## Why two skills together

When the words *"maybe"*, *"probably"*, *"likely"* appear, the agent is guessing. `/scientific-method` switches it into investigation mode — but an investigator grading its own homework inherits its own blind spots. That's what the reviewer is for:

- Investigate → `/scientific-method` forms hypotheses, derives testable consequences, runs experiments, and reports a PROVED / DISPROVED / INCONCLUSIVE verdict with the evidence attached.
- Review → `/investigation-review` — run in a **fresh context** (new session or subagent) — re-runs the evidence, challenges the strongest claim, and gates the verdict with PASS or numbered challenges.

Only a PASS-ed verdict graduates from hypothesis to fact. Each skill is also useful on its own.

## Install (one plugin)

Everything on this site installs from a single plugin — no curl, no shell. Paste these into Claude Code:

```
/plugin marketplace add blakesims/task-workflow-plugin
/plugin install task-workflow@task-workflow-marketplace
/reload-plugins
```

`/reload-plugins` activates it in the current session — no restart needed. Or use the **Copy install** button above to hand the whole job to your agent.

After install: invoke with `/task-workflow:scientific-method` or `/task-workflow:investigation-review` — or just say "use the scientific method" when Claude starts guessing.

## The two skills

### /scientific-method

Triggers on hedge words ("maybe", "probably", "likely"). Verifies the premise first, surfaces hidden assumptions, then loops hypothesis → experiment until it reaches a PROVED / DISPROVED / INCONCLUSIVE verdict - every claim backed by a command and its actual output.

### /investigation-review

Assumes the investigator cut corners until proven otherwise. Re-runs at least two pieces of evidence, checks what was not checked, challenges the strongest claim independently, then gates with PASS or numbered REVISE challenges. Best run in a fresh context.

## Using one skill on its own

The plugin namespaces every skill, so each is individually invocable — `/task-workflow:investigation-review` works standalone in a fresh context. The source files below stay readable and curl-able for review.