prototype-spikelisted
Install: claude install-skill wagneripjr/skills
# Prototype Spike
Invoke as `prototype-spike:prototype-spike` in Claude Code or `$prototype-spike:prototype-spike` in Codex.
Resolve `references/` from this installed `SKILL.md` directory. Use the host's native file,
search, edit, and terminal tools; browser evidence follows `references/verification.md`.
Build one self-contained HTML file that executes a requirement's interpretation against the real
system, for two audiences at once.
**Outward** it asks the client *is this what you meant?* — its output is deltas, not approval.
**Inward** it asks the team *does this technique work, and what should the contract be?* — its output
is a settled decision. States are indexed by **open questions**: some are acceptance criteria, some
are unproven mechanisms. Flipping a control re-runs the reading.
Three files, every run, written beside the prototype:
- `<anchor>-<slug>.html` — the prototype
- `harvest-table.md` — where every borrowed value came from
- `spike-report.md` — what the run settled, and where each answer belongs
`spike-report.md` is **opened at ANCHOR and appended to as you go** — not composed at the end. A run
that ends early still leaves its answers behind, and answers are the part that outlives the file: a
technique proven in a run that stopped short otherwise ends up recorded nowhere but a code comment.
Treat it the way you would a lab notebook, not a closing summary.
Four invariants:
- **Only invent what the requirement adds.** Structure, copy, icons, spacing, t