| name | strategic-review |
| type | skill |
| category | instruction |
| description | Unified multi-agent review of any artifact — a document, plan, proposal, or pull request. The calling agent deploys rbg, pauli, and marsha in parallel, then @james reconciles their findings into one verdict. Pass `comment` and/or `fix` to write the result back to the review surface. Use `--critic` for a fast pauli-only pre-hoc critique. |
| triggers | ["strategic review","review this document","review this proposal","review this plan","review this PR","review PR","review a pull request","adversarial review","/strategic-review","/strategic-review --critic","critic review"] |
| modifies_files | true |
| needs_task | false |
| mode | execution |
| domain | ["framework","quality-assurance"] |
| allowed-tools | Agent, Bash, Read, Glob, Grep, AskUserQuestion |
| version | 3.0.0 |
| permalink | skills-strategic-review |
/strategic-review — Unified Multi-Agent Review
Review any artifact — a document, plan, proposal, or pull request — from several expert
perspectives and return one reconciled verdict. Owned by james (reconciliation), but
you, the invoking agent, own the orchestration: you deploy the reviewers yourself, because a
subagent cannot spawn its own subagents. James is called only at the end, to reconcile.
Modes
- Default — full review: deploy
rbg + pauli + marsha in parallel, then @james reconciles.
--critic — solo pauli: one fast adversarial critique, no reconciliation. For a pre-hoc
sanity check on a plan or proposal before work starts.
Inputs
- The artifact: a file path, a PKB id, pasted text, or a pull request (an
owner/repo#N ref or URL).
- Action flags (optional):
comment, fix, or both. With no flag the review is advisory —
you return the verdict to the caller and change nothing.
Procedure (default mode)
1. Gather context
Identify and load the artifact. If it is a PR, load the diff, description, and any prior
unresolved review comments (gh pr view, gh pr diff). Select the matching review-context
descriptor from review-contexts/ and pass it to every reviewer:
review-contexts/pr-code.md — code PRs.review-contexts/pr-framework.md — framework PRs (skills, agents, hooks, specs).
2. Deploy the three reviewers — in parallel
Spawn all three in a single message (concurrent Agent calls). Give each the artifact, the
context, and the descriptor. You spawn them — not james. Subagents cannot spawn subagents, so
this fan-out must happen here, at the top level.
- rbg — axiom & rule compliance. Always runs.
- pauli — strategic critique: the premise test, then the "is this in the right place, or a
workaround for a root cause that belongs elsewhere?" architectural-fit lens (the 10 cognitive
moves). Always runs. Discipline and the worked specimen live in [[references/premise-test.md]].
