Add laboratory container image for function/profile experimentation

[WiggidyW]

Summary

Create a Docker container image for use with ObjectiveAI laboratory executions. The builder agent runs inside this container and experiments with different ObjectiveAI functions and profiles to find the optimal scoring configuration for a given dataset of tweets.

Background

ObjectiveAI laboratory executions spawn a builder agent inside a Docker container with an MCP server injected. The builder agent uses MCP tools to interact with its container environment — reading files, running commands, writing output. Our container image provides the environment the builder agent needs to experiment.

What the Builder Agent Does

The builder agent is the experimenter. Inside the container, it:

1. Reads the tweet dataset (scraped posts from posts_queue)
2. Creates an ObjectiveAI function definition
3. Creates an ObjectiveAI profile definition
4. Runs a function execution against the tweet dataset using the objectiveai CLI
5. Examines the resulting scores
6. Decides whether the scores match the user's intent
7. If not, modifies the function and/or profile and tries again
8. Outputs its best function + profile pair when satisfied

The evaluation agent then validates the builder's output against the expected schema and scores it.

Container Image Contents

• objectiveai CLI — for running function executions against the dataset
• Tweet dataset — scraped tweets serialized as JSON, available as a file in the container
• User's criteria — the psyop spec/description so the builder agent understands what "good scoring" means

Integration

A new CLI command in psychological-operations-cli:

psychological-operations lab <psyop-name>

This triggers a laboratory execution using the container image, passing the tweet dataset and psyop criteria to the builder agent. The winning function/profile pair from the lab gets saved into the psyop's stage configuration.

Acceptance Criteria

• Dockerfile that builds the container image
• Container includes objectiveai CLI binary
• Container accepts tweet dataset as a mounted/copied JSON file
• Evaluation script validates builder output (function + profile JSON)
• psychological-operations lab CLI command triggers the lab execution
• Documentation on the lab workflow