blueprintlisted
Install: claude install-skill maxgfr/skills
# blueprint
A plan fails for one of two reasons: it answered a question the user never
agreed to, or it asserted something about the repository that is not true. This skill is built against both.
All paths below are relative to this skill's directory.
## Three laws
1. **No design while the frontier is open.** Every unsettled decision that could
change observable behaviour is put to the user before an approach is proposed.
The grill stops on a structural condition, not on feeling thorough.
2. **No repository claim without `path:line`.** A fact you cannot cite is an
assumption, and it gets written down as one or asked about — never asserted.
3. **The artifact is the promise.** What `verify` reads later is the file, not
this conversation. Anything that lives only in the chat is lost.
Violating the letter of a law is violating its spirit.
## Invocations
Syntax: Codex uses `$blueprint`; the Claude plugin uses `/maxgfr:blueprint`;
a standalone Claude skill uses `/blueprint`. The table
shows the arguments after that host-specific name.
| | |
|---|---|
| no arguments | **Default.** Orient → grill → ground → design → write → approve. |
| `grill` | The interview only. Stops at the locked constraints; designs nothing. |
| `crosscheck` | + one read-only consultation of the other CLI agent before approval. |
| `<path>` | Harden an existing plan instead of writing a new one. |
| `auto` | After approval, run `build` then `verify` on the plan — same turn, no prompt between