explore-code
The explore-code skill facilitates exploratory code modifications within deep learning research repositories. It manages module transplants, backbone adaptations, and the insertion of LoRA or adapter layers on isolated branches or worktrees. The skill requires explicit user authorization and mandates the creation of rollback-aware records in the explore_outputs directory.
Is explore-code safe to install?
Review before installing: our audit of explore-code's source files found 0 shell commands, 0 external URLs, file reads and writes (medium risk). Every command and URL listed appears verbatim in the skill's source. The skill modifies local repository files and executes Python scripts to plan changes and write output reports.
How we audit skills: our security review methodology.
Who is this skill for?
Deep learning researchers performing exploratory code experiments.
What can you do with it?
- Transplanting modules between research repositories
- Adapting model backbones
- Inserting LoRA or adapter layers
- Combining low-risk code modules
- Documenting candidate code changes and rollback procedures
How good is this skill?
Quality score: 9/10. The documentation clearly defines the scope, boundaries, and required output artifacts for the skill. It explicitly lists the scripts executed, allowing for an accurate risk assessment.
What does the skill file contain?
# explore-code Use this as the Rigor Improve implementation leaf skill. The installed slug remains `explore-code` for compatibility. Use the shared operating principles in `../../references/agent-operating-principles.md`; this skill should guide bounded candidate code work without over-prescribing implementation details. ## When to apply - When the researcher explicitly authorizes exploratory code changes on an isolated branch or worktree. - When the task is source-anchored module transplant, backbone adaptation, LoRA or adapter insertion, or low-risk module combination. - When summary-lev...
Frequently asked questions
When should I use this skill?
Use this skill when you have explicitly authorized exploratory code changes on an isolated branch or worktree, such as transplanting modules or adapting backbones.
What output files does this skill generate?
The skill generates CHANGESET.md, SCIENTIFIC_CHANGELOG.md, COMPARABILITY_REPORT.md, TOP_RUNS.md, and status.json within the explore_outputs directory.
Can I use this for standard debugging or baseline reproduction?
No. The skill is restricted to exploratory work and should not be used for conservative debugging, trusted baseline reproduction, or environment setup.
Related skills
paper-context-resolver
140.4KDevelopers and researchers performing deep learning repository reproduction who encounter specific technical ambiguities in documentation
The paper-context-resolver skill provides targeted information from research papers to resolve specific gaps in deep learning repository reproduction. It focuses on technical details like dataset splits, preprocessing steps, evaluation protocols, checkpoint mappings, and runtime assumptions.
minimal-run-and-audit
139.8KResearchers and developers performing smoke tests or documented inference and evaluation runs on deep learning repositories
The minimal-run-and-audit skill executes specific commands for deep learning repository reproduction and generates standardized evidence reports. It creates output files including patch notes and comparability reports to document verification status.
ai-research-explore
101.9KResearchers conducting deep learning experiments who have established a task, dataset, benchmark, and evaluation method
The ai-research-explore skill facilitates candidate-only deep learning research exploration. It operates on a durable current_research anchor to manage experiments, rank ideas, and produce auditable outputs while maintaining scientific rigor.
analyze-project
91.3KResearchers and developers who need to understand the architecture and implementation details of a deep learning repository before modifying code
The analyze-project skill performs read-only audits of deep learning repositories. It maps model structures, identifies training and inference entrypoints, reviews configuration relationships, and flags suspicious implementation patterns.