verifylisted
Install: claude install-skill Shilun-Allan-Li/claude-math-tcs-agent
Args: !`python3 "${CLAUDE_PLUGIN_ROOT}/scripts/mathtcs.py" args verify $ARGUMENTS`
Project: !`python3 "${CLAUDE_PLUGIN_ROOT}/scripts/mathtcs.py" project detect`
Now: !`date -u +%Y-%m-%dT%H:%M:%SZ`
Plugin root: !`echo "${CLAUDE_PLUGIN_ROOT}"`
You are the **verify coordinator**. Elaboration is a formal check done by `mathtcs.py snapshot`; the two reviewers are model reviews; the combined report keeps the three apart and records that no human has approved anything. `MT` = `python3 "<Plugin root>/scripts/mathtcs.py"`; `NOW` = the timestamp above.
## Input contract
`Args.target` is a declaration id, a slug, or a Lean module path (relative to the project). `Args.flags.ids` narrows the set. If `Args.errors` is non-empty print them with `/math-tcs:verify <lean-file | id | slug> [--ids a,b]` and stop.
## Steps
1. **Resolve ids.** id → itself; slug → `MT manifest list --slug <slug>`; module path → `MT manifest list --module <module name>` (the module name is `<module_prefix>.<file stem>`). Keep entries whose status is `scaffolded`, `verified`, `needs_statement_review`, `deferred`, `unfinished` or `failed`; report `blocked`/`translated` ones as skipped with the reason.
2. For **each** declaration:
a. `MT snapshot <id> --at NOW` → `dir`, `context`, `statement_sha256`, `elaboration`. If it returns `skipped` (existing representation), record that and continue.
b. **Launch both reviewers in one message** (two Agent calls in the same turn): `math-tcs:semantic-reviewer` and `math-t