Intake Protocol

How AGENT-33 processes new information sources and converts them into actionable knowledge.

Repo Intake

When a new repository is provided for analysis:

1. Clone — Fetch the repository to a local working directory.
2. Dossier — Generate a structured dossier covering the repository's orchestration primitive, state model, tooling protocol, observability surfaces, safety/governance posture, and extensibility points.
3. Feature Matrix — Extract capabilities into a comparable feature matrix (orchestration, state, safety, tooling, observability, productization).
4. Gap Analysis — Compare extracted features against AGENT-33's current capabilities.
5. Improvement Proposals — Generate specific, testable proposals for any identified gaps.
6. Output — Store the dossier in engine memory and update the live feature matrix.

Usage: agent33 intake <repo-url>

User Guidance Intake

When receiving user instructions or corrections:

1. Parse Intent — Identify what the user wants changed (behavior, output format, policy)
2. Map to Files — Locate the specific prompts, templates, or configs that govern the behavior
3. Generate Edits — Produce minimal, targeted changes
4. Regression Test — Run affected workflows against test cases to verify no regressions
5. Apply — Commit changes with full provenance (user request → file change → test result)

Format/Situation Adaptation

When encountering a new data format or operational context:

1. Analyze Structure — Parse samples to identify schema, patterns, and edge cases
2. Check Tool Registry — Determine if existing tools can handle the format
3. Generate Extension — If no tool exists, generate a handler following the plugin system
4. Test — Validate against sample data
5. Register — Add to tool registry with capability declaration