shrt

shrt

Record, replay and verify ordered chains of internal API calls. A chain is an ordered list of RPC calls that reproduces a backend state. A safe spot is a run a person approved as correct; later replays are diffed against it, so a regression names the rpc that changed.

Start here

On a fresh clone .shrt/docs/ is missing: shrt init -agents=false -build=false restores it. Then read .shrt/docs/README.md: "Before the first command", then "Quickstart" to build a regression suite (let the planner compose the chains, do not hand-write them), and "Four rules that are never negotiable", which hold for every command. Commit .claude/ (this skill and the shrt-contract-author subagent); shrt init -force -build=false refreshes docs and kit after an upgrade and never touches your config.

Route

question	file
commands, exit codes, the loop, the rules	.shrt/docs/README.md
which keys exist, with types; ${...} forms; what verify compares	.shrt/docs/GRAMMAR.md
compose, fill, assert, probe, principals, author a contract, verify, find, slice	.shrt/docs/PLAYBOOK.md §1-§11
it did something strange	.shrt/docs/PITFALLS.md
contracts for many domains	one shrt-contract-author subagent per domain, then the before: pass of .shrt/docs/PLAYBOOK.md §7 over the whole library yourself: a per-domain author cannot see other domains

Every command prints its flags and exit codes with -h.