| name | code-investigation-guide |
| description | How to research and present findings about a codebase. Reading code, tracing flows, summarizing architecture. Use when understanding, exploring or explaining existing code before proposing changes.
|
Code Investigation
Approach
- Map the area: identify key files, entry points, data flow.
- Read the code: understand actual behaviour, not just names.
- Look for context: tests, config, dependencies, patterns.
- Summarize with "so what": what matters for the task at hand.
- Pause for feedback: present findings before proposing changes.
What to Look For
- Entry points and public API surface
- Data flow: where does input come from, where does output go
- Existing tests and what they cover
- Configuration and environment dependencies
- Patterns and conventions the codebase follows
- Error handling approach
- Dependencies and what they provide
Presenting Findings
- Reference file and line; don't dump raw code.
- Explain what matters, not everything you found.
- Structure as: what exists → how it works → what's relevant.
- Call out surprises, risks or constraints.
- Keep it concise; the user can ask for more detail.
Command Output Fidelity
RTK may compress output from bash commands. Pi's native
grep, find and ls tools bypass RTK entirely (they
aren't bash), so prefer those for searching and listing.
For piped commands (git log | head -20), RTK rewrites
the left side but passes the right side through unchanged.
The pipe target receives RTK-compressed input.
When running git commands through bash during investigation,
be aware that RTK caps git log at 10 entries, injects
--no-merges silently and truncates lines. If you need
full commit history or complete messages, prefix the
command with NORTK=1 to bypass compression:
NORTK=1 git log --oneline -20
The NORTK=1 prefix is stripped before execution; bash
never sees it as an environment variable. It works on any
bash command, not just git.
What Not to Do
- Don't dump entire files into the conversation.
- Don't describe code mechanically line-by-line.
- Don't propose changes before presenting findings.
- Don't skip investigation and jump to solutions.
