stage-evidencelisted
Install: claude install-skill m4ttstack/skills
# stage: evidence
{{stage.fields}}
## Run state
Contracts v2 and v3 (authoritative text: the parameterized-skills skill's convention reference).
- First action: `rt runs stage-start --stage evidence`
- Read consumed fields with `rt runs field get <key>` before
deriving or asking for them.
- Write each declared produce the moment it exists:
`rt runs field set <key> <value> --stage evidence`
- Last action on success: `rt runs stage-done --stage evidence`;
on failure: `rt runs stage-fail --stage evidence --reason
"<what actually failed>" --detail-path <path to whatever was captured
before the failure>` before you report it.
If `evidence-plan` is `none` (or starts with `none`), write `evidence` as
`{"plan": "none"}` and finish -- nothing to capture.
Capture the BEFORE per the plan now -- before implementation changes the
surface. Do not defer the before to ship time: reverting code on a running
dev server to reconstruct it produces stale, false befores. If the ticket
already embeds the broken state, that IS the before -- record its location
instead of recapturing.
## Gate `evidence`
Before any capture, when the domain rules below declare intake questions,
or the data source is anything other than the local default:
- `rt runs field set gate evidence --stage evidence`
- One sentence: what the plan asks for and what is unknown.
- The form: the domain's intake questions as it words them; the data
source when it is not local (**Proceed with `<source>`** / **Switc