Claude Code plugin for deep system discovery, implementation planning, and code generation — research, multi-LLM review, TDD-oriented section generation
Agent
Architecture and design expert for the code-review panel. Spawned for large or cross-module diffs, new modules, or boundary changes. Reviews coupling, dependency direction, API shape, and extensibility. Outputs the shared panel JSON.
Agent
Generates focused audit document content for a specific topic. Has web access for build-vs-buy research.
Agent
Claim-verification agent for the code-review panel, spawned after the experts return. The panel's only network stage. Web-verifies framework, library, API, statistical-method, SQL-dialect (Snowflake/BigQuery/Postgres), and version-currency claims in the merged findings against current official documentation, batching…
Agent
Sole code-review agent for /deep implement Phase 5. Reviews implementation sections in any language (Python, TypeScript/JavaScript, Go, SQL/dbt) against the quality rule-packs active for the target. Four-phase review workflow (context → high-level → line-by-line → summary), five review dimensions mapped to rule…
Agent
Data-engineering expert for the code-review panel. Spawned when the diff touches SQL, dbt models or schema .yml files, Spark, or pandas/polars ETL. Hunts silently-wrong query logic, non-idempotent incremental loads, join explosions, null handling, and warehouse-pack (SQL-/DBT-) violations. Outputs the shared panel…
Agent
Correctness and logic expert for the code-review panel. Always spawned. Deep-dives algorithms, control flow, edge cases, and invariants in changed code — the things linters and type checkers cannot see. Outputs the shared panel JSON.
Agent
Machine-learning expert for the code-review panel. Spawned when the diff touches ML frameworks (torch, tensorflow, sklearn, xgboost, lightgbm, transformers), training/eval scripts, or notebooks. Hunts data leakage, evaluation flaws, reproducibility gaps, and tensor bugs. Outputs the shared panel JSON.
Agent
MLOps and ML-infrastructure expert for the code-review panel. Spawned when the diff touches pipelines (Airflow/Dagster/Prefect), experiment tracking (MLflow/W&B), model serving, feature stores, Dockerfiles/K8s, or CI for ML. Hunts versioning gaps, irreproducible environments, unsafe rollouts, and missing monitoring.…
Agent
Reviews implementation plans (fallback when external LLMs unavailable).
Agent
LLM prompt-engineering expert for the code-review panel. Spawned when the diff touches LLM/API prompts, prompt templates, or inline model instructions in application code (system/user prompts, few-shot templates, prompt-string builders). Reviews prompting quality, output contracts, context economy, injection surface…
Agent
Final verification gate for the code-review panel — always the last subagent. Re-reads the actual code for every merged finding, kills phantom bugs, fixes wrong file/line references, deduplicates overlapping expert findings, and normalizes severity. Only findings it approves reach the user.
Agent
Generates self-contained implementation section content. Outputs raw markdown. Used by /deep-plan for parallel section generation.
Agent
Skill and agent-definition expert for the code-review panel. Spawned when the diff touches SKILL.md files, agent definitions, or hook prompts. Audits them against the "Don't Ship Skills Without Evals" rubric — description-as-trigger, lean progressive-disclosure structure, directives over essays, negative cases…
Agent
Inspects installed Claude Code skills at /deep hook points and decides which ones to auto-invoke, which to surface for user approval, and which to skip. Used by /deep plan, /deep implement, and /deep auto.
Agent
Statistics and data-science methodology expert for the code-review panel. Spawned when the diff touches statistical tests, experiment/A-B analysis, metric definitions, sampling, or forecasting code (scipy.stats, statsmodels, experiment frameworks). Hunts invalid inference, biased sampling, and misleading aggregation.…
Agent
Decides which session artifacts deserve long-lived storage in the Obsidian vault and writes the survivors via the obsidian-vault skill. Invoked at the end of every /deep mode (discovery, plan, implement, auto).