Genes
Retain a reusable approach rather than every detail of one task.
A Gene describes a strategy, matching signals, and applicability boundaries. Evolver can recall relevant Genes from workspace signals and assess them against the current task.
What belongs in a Gene
A useful approach has a recognizable problem type, explainable steps, and a validation method. Repeated dependency-compatibility problems may justify a diagnostic strategy; one lucky change is not proof of reuse.
| Content | Question to answer |
|---|---|
| Signals | When should this approach be considered? |
| Preconditions and limits | Which environments fit, and when should work stop? |
| Steps | What should be checked and handled first? |
| Validation | How do you confirm a fix without regressions? |
| Failure handling | What evidence should be retained, and how can work be recovered? |
This is a content checklist, not an importable JSON schema.
Example: a build fails after an upgrade
Signal: A dependency upgrade introduces a new build error.
Approach: Preserve the first error and version information, check compatibility requirements, isolate the smallest relevant difference, and change only related configuration or code.
Validation: Rebuild in the same environment and check affected behavior.
Exclusion: If missing credentials or an unreachable service caused the failure, repair the environment instead of masking the problem with a dependency downgrade.
The Gene retains the diagnostic method. Exact files changed and test outputs belong in the execution evidence and process record.
Check before reuse
- Compare actual failure signals, not just title keywords.
- Review environment, dependencies, and permission requirements.
- Adapt the strategy to the current project and define allowed changes.
- Validate afterward and record differences from the original experience.
Recalling a Gene does not authorize every step it describes. Deletion, network access, and publication retain the current task's permission boundaries.
Updates and invalidation
Investigate failures in new environments. Distinguish changed conditions, outdated steps, and missing validation; do not rewrite a failed outcome as success.
Retain provenance and version relationships so older outcomes remain explainable. For review, withdrawal, and upgrades of external assets, see Asset governance.
Storage and sharing
Local GEP asset locations depend on configuration; see Asset paths. Do not overwrite live asset files.
For structured extraction, use tools supported by the installed version. See Distilling Genes, Skills, and Recipes for commands and quality requirements. Check privacy, references, and evidence before publication.
Related docs
- Capsules: Results of applying a strategy.
- Evolution events: Selection and validation history.