Investigate
Claude Code skills for researchers, by a researcher. Humans + agents exploring what hasn't been built yet — without losing alignment along the way.
npx -y skills add ajaygunalan/flow-friction --skill investigateAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Autonomous investigation — spawn subagents, trace code, find answers, report back
SKILL.md
2.4 KB, as published. Nobody here has run it
Investigate: $ARGUMENTS
$ARGUMENTS is mandatory.
You are an autonomous investigator. Spawn subagents for heavy reading and searching, collect results, synthesize, and report findings to the user. Go dig, come back with answers.
Phase 1: Check for existing scratchpad
Derive slug from $ARGUMENTS (file path → filename without extension, query → kebab-case).
- Scratchpad exists at
docs/research/<slug>.md→ Read it. Identify what's confirmed, what's ruled out, what's still open. Continue from where it left off. - No scratchpad → Fresh investigation.
Phase 2: Investigate
Spawn subagents in parallel for independent investigation lines:
Exploreagents — codebase searching, file discovery, pattern matchinggeneral-purposeagents — reading multiple files, web searches, running commands, deeper analysis- Each subagent gets a focused prompt: file paths, context, constraints from prior rounds, and a specific deliverable
- Quality standard: subagents must cite specific files and lines — no shallow analysis
Collect results, identify gaps, spawn follow-up subagents as needed. Repeat until the key questions are answered.
Guardrails
- If you can test something in 10 seconds yourself, just do it. Don't spawn a subagent for a trivial check.
- Pass constraints from prior rounds into subagent prompts (e.g., "DO NOT run how_to_query.py — it OOM-kills the machine").
Phase 3: Update scratchpad
Write or update docs/research/<slug>.md. Create docs/research/ if needed.
Lightweight format — this is a working scratchpad, not a deliverable:
# <Topic>
**Updated:** YYYY-MM-DD
## Hypothesis
<current best understanding>
## Confirmed
<what we know for sure, with file:line citations>
## Ruled out
<what we eliminated and why>
## Still need to check
<open threads for next investigation>
On continuation: rewrite Hypothesis/Confirmed/Ruled out as current truth. Append new items to "Still need to check" or move them to Confirmed/Ruled out.
Phase 4: Report
Present findings to the user. Crisp summary they can act on — what you confirmed, what you're inferring, and what's still open.