Skip to content

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

Skill Contract canonical-skill-v1

Score a strategy against the rubric and run independent forked reviewers for detailed adversarial prose.

Identity

Functions
review
Success
  • Produces a review file with numeric scores across all rubric dimensions.
  • Runs independent reviewer agents for feasibility, testability, scope, and architecture.

Optimization Targets

task_success judge SKILL.md @ 0e102d4
output_quality judge SKILL.md @ 0e102d4

Invariants

Must Preserve
  • Do not skip rubric criteria or invent findings not supported by evidence.
  • Do not modify the strategy being reviewed.
Fixed Context
toolsRead, Write, Edit, Glob, Grep, Bash, Skill, Agent
clipython3
knowledgerepository_contentpublic, task_inputtask_private, tool_outputtask_private

Traceability

Diagram

strategy-review 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