| name | blindspots |
| description | This skill should be used when the user is running blindspots testing, when a ".blindspots/" directory exists in the project, when the user mentions "blindspots", "dogfood", "user-trial", "experiment", or "interview" in a testing context, or when working in a git branch starting with "blindspots/". |
Blindspots — Find Your Product's Blind Spots
This skill provides context and guidance during an active blindspots session in Codex.
Modes
| Mode | Purpose | Has Specs? | Fixes Bugs? |
|---|
blindspots:setup | Interactive config and persona generation | N/A | No |
blindspots:dogfood | Test against specs, find regressions | Yes | Yes |
blindspots:user-trial | Persona discovers the product with no insider knowledge | No | No |
blindspots:experiment | A/B compare two variants with user trial personas | No | No |
blindspots:interview | Ask personas questions about the product | Depends | No |
The .blindspots/ Directory
Config is checked in; session output is gitignored.
Config (checked in)
| File | Purpose |
|---|
.blindspots/config.md | Project testing config (setup, start, explore, diagnose, test, specs) |
.blindspots/personas.md | Pool of user personas |
Output (gitignored)
| File | Mode | Purpose |
|---|
.blindspots/dogfood-journals/<persona>.md | dogfood | Per-persona session log |
.blindspots/user-trials/<persona>/ | user-trial | Per-persona trial output (journal, discovered-specs, reactions, comparison) |
.blindspots/screenshots/ | all | Evidence |
.blindspots/experiments/<name>/ | experiment | A/B test output |
.blindspots/interviews/<name>.md | interview | Interview transcripts |
Roles
- Browser explorer — Browser-based user trial explorer. Avoid source-code inspection.
- Terminal explorer — Terminal-based user trial explorer. Use shell and limited file reads.
- Trial comparer — Analyzes discovered specs vs actual specs.
- Experiment verdict — Synthesizes A/B test results into ship/no-ship verdict.
- Interviewer — Conducts persona Q&A sessions.
Shared Steps
When the user asks for any Blindspots mode, first make sure setup exists. The
shared startup pattern:
- Check for
.blindspots/config.md — run blindspots:setup if missing
- Check for
.blindspots/personas.md — run blindspots:setup if missing
- Select persona — method varies by mode:
blindspots:dogfood, blindspots:user-trial: random, or requested persona
blindspots:interview: show persona list and ask user to pick
blindspots:experiment: prompt interactively for test name, variants, user type, and run count
- Run setup commands from config
- Infer interface mode from
## Start (browser or terminal)
- Mode-specific launch with
BLINDSPOTS_DEPTH=1
Reference Files
references/dogfood-cycle.md — Guided testing cycle (EXPLORE → USE → DIAGNOSE → FIX → COMMIT)
references/compare-prompt.md — Spec comparison methodology
references/persona-template.md — Persona generation template
references/experiment-verdict-prompt.md — A/B verdict methodology