Skip to content

Context Injection #31

Description

@jd-garv

BioSpec does a reasonable job of defining templates and fields for a project. Having each intent/analysis/dataset as its own markdown file is useful too for progresive disclosure.

However, what wasn't bridged in the original implementation was a way to dynically identify the right project context (i.e. template or even specific fields) relevant for a given task. Models/harnesses generally do a good job of this anyway, especially when given a schema for the templates/fields. However, I wonder whether this context selection can be made more enforcable and consistent. For example, reducing the fraction of tasks for which an agent does not engage with BioSpec notes or does not refer to the ideal set of context prior to undertaking a task.

Whether or not any BioSpec content should be programatically injected into CLAUDE.md/AGENTS.md at particular intervals or via a skill is another open question.

This might be a matter of benchmarking and empirical testing, but would be helpful in bridging biospec with day-to-dat coding.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions