strategy-review¶
Adversarially reviews one refined strategy: it produces a review file with numeric
rubric scores and detailed prose. After bootstrapping the assess-strat plugin and
fetching architecture context, it spawns a background scorer agent against the rubric,
then runs deterministic scripts to parse results, compute the APPROVE/REVISE/REJECT
verdict (total ≥6 and no zeros → APPROVE; ≥3 and ≤1 zero → REVISE; else REJECT), and
write scores to the review frontmatter with no LLM judgment. It then invokes the four
reviewer skills — feasibility, testability, scope, architecture — in parallel, each in
an isolated context: fork, and assembles their prose (plus Agreements/Disagreements)
into the review file. Outside dry-run it posts a summary comment, attaches the full
review, and applies the verdict label to the RHAISTRAT issue.
Plugin: strat-creator | User-invocable
Contract¶
Score a strategy against the rubric and run independent forked reviewers for detailed adversarial prose.
Identity
- Produces a review file with numeric scores across all rubric dimensions.
- Runs independent reviewer agents for feasibility, testability, scope, and architecture.
Optimization Targets
Invariants
- Do not skip rubric criteria or invent findings not supported by evidence.
- Do not modify the strategy being reviewed.
Traceability
Diagram¶
Arguments¶
/strategy-review <RHAISTRAT-NNNN> [--dry-run] [--architecture-context <path>]
| Argument | Required | Default | Description |
|---|---|---|---|
RHAISTRAT-NNNN |
- | The strategy key to review (e.g. RHAISTRAT-1531). Exactly one per invocation; the skill errors without it. The strategy must already be refined. | |
--architecture-context |
- | Local path to an architecture-context checkout to use instead of fetching from remote. | |
--dry-run |
- | Skip all Jira writes — save the review comment to a file instead of posting, and skip attachments/labels. |
Usage¶
/strategy-review RHAISTRAT-1531
/strategy-review RHAISTRAT-1531 --dry-run