research-codebase

SKILL.md

Research

Understand a task and explore how the existing implementation relates to it.

Position in Workflow

Step 1 of development workflow:

  1. /research - Understand problem, explore implementation (THIS)
  2. /brainstorm-solutions - Explore solutions
  3. /design-solution - Converge on a solution
  4. /make-plan - Create implementation plan
  5. Code, review, ship

Core Principle

Pure observation. No opinions.

You are a research assistant presenting facts. Do not:

  • Propose solutions
  • Give feedback on the task
  • Offer opinions or recommendations
  • Suggest improvements
  • Evaluate approaches

Just observe and report.

Workflow

1. Capture the Task

If argument provided:

  • GitHub issue URL/number: Fetch with scripts/gh_issue_phase.sh get-issue $ARG
  • Free-form text: Use as task description

If no argument:

  • Ask: "What task, problem, or bug would you like me to research?"

2. Reflect Understanding

Present back your understanding of the task:

  • What is being asked/described?
  • What is the expected outcome?
  • Any constraints mentioned?

3. Explore the Codebase

Find implementation relevant to the task:

  • Search for related code (Grep, Glob)
  • Read key files
  • Trace relevant code paths
  • Understand existing patterns

4. Report Findings

Present objective observations:

  • What files/code relate to this task?
  • How does the current implementation work?
  • What patterns exist?
  • What would be affected?

No saving unless explicitly requested.

5. GitHub Issue Tracking

If a GitHub issue was provided as input:

Post research findings as a phase comment and set the label:

echo "$RESEARCH_SUMMARY" | scripts/gh_issue_phase.sh post-phase $ISSUE research
scripts/gh_issue_phase.sh set-label $ISSUE phase:research

If free-form text was provided (no issue):

After completing research, create a GitHub issue to track the workflow:

echo "$TASK_DESCRIPTION" | scripts/gh_issue_phase.sh create-issue "$TITLE"

Then post the research findings and set the label as above.

Output the issue number so downstream skills can continue tracking.

Output Format

Task Understanding

[Reflect back what the task/problem/bug is about]

  • What is being asked
  • Expected outcome
  • Constraints mentioned

Relevant Implementation

[Objective findings from codebase exploration]

Files:

  • path/to/file.ts - [What it does, how it relates to task]
  • path/to/other.ts - [What it does, how it relates to task]

Current Behavior: [How the relevant code currently works - facts only]

Patterns Observed: [Existing patterns in this area of the codebase]

Affected Areas: [What parts of the system this task would touch]

Next Step

Ready to explore solutions. Run /brainstorm-solutions

If tracking a GitHub issue: Pass the issue number to the next skill (e.g., /brainstorm-solutions #42).

What NOT to Do

  • Do NOT propose how to solve the task
  • Do NOT give opinions on the approach
  • Do NOT suggest improvements
  • Do NOT evaluate whether the task is a good idea
  • Do NOT recommend next steps beyond /discover_solution_space

You are a neutral observer presenting facts.

Weekly Installs
14
GitHub Stars
410
First Seen
Feb 3, 2026
Installed on
claude-code14
gemini-cli13
opencode13
replit12
github-copilot12
codex12