Comprehension
Comprehension is Spine’s free wedge: eight read-only tools you can use from Claude Code or Codex. They’re deterministic, use no LLM, need no credentials, and write nothing. Ask in plain language — the assistant picks the tool.
The eight tools
| Tool | The question it answers |
|---|---|
map_repo | “I’ve never seen this repo — what is it?” |
blast_radius | “What breaks if I change this?” |
explain_symbol | “What is this and who uses it?” |
investigate | “Where does this ticket land in the code?” |
localize | “Here’s a stack trace — where’s the bug?” |
regression_gaps | “What will break silently, with no test to catch it?” |
root_cause | “Why is this happening, and what’s the fix shape?” |
docs_for | “Which docs describe this — and are they still true?” |
A real transcript
you › where would a fix for the "sessions expire early" bug go?
assistant → investigate("sessions expire early")
Likely site: auth.validate app/auth.py:60 → reads secret_keyReachable, untested: commit db.py:88Related doc: docs/auth.md — driftedRead-only by design
These 8 tools are the read-only half of Spine’s 17 MCP tools. The remaining tools perform SDLC actions — and those write, only after a human gate. See Governance & gates and the MCP tools reference.