simsci

Plugin simsci from marketplace vivarium-ai-tools in ihmeuw/vivarium-suite. Commands read from the repo's manifest — 3 days ago.

Skills (10)

• _finalize-core

  Internal building block invoked by /simsci:pr-prep, /simsci:framework-development's Phase 5, and a model-development loop: takes work that is already reviewed and validated and prepares it for review — leftover-finding triage, the test summary posted on each changed file, the PR approval gate, a rev

• _review-core

  Internal building block invoked by /simsci:pr-prep (and /simsci:framework-development's review phase): runs the multi-agent review fan-out + functional-correctness pass + per-finding confidence scoring + synthesis on a diff the caller has already gathered. Not a review entry point.

• change-propagation

  Propagate an adapted copy of a reference file or directory across several targets — one or more packages in one or more repositories — in parallel, then file one draft PR per repository. Use when the user wants to roll a piece of boilerplate or config out to many places at once. Not for a single-tar

• commit-splitter

  Split a bulk uncommitted diff into a sequence of small, reviewable commits — and, when scope warrants, into multiple PR-sized branches. Use when the user asks to break up uncommitted changes for review, or references PR-size limits. May also be surfaced as a suggestion when a large pending diff is a

• framework-development

  Guided design→implement→verify→PR loop for a well-scoped feature.

• git-rescue

  Diagnose and untangle a messy git situation (stuck rebases, stacked-branch conflicts after squash-merge, divergent history, dropped commits).

• pr-prep

  Take a change you have already written from raw branch to a PR ready for review: multi-agent review, a per-finding disposition (fix now / ticket / drop), apply the fixes, re-validate, then organize the commits and open the draft PR. Use when the user says \"review this and open a PR\", \"prep this f

• regression-debugger

  Trace a regression across repositories to identify the behavioral change causing it.

• type-hinter

  Type-hint a target — a whole package, a sub-folder, or individual files — until it conforms to the package's mypy config. Runs as the lead of an agent team: resolves the inter-file dependency graph, spawns one autonomous teammate per file, and lets teammates negotiate shared type contracts directly.

• workflow-assessment

  Assess a finished run of an agentic workflow — did the multi-agent handoffs and tool invocations happen the way the workflow's definition says they should? Reads the Claude Code session transcripts and judges the trace against the workflow definition across coverage, ordering/gates, parallelism, han

Subagents (16)

• _diff_analyzer

  Use when: analyzing diffs between branches across one or more repos, summarizing code changes relevant to a regression, identifying which changes could affect program behavior/outputs.

• _feature_implementer

  Use when: implementing a stubbed interface in black-box TDD, working from the signatures, the design spec, and body-less test stubs as read-only criteria — never from the filled-in test assertions.

• _hypothesis_tester

  Test a single hypothesis about the cause of a regression by comparing old and new code.

• _propagate_target

  Use when: adapting a reference file/dir into one propagation target (a package or path in a repository) and reporting the result back to the lead. Spawned by the change-propagation skill, one worker per target.

• _review_design

  Use when: reviewing code for design and data structure choices, algorithmic efficiency, whether the right abstractions are used, representation trade-offs, API surface design.

• _review_documentation

  Use when: reviewing code for documentation quality, including docstrings, comments, README/changelog updates, and public API documentation accuracy.

• _review_dry

  Use when: reviewing code for DRY violations, duplicated logic, repeated patterns, opportunities to extract helpers or shared utilities, near-identical code blocks.

• _review_maintainability

  Use when: reviewing code for maintainability, readability, documentation quality, implicit assumptions, magic numbers, docstring accuracy, coupling between components.

• _review_scorer

  Use when: independently scoring a single code-review finding for confidence (0-100) against the verbatim rubric, so low-confidence findings can be filtered out before synthesis. Spawned one-per-finding by simsci:_review-core.

• _review_tests

  Use when: reviewing code for test coverage, test quality, edge cases, and whether tests validate the intended behavior without being brittle.

• _split_proposer

  Use when: proposing how to split an uncommitted working-tree diff into a sequence of reviewable commits, and into PR-sized branches when the total scope warrants it. Returns a structured plan without executing any git changes.

• _test_summarizer

  Use when: summarizing a PR's tests — one table per changed source file, ordered top down, mapping each case the file must handle to the test that covers it and marking the cases nothing covers. The caller posts each table as a comment on its file.

• _test_writer

  Use when: fleshing out body-less test stubs into real tests in black-box TDD, working from the public signatures and the design — never from an implementation.

• _trace_extractor

  Use when: extracting a compact orchestration trace (sub-agent dispatches, skill invocations, gates, notable tool calls) from one Claude Code session transcript JSONL, keeping raw transcript bulk out of the orchestrator's context.

• _type_hint_file

  Use when type-hinting a single Python file to conform to its package's mypy config. Spawned by the /simsci:type-hinter slash command as one autonomous teammate per target file; teammates coordinate cross-file type contracts directly with each other.

• _validator

  Use when: running a package's test/lint/type suite and reporting a compact PASS/FAIL verdict, keeping verbose suite output out of the orchestrator's context.

MCP servers (1)

• jira

  runs: uv run --quiet --script ${CLAUDE_PLUGIN_ROOT}/mcp-servers/jira/server.py