| name | my-review |
| description | Deep, security-weighted review of your changes or a PR — routes the reviewer model to the change's complexity (opus by default; fable for a complex diff, on ask), then spawns the my-review agent (xhigh reasoning) on that model. Read-only, report-only. Use for "/my-review", "/my-review <PR#>". |
| argument-hint | [PR number/URL] [--complexity trivial|standard|complex]; empty = review the local working diff |
| allowed-tools | Agent, AskUserQuestion, Bash(git:*), Bash(gh:*) |
You are the main-thread launcher for the my-review agent. Pick the reviewer model
forward-or-judge — never a blind prompt — then spawn the agent. Three steps:
- Get a tier (forward-or-judge). In priority order:
--complexity <tier> in $ARGUMENTS wins — take that tier verbatim.
- else a tier already confirmed this session (an earlier
/classify-task or /pipeline
run in this conversation) — reuse it.
- else judge the tier yourself from a cheap diff peek:
git diff HEAD --stat for the local
working diff, or gh pr diff <N> --stat for a PR — the size and spread of the change decide
trivial / standard / complex. This skill stays dependency-free of the workflow plugin: do
not invoke its classify-task skill or its resolve-tier.sh helper.
- Pick the model from the tier.
- complex →
AskUserQuestion: opus (default — cheaper, faster) vs fable (deepest,
slowest).
- not complex (trivial / standard) → opus, no prompt.
- Spawn the reviewer with one
Agent call: subagent_type: personal-tools:my-review,
model: "<pick>". Hand it the target — a PR number/URL if $ARGUMENTS names one, else the
local working diff (git diff HEAD). Relay its report verbatim.
/pipeline does not invoke this skill — there the complexity tier decides the reviewer model
directly. The agents/my-review.md frontmatter now pins model: opus as the fallback for any
direct spawn that omits an override.