---

name: go_explore_archive_selection_update description: Maintain and sample Go-Explore archive entries using score, length, and exploration metadata.

Go-Explore Archive Selection and Update

Use this skill when implementing Phase 1 archive mechanics for Go-Explore-style exploration. It owns insertion, replacement, and deterministic sampling of archive cells. Do not use it for cell encoding or environment stepping.

Inputs

• A candidate trajectory record with cell_key, state, actions, score, and optional length.
• An archive dictionary keyed by encoded cells.
• Optional seeded random generator and selection settings.

Outputs

• An updated archive entry when the candidate is new or improves an existing entry.
• A selected archive entry for return-then-explore.

Workflow

1. Insert candidates for unseen cells.
2. Replace existing entries only if the candidate has higher cumulative score or equal score with a shorter trajectory.
3. Track update and selection counts for auditability.
4. Sample cells with a deterministic seeded weighted rule that favors high score and low selection count.
5. Return both the selected key and entry so recovery logs can prove which frontier was explored.

Validation

Run python tests/test_archive.py or validate with validate_skill_tree.py --run-tests.

Limitations

This skill intentionally keeps the weighting simple for bounded recovery. Full Atari-scale Go-Explore can use richer weights, but the replacement invariant should remain the same.