Skip to content
Trurlic

Bootstrap

Bootstrap makes an existing codebase’s implicit architecture explicit. It runs as the bootstrap task type through your coding agent over MCP — the agent reads your source and records components, decisions, and patterns on its own, no interactive dialogue. It is always agent mode, so everything it records lands attribution="agent", flagged ⚠ for your review.

There is no trurlic bootstrap command — bootstrapping happens entirely through the agent.

Run it

Section titled “Run it”
Terminal window
trurlic init      # create .trurlic/ if you haven't
trurlic serve     # start the MCP server (or just have your IDE agent connected)

Then ask your agent:

Bootstrap the architecture with Trurlic.

The agent calls advance(component="project", task_type="bootstrap", mode="agent") and follows the autonomous step sequence:

scan_project → extract_decisions (per component) → project_rules → pattern_detection → ready
  • scan_project — read the tree, identify components and connections, register them.
  • extract_decisions — for each component, mine the source for architectural choices and record them.
  • project_rules — record project-wide, cross-cutting rules.
  • pattern_detection — synthesize patterns from repeated approaches.

None of these steps are gated — the agent runs straight through to ready: true.

Review what it found

Section titled “Review what it found”

Everything bootstrap records is agent · unreviewed. Treat the output as a first draft to audit, not ground truth:

Terminal window
trurlic map       # visual review; agent decisions are flagged ⚠
trurlic status    # counts and open issues
trurlic check     # validate graph integrity

Then, in your agent, promote or revise each decision as you confirm it:

  • Promote a decision you’ve reviewed and agree with — update_decision(name, mode="promote") flips its attribution to user.
  • Revise one that’s wrong — update_decision(name, mode="revise", …) edits it and versions the old content into history.

Stale or anchorless agent decisions can also be swept with trurlic gc once you’re done reviewing.

See Core Concepts for agent vs interactive mode and Workflow Engine for the step sequences.