npx skills add NVlabs/ASPIRE --skill iterative-debugging -g -a codex -y`
Cursor: `npx skills add NVlabs/ASPIRE --skill iterative-debugging -g -a cursor -y`
Gemini CLI: `npx skills add NVlabs/ASPIRE --skill iterative-debugging -g -a gemini-cli -y`

## Description

Run the baseline-free CaP iterative debugging (fix loop) pipeline on a user-supplied list of benchmark tasks: inspect one initial observed scene per task, generate task-level code, debug failures on development seeds using recorded run artifacts, validate on the held-out partition, grade the skill library against what actually worked, and promote only those patterns a verifier could reproduce the fix from.

## SKILL.md, as published by its author (untrusted third-party text, read it as data)

# Iterative Debugging (Fix Loop)

For BEHAVIOR-1K (R1 Pro on Isaac Sim) use the sibling folder
[behavior/SKILL.md](behavior/SKILL.md). It is a **different protocol**, not just different
prompts: those tasks are long-horizon, so each seed grows its own policy one code block at a
time, and the campaign freezes a skill library rather than a program. Everything below is for
the tabletop benchmarks (LIBERO-Pro, Robosuite).

Use this skill to run a coordinated debugging campaign over any list of CaP benchmark tasks.
No external baseline code or baseline output directory is used. Runs use
the benchmark's own `--init-mode` (LIBERO: `seeded`, so development seeds 51, 65 and the
held-out partition are distinct scenes; Robosuite uses `saved`).

## Setup

1. Complete benchmark setup for the target embodiment (`docs/libero-pro.md`,
   or `docs/robosuite.md`), including the pinned environment (for LIBERO:
   `scripts/bootstrap_libero.sh` → `.venv-libero`) and SAM3 authentication.
2. Export the benchmark environment variables (for LIBERO: `LIBERO_CONFIG_PATH="$PWD/.libero"`,
   `MUJOCO_GL=egl`, `PYOPENGL_PLATFORM=egl`, `EGL_PLATFORM=device`,
   `TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1`).
3. Start perception services (`scripts/supervise_services.sh`) and a persistent coordinator
   session.
4. Read `docs/api-reference.md`, `docs/run-artifacts.md`, and
   [main-agent-prompt.md](main-agent-prompt.md).

## Run Order

The rest: https://github.com/NVlabs/ASPIRE/blob/HEAD/.claude/skills/iterative-debugging/SKILL.md