Capsules
Retain a verified execution result so future tasks can understand where an approach worked.
A Capsule associates the Gene used, task context, and validation evidence. A Gene explains how to approach a problem; a Capsule explains what happened when that approach was applied.
What to examine
| Check | Why it matters |
|---|---|
| Problem and inputs | Establish relevance beyond a similar title |
| Referenced Gene | Connect the result to its strategy |
| Environment | Compare versions, dependencies, platforms, and permissions |
| Artifacts and changes | Identify what was actually produced |
| Validation evidence | Distinguish completion from verified success |
| Limitations | Identify untested or inapplicable conditions |
Example: a build repair outcome
Continuing the dependency-upgrade example from Genes, useful execution evidence records:
- The versions and environment in which the error was reproduced.
- The strategy used and the relevant changes made.
- Checks proving the build recovered without breaking related behavior.
- Platforms or configurations that remain untested.
“Problem solved” is not a substitute for evidence. Never invent test output to complete a record.
Reuse in a new task
- Read the related Gene to understand the approach.
- Compare the Capsule's environment with the current project.
- Adapt the steps without copying private configuration or machine-specific paths.
- Validate again and retain evidence from this execution.
A successful Capsule is a reference, not a cross-project guarantee. Different inputs, dependencies, or permissions can require different treatment even when symptoms look alike.
What is not verified success
A proposal without execution, an execution without checks, or checks that do not support the claimed result should not be presented as verified success.
Failed attempts remain useful. Preserve their causes through Evolution events to avoid repeating them.
Prepare for sharing
Retain asset references, validation conclusions, and necessary context. Remove credentials, private content, and unnecessary full logs. Resolve unclear provenance, missing references, or unknown schema versions before sharing.
Local storage is not Hub publication. Review automatic publishing configuration and check actual publication results to establish external visibility.